// Copyright (c) 2012-2022 Wojciech Figat. All rights reserved.

#include "Model.h"
#include "Engine/Core/Log.h"
#include "Engine/Core/Math/Int3.h"
#include "Engine/Core/RandomStream.h"
#include "Engine/Engine/Engine.h"
#include "Engine/Serialization/MemoryReadStream.h"
#include "Engine/Content/WeakAssetReference.h"
#include "Engine/Content/Upgraders/ModelAssetUpgrader.h"
#include "Engine/Content/Factories/BinaryAssetFactory.h"
#include "Engine/Core/Math/Int2.h"
#include "Engine/Debug/DebugDraw.h"
#include "Engine/Graphics/RenderTools.h"
#include "Engine/Graphics/RenderTask.h"
#include "Engine/Graphics/Models/ModelInstanceEntry.h"
#include "Engine/Streaming/StreamingGroup.h"
#include "Engine/Debug/Exceptions/ArgumentOutOfRangeException.h"
#include "Engine/Graphics/Async/GPUTask.h"
#include "Engine/Graphics/Textures/GPUTexture.h"
#include "Engine/Profiler/ProfilerCPU.h"
#include "Engine/Renderer/DrawCall.h"
#include "Engine/Threading/JobSystem.h"
#include "Engine/Threading/Threading.h"
#include "Engine/Tools/ModelTool/MeshAccelerationStructure.h"
#if GPU_ENABLE_ASYNC_RESOURCES_CREATION
#include "Engine/Threading/ThreadPoolTask.h"
#define STREAM_TASK_BASE ThreadPoolTask
#else
#include "Engine/Threading/MainThreadTask.h"
#define STREAM_TASK_BASE MainThreadTask
#endif

#define CHECK_INVALID_BUFFER(buffer) \
    if (buffer->IsValidFor(this) == false) \
	{ \
		LOG(Warning, "Invalid Model Instance Buffer size {0} for Model {1}. It should be {2}. Manual update to proper size.", buffer->Count(), ToString(), MaterialSlots.Count()); \
		buffer->Setup(this); \
	}

REGISTER_BINARY_ASSET_ABSTRACT(ModelBase, "FlaxEngine.ModelBase");

/// <summary>
/// Model LOD streaming task.
/// </summary>
class StreamModelLODTask : public STREAM_TASK_BASE
{
private:
    WeakAssetReference<Model> _asset;
    int32 _lodIndex;
    FlaxStorage::LockData _dataLock;

public:
    /// <summary>
    /// Init
    /// </summary>
    /// <param name="model">Parent model</param>
    /// <param name="lodIndex">LOD to stream index</param>
    StreamModelLODTask(Model* model, int32 lodIndex)
        : _asset(model)
        , _lodIndex(lodIndex)
        , _dataLock(model->Storage->Lock())
    {
    }

public:
    // [ThreadPoolTask]
    bool HasReference(Object* resource) const override
    {
        return _asset == resource;
    }

protected:
    // [ThreadPoolTask]
    bool Run() override
    {
        AssetReference<Model> model = _asset.Get();
        if (model == nullptr)
        {
            return true;
        }

        // Get data
        BytesContainer data;
        model->GetLODData(_lodIndex, data);
        if (data.IsInvalid())
        {
            LOG(Warning, "Missing data chunk");
            return true;
        }
        MemoryReadStream stream(data.Get(), data.Length());

        // Note: this is running on thread pool task so we must be sure that updated LOD is not used at all (for rendering)

        // Load model LOD (initialize vertex and index buffers)
        if (model->LODs[_lodIndex].Load(stream))
        {
            LOG(Warning, "Cannot load LOD{1} for model \'{0}\'", model->ToString(), _lodIndex);
            return true;
        }

        // Update residency level
        model->_loadedLODs++;

        return false;
    }

    void OnEnd() override
    {
        // Unlink
        if (_asset)
        {
            ASSERT(_asset->_streamingTask == this);
            _asset->_streamingTask = nullptr;
            _asset = nullptr;
        }
        _dataLock.Release();

        // Base
        STREAM_TASK_BASE::OnEnd();
    }
};

REGISTER_BINARY_ASSET_WITH_UPGRADER(Model, "FlaxEngine.Model", ModelAssetUpgrader, true);

Model::Model(const SpawnParams& params, const AssetInfo* info)
    : ModelBase(params, info, StreamingGroups::Instance()->Models())
{
}

Model::~Model()
{
    ASSERT(_streamingTask == nullptr);
}

bool Model::Intersects(const Ray& ray, const Matrix& world, float& distance, Vector3& normal, Mesh** mesh, int32 lodIndex)
{
    return LODs[lodIndex].Intersects(ray, world, distance, normal, mesh);
}

BoundingBox Model::GetBox(const Matrix& world, int32 lodIndex) const
{
    return LODs[lodIndex].GetBox(world);
}

BoundingBox Model::GetBox(int32 lodIndex) const
{
    return LODs[lodIndex].GetBox();
}

void Model::Draw(const RenderContext& renderContext, MaterialBase* material, const Matrix& world, StaticFlags flags, bool receiveDecals) const
{
    if (!CanBeRendered())
        return;

    // Select a proper LOD index (model may be culled)
    const BoundingBox box = GetBox(world);
    BoundingSphere sphere;
    BoundingSphere::FromBox(box, sphere);
    int32 lodIndex = RenderTools::ComputeModelLOD(this, sphere.Center, sphere.Radius, renderContext);
    if (lodIndex == -1)
        return;
    lodIndex += renderContext.View.ModelLODBias;
    lodIndex = ClampLODIndex(lodIndex);

    // Draw
    LODs[lodIndex].Draw(renderContext, material, world, flags, receiveDecals);
}

void Model::Draw(const RenderContext& renderContext, const Mesh::DrawInfo& info)
{
    ASSERT(info.Buffer);
    if (!CanBeRendered())
        return;
    const auto frame = Engine::FrameCount;
    const auto modelFrame = info.DrawState->PrevFrame + 1;
    CHECK_INVALID_BUFFER(info.Buffer);

    // Select a proper LOD index (model may be culled)
    int32 lodIndex;
    if (info.ForcedLOD != -1)
    {
        lodIndex = info.ForcedLOD;
    }
    else
    {
        lodIndex = RenderTools::ComputeModelLOD(this, info.Bounds.Center, info.Bounds.Radius, renderContext);
        if (lodIndex == -1)
        {
            // Handling model fade-out transition
            if (modelFrame == frame && info.DrawState->PrevLOD != -1)
            {
                // Check if start transition
                if (info.DrawState->LODTransition == 255)
                {
                    info.DrawState->LODTransition = 0;
                }

                RenderTools::UpdateModelLODTransition(info.DrawState->LODTransition);

                // Check if end transition
                if (info.DrawState->LODTransition == 255)
                {
                    info.DrawState->PrevLOD = lodIndex;
                }
                else
                {
                    const auto prevLOD = ClampLODIndex(info.DrawState->PrevLOD);
                    const float normalizedProgress = static_cast<float>(info.DrawState->LODTransition) * (1.0f / 255.0f);
                    LODs[prevLOD].Draw(renderContext, info, normalizedProgress);
                }
            }

            return;
        }
    }
    lodIndex += info.LODBias + renderContext.View.ModelLODBias;
    lodIndex = ClampLODIndex(lodIndex);

    // Check if it's the new frame and could update the drawing state (note: model instance could be rendered many times per frame to different viewports)
    if (modelFrame == frame)
    {
        // Check if start transition
        if (info.DrawState->PrevLOD != lodIndex && info.DrawState->LODTransition == 255)
        {
            info.DrawState->LODTransition = 0;
        }

        RenderTools::UpdateModelLODTransition(info.DrawState->LODTransition);

        // Check if end transition
        if (info.DrawState->LODTransition == 255)
        {
            info.DrawState->PrevLOD = lodIndex;
        }
    }
    // Check if there was a gap between frames in drawing this model instance
    else if (modelFrame < frame || info.DrawState->PrevLOD == -1)
    {
        // Reset state
        info.DrawState->PrevLOD = lodIndex;
        info.DrawState->LODTransition = 255;
    }

    // Draw
    if (info.DrawState->PrevLOD == lodIndex)
    {
        LODs[lodIndex].Draw(renderContext, info, 0.0f);
    }
    else if (info.DrawState->PrevLOD == -1)
    {
        const float normalizedProgress = static_cast<float>(info.DrawState->LODTransition) * (1.0f / 255.0f);
        LODs[lodIndex].Draw(renderContext, info, 1.0f - normalizedProgress);
    }
    else
    {
        const auto prevLOD = ClampLODIndex(info.DrawState->PrevLOD);
        const float normalizedProgress = static_cast<float>(info.DrawState->LODTransition) * (1.0f / 255.0f);
        LODs[prevLOD].Draw(renderContext, info, normalizedProgress);
        LODs[lodIndex].Draw(renderContext, info, normalizedProgress - 1.0f);
    }
}

bool Model::SetupLODs(const Span<int32>& meshesCountPerLod)
{
    ScopeLock lock(Locker);

    // Validate input and state
    if (!IsVirtual())
    {
        LOG(Error, "Only virtual models can be updated at runtime.");
        return true;
    }

    return Init(meshesCountPerLod);
}

#if USE_EDITOR

bool Model::Save(bool withMeshDataFromGpu, const StringView& path)
{
    // Validate state
    if (WaitForLoaded())
    {
        LOG(Error, "Asset loading failed. Cannot save it.");
        return true;
    }
    if (IsVirtual() && path.IsEmpty())
    {
        LOG(Error, "To save virtual asset asset you need to specify the target asset path location.");
        return true;
    }
    if (withMeshDataFromGpu && IsInMainThread())
    {
        LOG(Error, "To save model with GPU mesh buffers it needs to be called from the other thread (not the main thread).");
        return true;
    }
    if (IsVirtual() && !withMeshDataFromGpu)
    {
        LOG(Error, "To save virtual model asset you need to specify 'withMeshDataFromGpu' (it has no other storage container to get data).");
        return true;
    }

    ScopeLock lock(Locker);

    // Create model data header
    MemoryWriteStream headerStream(1024);
    MemoryWriteStream* stream = &headerStream;
    {
        // Min Screen Size
        stream->WriteFloat(MinScreenSize);

        // Amount of material slots
        stream->WriteInt32(MaterialSlots.Count());

        // For each material slot
        for (int32 materialSlotIndex = 0; materialSlotIndex < MaterialSlots.Count(); materialSlotIndex++)
        {
            auto& slot = MaterialSlots[materialSlotIndex];

            const auto id = slot.Material.GetID();
            stream->Write(&id);
            stream->WriteByte(static_cast<byte>(slot.ShadowsMode));
            stream->WriteString(slot.Name, 11);
        }

        // Amount of LODs
        const int32 lods = LODs.Count();
        stream->WriteByte(lods);

        // For each LOD
        for (int32 lodIndex = 0; lodIndex < lods; lodIndex++)
        {
            auto& lod = LODs[lodIndex];

            // Screen Size
            stream->WriteFloat(lod.ScreenSize);

            // Amount of meshes
            const int32 meshes = lod.Meshes.Count();
            stream->WriteUint16(meshes);

            // For each mesh
            for (int32 meshIndex = 0; meshIndex < meshes; meshIndex++)
            {
                const auto& mesh = lod.Meshes[meshIndex];

                // Material Slot index
                stream->WriteInt32(mesh.GetMaterialSlotIndex());

                // Box
                const auto box = mesh.GetBox();
                stream->Write(&box);

                // Sphere
                const auto sphere = mesh.GetSphere();
                stream->Write(&sphere);

                // Has Lightmap UVs
                stream->WriteBool(mesh.HasLightmapUVs());
            }
        }
    }

    // Use a temporary chunks for data storage for virtual assets
    FlaxChunk* tmpChunks[ASSET_FILE_DATA_CHUNKS];
    Platform::MemoryClear(tmpChunks, sizeof(tmpChunks));
    Array<FlaxChunk> chunks;
    if (IsVirtual())
        chunks.Resize(ASSET_FILE_DATA_CHUNKS);
#define GET_CHUNK(index) (IsVirtual() ? tmpChunks[index] = &chunks[index] : GetOrCreateChunk(index))

    // Check if use data from drive or from GPU
    if (withMeshDataFromGpu)
    {
        // Download all meshes buffers
        Array<Task*> tasks;
        for (int32 lodIndex = 0; lodIndex < LODs.Count(); lodIndex++)
        {
            auto& lod = LODs[lodIndex];

            const int32 meshesCount = lod.Meshes.Count();
            struct MeshData
            {
                BytesContainer VB0;
                BytesContainer VB1;
                BytesContainer VB2;
                BytesContainer IB;

                uint32 DataSize() const
                {
                    return VB0.Length() + VB1.Length() + VB2.Length() + IB.Length();
                }
            };
            Array<MeshData> meshesData;
            meshesData.Resize(meshesCount);
            tasks.EnsureCapacity(meshesCount * 4);

            for (int32 meshIndex = 0; meshIndex < meshesCount; meshIndex++)
            {
                const auto& mesh = lod.Meshes[meshIndex];
                auto& meshData = meshesData[meshIndex];

                // Vertex Buffer 0 (required)
                auto task = mesh.DownloadDataGPUAsync(MeshBufferType::Vertex0, meshData.VB0);
                if (task == nullptr)
                    return true;
                task->Start();
                tasks.Add(task);

                // Vertex Buffer 1 (required)
                task = mesh.DownloadDataGPUAsync(MeshBufferType::Vertex1, meshData.VB1);
                if (task == nullptr)
                    return true;
                task->Start();
                tasks.Add(task);

                // Vertex Buffer 2 (optional)
                task = mesh.DownloadDataGPUAsync(MeshBufferType::Vertex2, meshData.VB2);
                if (task)
                {
                    task->Start();
                    tasks.Add(task);
                }

                // Index Buffer (required)
                task = mesh.DownloadDataGPUAsync(MeshBufferType::Index, meshData.IB);
                if (task == nullptr)
                    return true;
                task->Start();
                tasks.Add(task);
            }

            // Wait for all
            if (Task::WaitAll(tasks))
                return true;
            tasks.Clear();

            // Create meshes data
            {
                int32 dataSize = meshesCount * (2 * sizeof(uint32) + sizeof(bool));
                for (int32 meshIndex = 0; meshIndex < meshesCount; meshIndex++)
                {
                    dataSize += meshesData[meshIndex].DataSize();
                }

                MemoryWriteStream meshesStream(dataSize);

                for (int32 meshIndex = 0; meshIndex < meshesCount; meshIndex++)
                {
                    const auto& mesh = lod.Meshes[meshIndex];
                    const auto& meshData = meshesData[meshIndex];

                    uint32 vertices = mesh.GetVertexCount();
                    uint32 triangles = mesh.GetTriangleCount();
                    bool hasColors = meshData.VB2.IsValid();
                    uint32 vb0Size = vertices * sizeof(VB0ElementType);
                    uint32 vb1Size = vertices * sizeof(VB1ElementType);
                    uint32 vb2Size = vertices * sizeof(VB2ElementType);
                    uint32 indicesCount = triangles * 3;
                    bool shouldUse16BitIndexBuffer = indicesCount <= MAX_uint16;
                    bool use16BitIndexBuffer = mesh.Use16BitIndexBuffer();
                    uint32 ibSize = indicesCount * (use16BitIndexBuffer ? sizeof(uint16) : sizeof(uint32));

                    if (vertices == 0 || triangles == 0)
                    {
                        LOG(Warning, "Cannot save model with empty meshes.");
                        return true;
                    }
                    if ((uint32)meshData.VB0.Length() < vb0Size)
                    {
                        LOG(Warning, "Invalid vertex buffer 0 size.");
                        return true;
                    }
                    if ((uint32)meshData.VB1.Length() < vb1Size)
                    {
                        LOG(Warning, "Invalid vertex buffer 1 size.");
                        return true;
                    }
                    if (hasColors && (uint32)meshData.VB2.Length() < vb2Size)
                    {
                        LOG(Warning, "Invalid vertex buffer 2 size.");
                        return true;
                    }
                    if ((uint32)meshData.IB.Length() < ibSize)
                    {
                        LOG(Warning, "Invalid index buffer size.");
                        return true;
                    }

                    meshesStream.WriteUint32(vertices);
                    meshesStream.WriteUint32(triangles);

                    meshesStream.WriteBytes(meshData.VB0.Get(), vb0Size);
                    meshesStream.WriteBytes(meshData.VB1.Get(), vb1Size);

                    meshesStream.WriteBool(hasColors);

                    if (hasColors)
                    {
                        meshesStream.WriteBytes(meshData.VB2.Get(), vb2Size);
                    }

                    if (shouldUse16BitIndexBuffer == use16BitIndexBuffer)
                    {
                        meshesStream.WriteBytes(meshData.IB.Get(), ibSize);
                    }
                    else if (shouldUse16BitIndexBuffer)
                    {
                        auto ib = (const int32*)meshData.IB.Get();
                        for (uint32 i = 0; i < indicesCount; i++)
                        {
                            meshesStream.WriteUint16(ib[i]);
                        }
                    }
                    else
                    {
                        CRASH;
                    }
                }

                // Override LOD data chunk with the fetched GPU meshes memory
                auto lodChunk = GET_CHUNK(MODEL_LOD_TO_CHUNK_INDEX(lodIndex));
                if (lodChunk == nullptr)
                    return true;
                lodChunk->Data.Copy(meshesStream.GetHandle(), meshesStream.GetPosition());
            }
        }
    }
    else
    {
        ASSERT(!IsVirtual());

        // Load all chunks with a mesh data
        for (int32 lodIndex = 0; lodIndex < LODs.Count(); lodIndex++)
        {
            if (LoadChunk(MODEL_LOD_TO_CHUNK_INDEX(lodIndex)))
                return true;
        }
    }

    // Set mesh header data
    auto headerChunk = GET_CHUNK(0);
    ASSERT(headerChunk != nullptr);
    headerChunk->Data.Copy(headerStream.GetHandle(), headerStream.GetPosition());

#undef GET_CHUNK

    // Save
    AssetInitData data;
    data.SerializedVersion = SerializedVersion;
    if (IsVirtual())
        Platform::MemoryCopy(_header.Chunks, tmpChunks, sizeof(_header.Chunks));
    const bool saveResult = path.HasChars() ? SaveAsset(path, data) : SaveAsset(data, true);
    if (IsVirtual())
        Platform::MemoryClear(_header.Chunks, sizeof(_header.Chunks));
    if (saveResult)
    {
        LOG(Error, "Cannot save \'{0}\'", ToString());
        return true;
    }

    return false;
}

#endif

bool Model::GenerateSDF(float resolutionScale, int32 lodIndex)
{
    if (!HasAnyLODInitialized())
        return true;
    if (IsInMainThread() && IsVirtual())
    {
        // TODO: could be supported if algorithm could run on a GPU and called during rendering
        LOG(Warning, "Cannot generate SDF for virtual models on a main thread.");
        return true;
    }
    PROFILE_CPU();
    auto startTime = Platform::GetTimeSeconds();
    ScopeLock lock(Locker);

    // Setup SDF texture properties
    lodIndex = Math::Clamp(lodIndex, HighestResidentLODIndex(), LODs.Count() - 1);
    auto& lod = LODs[lodIndex];
    BoundingBox bounds = lod.GetBox();
    Vector3 size = bounds.GetSize();
    SDF.WorldUnitsPerVoxel = 10 / Math::Max(resolutionScale, 0.0001f);
    Int3 resolution(Vector3::Ceil(Vector3::Clamp(size / SDF.WorldUnitsPerVoxel, 4, 256)));
    Vector3 uvwToLocalMul = size;
    Vector3 uvwToLocalAdd = bounds.Minimum;
    SDF.LocalToUVWMul = Vector3::One / uvwToLocalMul;
    SDF.LocalToUVWAdd = -uvwToLocalAdd / uvwToLocalMul;
    SDF.MaxDistance = size.MaxValue();
    SDF.LocalBoundsMin = bounds.Minimum;
    SDF.LocalBoundsMax = bounds.Maximum;
    // TODO: maybe apply 1 voxel margin around the geometry?
    const int32 maxMips = 3;
    const int32 mipCount = Math::Min(MipLevelsCount(resolution.X, resolution.Y, resolution.Z, true), maxMips);
    if (!SDF.Texture)
        SDF.Texture = GPUTexture::New();
    // TODO: use 8bit format for smaller SDF textures (eg. res<100)
    if (SDF.Texture->Init(GPUTextureDescription::New3D(resolution.X, resolution.Y, resolution.Z, PixelFormat::R16_Float, GPUTextureFlags::ShaderResource | GPUTextureFlags::UnorderedAccess, mipCount)))
    {
        SAFE_DELETE_GPU_RESOURCE(SDF.Texture);
        return true;
    }

    // TODO: support GPU to generate model SDF on-the-fly (if called during rendering)

    // Setup acceleration structure for fast ray tracing the mesh triangles
    MeshAccelerationStructure scene;
    scene.Add(this, lodIndex);
    scene.BuildBVH();

    // Allocate memory for the distant field
    Array<Half> voxels;
    voxels.Resize(resolution.X * resolution.Y * resolution.Z);
    Vector3 xyzToLocalMul = uvwToLocalMul / Vector3(resolution);
    Vector3 xyzToLocalAdd = uvwToLocalAdd;

    // TODO: use optimized sparse storage for SDF data as hierarchical bricks
    // https://graphics.pixar.com/library/IrradianceAtlas/paper.pdf
    // http://maverick.inria.fr/Membres/Cyril.Crassin/thesis/CCrassinThesis_EN_Web.pdf
    // http://ramakarl.com/pdfs/2016_Hoetzlein_GVDB.pdf
    // https://www.cse.chalmers.se/~uffe/HighResolutionSparseVoxelDAGs.pdf
    // then use R8 format and brick size of 8x8x8

    // Brute-force for each voxel to calculate distance to the closest triangle with point query and distance sign by raycasting around the voxel
    const int32 sampleCount = 12;
    Array<Vector3> sampleDirections;
    sampleDirections.Resize(sampleCount);
    {
        RandomStream rand;
        sampleDirections.Get()[0] = Vector3::Up;
        sampleDirections.Get()[1] = Vector3::Down;
        sampleDirections.Get()[2] = Vector3::Left;
        sampleDirections.Get()[3] = Vector3::Right;
        sampleDirections.Get()[4] = Vector3::Forward;
        sampleDirections.Get()[5] = Vector3::Backward;
        for (int32 i = 6; i < sampleCount; i++)
            sampleDirections.Get()[i] = rand.GetUnitVector();
    }
    Function<void(int32)> sdfJob = [this, &resolution, &sampleDirections, &scene, &voxels, &xyzToLocalMul, &xyzToLocalAdd](int32 z)
    {
        PROFILE_CPU_NAMED("Model SDF Job");
        const float encodeScale = 1.0f / SDF.MaxDistance;
        float hitDistance;
        Vector3 hitNormal, hitPoint;
        Triangle hitTriangle;
        const int32 zAddress = resolution.Y * resolution.X * z;
        for (int32 y = 0; y < resolution.Y; y++)
        {
            const int32 yAddress = resolution.X * y + zAddress;
            for (int32 x = 0; x < resolution.X; x++)
            {
                float minDistance = SDF.MaxDistance;
                Vector3 voxelPos = Vector3((float)x, (float)y, (float)z) * xyzToLocalMul + xyzToLocalAdd;

                // Point query to find the distance to the closest surface
                scene.PointQuery(voxelPos, minDistance, hitPoint, hitTriangle);

                // Raycast samples around voxel to count triangle backfaces hit
                int32 hitBackCount = 0, hitCount = 0;
                for (int32 sample = 0; sample < sampleDirections.Count(); sample++)
                {
                    Ray sampleRay(voxelPos, sampleDirections[sample]);
                    if (scene.RayCast(sampleRay, hitDistance, hitNormal, hitTriangle))
                    {
                        hitCount++;
                        const bool backHit = Vector3::Dot(sampleRay.Direction, hitTriangle.GetNormal()) > 0;
                        if (backHit)
                            hitBackCount++;
                    }
                }

                float distance = minDistance;
                // TODO: surface thickness threshold? shift reduce distance for all voxels by something like 0.01 to enlarge thin geometry
                //if ((float)hitBackCount > )hitCount * 0.3f && hitCount != 0)
                if ((float)hitBackCount > (float)sampleDirections.Count() * 0.6f && hitCount != 0)
                {
                    // Voxel is inside the geometry so turn it into negative distance to the surface
                    distance *= -1;
                }
                const int32 xAddress = x + yAddress;
                voxels.Get()[xAddress] = Float16Compressor::Compress(distance * encodeScale);
            }
        }
    };
    JobSystem::Execute(sdfJob, resolution.Z);

    // Upload data to the GPU
    BytesContainer data;
    data.Link((byte*)voxels.Get(), voxels.Count() * sizeof(Half));
    auto task = SDF.Texture->UploadMipMapAsync(data, 0, resolution.X * sizeof(Half), data.Length(), true);
    if (task)
        task->Start();

    // Generate mip maps
    Array<Half> voxelsMip;
    for (int32 mipLevel = 1; mipLevel < mipCount; mipLevel++)
    {
        Int3 resolutionMip = Int3::Max(resolution / 2, Int3::One);
        voxelsMip.Resize(resolutionMip.X * resolutionMip.Y * resolutionMip.Z);

        // Downscale mip
        Function<void(int32)> mipJob = [this, &voxelsMip, &voxels, &resolution, &resolutionMip](int32 z)
        {
            PROFILE_CPU_NAMED("Model SDF Mip Job");
            const float encodeScale = 1.0f / SDF.MaxDistance;
            const float decodeScale = SDF.MaxDistance;
            const int32 zAddress = resolutionMip.Y * resolutionMip.X * z;
            for (int32 y = 0; y < resolutionMip.Y; y++)
            {
                const int32 yAddress = resolutionMip.X * y + zAddress;
                for (int32 x = 0; x < resolutionMip.X; x++)
                {
                    // Linear box filter around the voxel
                    // TODO: use min distance for nearby texels (texel distance + distance to texel)
                    float distance = 0;
                    for (int32 dz = 0; dz < 2; dz++)
                    {
                        const int32 dzAddress = (z * 2 + dz) * (resolution.Y * resolution.X);
                        for (int32 dy = 0; dy < 2; dy++)
                        {
                            const int32 dyAddress = (y * 2 + dy) * (resolution.X) + dzAddress;
                            for (int32 dx = 0; dx < 2; dx++)
                            {
                                const int32 dxAddress = (x * 2 + dx) + dyAddress;
                                const float d = Float16Compressor::Decompress(voxels.Get()[dxAddress]) * decodeScale;
                                distance += d;
                            }
                        }
                    }
                    distance *= 1.0f / 8.0f;

                    const int32 xAddress = x + yAddress;
                    voxelsMip.Get()[xAddress] = Float16Compressor::Compress(distance * encodeScale);
                }
            }
        };
        JobSystem::Execute(mipJob, resolutionMip.Z);

        // Upload to the GPU
        data.Link((byte*)voxelsMip.Get(), voxelsMip.Count() * sizeof(Half));
        task = SDF.Texture->UploadMipMapAsync(data, mipLevel, resolutionMip.X * sizeof(Half), data.Length(), true);
        if (task)
            task->Start();

        // Go down
        voxelsMip.Swap(voxels);
        resolution = resolutionMip;
    }

#if !BUILD_RELEASE
    auto endTime = Platform::GetTimeSeconds();
    LOG(Info, "Generated SDF {}x{}x{} ({} kB) in {}ms for {}", resolution.X, resolution.Y, resolution.Z, SDF.Texture->GetMemoryUsage() / 1024, (int32)((endTime - startTime) * 1000.0), GetPath());
#endif
    return false;
}

bool Model::Init(const Span<int32>& meshesCountPerLod)
{
    if (meshesCountPerLod.IsInvalid() || meshesCountPerLod.Length() > MODEL_MAX_LODS)
    {
        Log::ArgumentOutOfRangeException();
        return true;
    }

    // Dispose previous data and disable streaming (will start data uploading tasks manually)
    StopStreaming();

    // Setup
    MaterialSlots.Resize(1);
    MinScreenSize = 0.0f;
    SAFE_DELETE_GPU_RESOURCE(SDF.Texture);

    // Setup LODs
    for (int32 lodIndex = 0; lodIndex < LODs.Count(); lodIndex++)
    {
        LODs[lodIndex].Dispose();
    }
    LODs.Resize(meshesCountPerLod.Length());
    _loadedLODs = meshesCountPerLod.Length();

    // Setup meshes
    for (int32 lodIndex = 0; lodIndex < meshesCountPerLod.Length(); lodIndex++)
    {
        auto& lod = LODs[lodIndex];
        lod._model = this;
        lod.ScreenSize = 1.0f;
        const int32 meshesCount = meshesCountPerLod[lodIndex];
        if (meshesCount <= 0 || meshesCount > MODEL_MAX_MESHES)
            return true;

        lod.Meshes.Resize(meshesCount);
        for (int32 meshIndex = 0; meshIndex < meshesCount; meshIndex++)
        {
            lod.Meshes[meshIndex].Init(this, lodIndex, meshIndex, 0, BoundingBox::Zero, BoundingSphere::Empty, true);
        }
    }

    return false;
}

void Model::SetupMaterialSlots(int32 slotsCount)
{
    ModelBase::SetupMaterialSlots(slotsCount);

    // Adjust meshes indices for slots
    for (int32 lodIndex = 0; lodIndex < LODs.Count(); lodIndex++)
    {
        for (int32 meshIndex = 0; meshIndex < LODs[lodIndex].Meshes.Count(); meshIndex++)
        {
            auto& mesh = LODs[lodIndex].Meshes[meshIndex];
            if (mesh.GetMaterialSlotIndex() >= slotsCount)
                mesh.SetMaterialSlotIndex(slotsCount - 1);
        }
    }
}

int32 Model::GetLODsCount() const
{
    return LODs.Count();
}

void Model::GetMeshes(Array<MeshBase*>& meshes, int32 lodIndex)
{
    auto& lod = LODs[lodIndex];
    meshes.Resize(lod.Meshes.Count());
    for (int32 meshIndex = 0; meshIndex < lod.Meshes.Count(); meshIndex++)
        meshes[meshIndex] = &lod.Meshes[meshIndex];
}

void Model::InitAsVirtual()
{
    // Init with a single LOD and one mesh
    int32 meshesCount = 1;
    Init(ToSpan(&meshesCount, 1));

    // Base
    BinaryAsset::InitAsVirtual();
}

#if USE_EDITOR

void Model::GetReferences(Array<Guid>& output) const
{
    // Base
    BinaryAsset::GetReferences(output);

    for (int32 i = 0; i < MaterialSlots.Count(); i++)
    {
        output.Add(MaterialSlots[i].Material.GetID());
    }
}

#endif

int32 Model::GetMaxResidency() const
{
    return LODs.Count();
}

int32 Model::GetCurrentResidency() const
{
    return _loadedLODs;
}

int32 Model::GetAllocatedResidency() const
{
    return LODs.Count();
}

bool Model::CanBeUpdated() const
{
    // Check if is ready and has no streaming tasks running
    return IsInitialized() && _streamingTask == nullptr;
}

Task* Model::UpdateAllocation(int32 residency)
{
    // Models are not using dynamic allocation feature
    return nullptr;
}

Task* Model::CreateStreamingTask(int32 residency)
{
    ScopeLock lock(Locker);

    ASSERT(IsInitialized() && Math::IsInRange(residency, 0, LODs.Count()) && _streamingTask == nullptr);
    Task* result = nullptr;
    const int32 lodCount = residency - GetCurrentResidency();

    // Switch if go up or down with residency
    if (lodCount > 0)
    {
        // Allow only to change LODs count by 1
        ASSERT(Math::Abs(lodCount) == 1);

        int32 lodIndex = HighestResidentLODIndex() - 1;

        // Request LOD data
        result = (Task*)RequestLODDataAsync(lodIndex);

        // Add upload data task
        _streamingTask = New<StreamModelLODTask>(this, lodIndex);
        if (result)
            result->ContinueWith(_streamingTask);
        else
            result = _streamingTask;
    }
    else
    {
        // Do the quick data release
        for (int32 i = HighestResidentLODIndex(); i < LODs.Count() - residency; i++)
            LODs[i].Unload();
        _loadedLODs = residency;
    }

    return result;
}

Asset::LoadResult Model::load()
{
    // Get header chunk
    auto chunk0 = GetChunk(0);
    if (chunk0 == nullptr || chunk0->IsMissing())
        return LoadResult::MissingDataChunk;
    MemoryReadStream headerStream(chunk0->Get(), chunk0->Size());
    ReadStream* stream = &headerStream;

    // Min Screen Size
    stream->ReadFloat(&MinScreenSize);

    // Amount of material slots
    int32 materialSlotsCount;
    stream->ReadInt32(&materialSlotsCount);
    if (materialSlotsCount <= 0 || materialSlotsCount > 4096)
        return LoadResult::InvalidData;
    MaterialSlots.Resize(materialSlotsCount, false);

    // For each material slot
    for (int32 materialSlotIndex = 0; materialSlotIndex < materialSlotsCount; materialSlotIndex++)
    {
        auto& slot = MaterialSlots[materialSlotIndex];

        // Material
        Guid materialId;
        stream->Read(&materialId);
        slot.Material = materialId;

        // Shadows Mode
        slot.ShadowsMode = static_cast<ShadowsCastingMode>(stream->ReadByte());

        // Name
        stream->ReadString(&slot.Name, 11);
    }

    // Amount of LODs
    byte lods;
    stream->ReadByte(&lods);
    if (lods == 0 || lods > MODEL_MAX_LODS)
        return LoadResult::InvalidData;
    LODs.Resize(lods);

    // For each LOD
    for (int32 lodIndex = 0; lodIndex < lods; lodIndex++)
    {
        auto& lod = LODs[lodIndex];
        lod._model = this;

        // Screen Size
        stream->ReadFloat(&lod.ScreenSize);

        // Amount of meshes
        uint16 meshesCount;
        stream->ReadUint16(&meshesCount);
        if (meshesCount == 0 || meshesCount > MODEL_MAX_MESHES)
            return LoadResult::InvalidData;
        ASSERT(lodIndex == 0 || LODs[0].Meshes.Count() >= meshesCount);

        // Allocate memory
        lod.Meshes.Resize(meshesCount, false);

        // For each mesh
        for (uint16 meshIndex = 0; meshIndex < meshesCount; meshIndex++)
        {
            // Material Slot index
            int32 materialSlotIndex;
            stream->ReadInt32(&materialSlotIndex);
            if (materialSlotIndex < 0 || materialSlotIndex >= materialSlotsCount)
            {
                LOG(Warning, "Invalid material slot index {0} for mesh {1}. Slots count: {2}.", materialSlotIndex, meshIndex, materialSlotsCount);
                return LoadResult::InvalidData;
            }

            // Box
            BoundingBox box;
            stream->Read(&box);

            // Sphere
            BoundingSphere sphere;
            stream->Read(&sphere);

            // Has Lightmap UVs
            bool hasLightmapUVs = stream->ReadBool();

            lod.Meshes[meshIndex].Init(this, lodIndex, meshIndex, materialSlotIndex, box, sphere, hasLightmapUVs);
        }
    }

#if BUILD_DEBUG || BUILD_DEVELOPMENT
    // Validate LODs
    for (int32 lodIndex = 1; lodIndex < LODs.Count(); lodIndex++)
    {
        const auto prevSS = LODs[lodIndex - 1].ScreenSize;
        const auto thisSS = LODs[lodIndex].ScreenSize;
        if (prevSS <= thisSS)
        {
            LOG(Warning, "Model LOD {0} has invalid screen size compared to LOD {1} (asset: {2})", lodIndex, lodIndex - 1, ToString());
        }
    }
#endif

    // Request resource streaming
    StartStreaming(true);

    return LoadResult::Ok;
}

void Model::unload(bool isReloading)
{
    // End streaming (if still active)
    if (_streamingTask != nullptr)
    {
        // Cancel streaming task
        _streamingTask->Cancel();
        _streamingTask = nullptr;
    }

    // Cleanup
    SAFE_DELETE_GPU_RESOURCE(SDF.Texture);
    MaterialSlots.Resize(0);
    for (int32 i = 0; i < LODs.Count(); i++)
        LODs[i].Dispose();
    LODs.Clear();
    _loadedLODs = 0;
}

bool Model::init(AssetInitData& initData)
{
    // Validate
    if (initData.SerializedVersion != SerializedVersion)
    {
        LOG(Error, "Invalid serialized model version.");
        return true;
    }

    return false;
}

AssetChunksFlag Model::getChunksToPreload() const
{
    // Note: we don't preload any LODs here because it's done by the Streaming Manager
    return GET_CHUNK_FLAG(0);
}

void ModelBase::SetupMaterialSlots(int32 slotsCount)
{
    CHECK(slotsCount >= 0 && slotsCount < 4096);
    if (!IsVirtual() && WaitForLoaded())
        return;

    ScopeLock lock(Locker);

    const int32 prevCount = MaterialSlots.Count();
    MaterialSlots.Resize(slotsCount, false);

    // Initialize slot names
    for (int32 i = prevCount; i < slotsCount; i++)
        MaterialSlots[i].Name = String::Format(TEXT("Material {0}"), i + 1);
}

MaterialSlot* ModelBase::GetSlot(const StringView& name)
{
    MaterialSlot* result = nullptr;
    for (auto& slot : MaterialSlots)
    {
        if (slot.Name == name)
        {
            result = &slot;
            break;
        }
    }
    return result;
}