#pragma once

#include "WorldMesher.h"

#include <condition_variable>
#include <array>
#include <unordered_set>

struct FStandaloneTileKeyHash
{
    std::size_t operator()(const FVoxelTileKey& Key) const noexcept
    { return static_cast<std::size_t>(GetTypeHash(Key)); }
};

struct FStandaloneDesiredTileSet
{
    std::vector<FVoxelTileKey> Tiles;
    FIntVector CenterChunk=FIntVector::ZeroValue;
    int32 MinimumChunkZ=MIN_int32;
    int32 MaximumChunkZ=MAX_int32;
    int32 StrateTopChunkZ=0;
    int32 StrateBottomChunkZ=0;
    bool bHasStrate=false;
};

struct FStandaloneCpuTile
{
    FVoxelTileKey Key;
    uint64 RequestedCrossing=0;
    std::vector<FStandaloneCpuVertex> Vertices;
    std::vector<uint32> Indices;
    uint64 DensitySamples=0;
    double WorkerSeconds=0.0;
    std::array<uint64,VoxelDensityProfile::BucketCount> ProfileCycles{};
};

struct FStandaloneStreamingUpdate
{
    FIntVector CenterChunk=FIntVector::ZeroValue;
    int32 DesiredTiles=0;
    int32 RetainedResidentTiles=0;
    int32 NewJobs=0;
    int32 CancelledQueuedJobs=0;
    int32 FullResTiles=0;
    std::vector<std::pair<uint64,uint64>> CancelledByCrossing;
};

struct FStandaloneTileJob
{
    FVoxelTileKey Key;
    int32 BandZMin=MIN_int32;
    int32 BandZMax=MAX_int32;
    uint64 RequestedCrossing=0;
    double PriorityDistanceSquared=0.0;
};

struct FStandaloneTileJobHeapOrder
{
    bool operator()(const FStandaloneTileJob& A,const FStandaloneTileJob& B)const
    {
        if(A.PriorityDistanceSquared!=B.PriorityDistanceSquared)
            return A.PriorityDistanceSquared>B.PriorityDistanceSquared;
        if(A.Key.Level!=B.Key.Level)return A.Key.Level>B.Key.Level;
        if(A.Key.Coord.Z!=B.Key.Coord.Z)return A.Key.Coord.Z>B.Key.Coord.Z;
        if(A.Key.Coord.Y!=B.Key.Coord.Y)return A.Key.Coord.Y>B.Key.Coord.Y;
        return A.Key.Coord.X>B.Key.Coord.X;
    }
};

namespace StandaloneWorldMesher
{
    inline double TileDistanceSquaredChunks(const FVoxelTileKey& Tile,const FIntVector& Center)
    {
        const double Scale=static_cast<double>(1<<Tile.Level);
        const double DX=(static_cast<double>(Tile.Coord.X)+0.5)*Scale-Center.X;
        const double DY=(static_cast<double>(Tile.Coord.Y)+0.5)*Scale-Center.Y;
        const double DZ=(static_cast<double>(Tile.Coord.Z)+0.5)*Scale-Center.Z;
        return DX*DX+DY*DY+DZ*DZ;
    }

    inline FStandaloneDesiredTileSet BuildDesiredSet(
        const FStandaloneOwnerWorld& World,const FVector& CameraPositionVoxels)
    {
        FStandaloneDesiredTileSet Desired;
        Desired.CenterChunk=CameraChunk(CameraPositionVoxels);
        if(World.Settings->bClampViewToStrate)
            Desired.bHasStrate=World.Manager->GetStrateChunkZBounds(
                Desired.CenterChunk.Z,Desired.StrateTopChunkZ,Desired.StrateBottomChunkZ);
        VoxelClipmapDesiredTiles::ResolveVerticalBounds(
            Desired.CenterChunk.Z,World.Settings->ViewDistanceUp,World.Settings->ViewDistanceDown,
            World.Settings->StrateViewMarginChunks,World.Settings->bClampViewToStrate,
            Desired.bHasStrate,Desired.StrateTopChunkZ,Desired.StrateBottomChunkZ,
            Desired.MinimumChunkZ,Desired.MaximumChunkZ);
        VoxelClipmapDesiredTiles::FParameters Parameters;
        Parameters.ClipRadius=World.ClipRadius;
        Parameters.MaxClipLevel=FMath::Clamp(World.MaxClipLevel,0,8);
        Parameters.RenderDistanceChunks=World.RenderDistanceChunks;
        TArray<FVoxelTileKey> Tiles;
        VoxelClipmapDesiredTiles::Build(Desired.CenterChunk,Desired.MinimumChunkZ,
            Desired.MaximumChunkZ,Parameters,Tiles);
        Desired.Tiles.reserve(static_cast<size_t>(Tiles.Num()));
        for(const FVoxelTileKey& Tile:Tiles)Desired.Tiles.push_back(Tile);
        std::sort(Desired.Tiles.begin(),Desired.Tiles.end(),[&](const FVoxelTileKey& A,const FVoxelTileKey& B)
        {
            const double DA=TileDistanceSquaredChunks(A,Desired.CenterChunk);
            const double DB=TileDistanceSquaredChunks(B,Desired.CenterChunk);
            return DA==DB?A.Level<B.Level:DA<DB;
        });
        return Desired;
    }
}

class FStandaloneStreamingGenerator
{
public:
    explicit FStandaloneStreamingGenerator(FStandaloneOwnerWorld& InWorld,int32 InWorkerCount=6)
        :World(&InWorld),WorkerCount(FMath::Clamp(InWorkerCount,1,32))
    {
        VoxelGenLOD::InitializeConsoleSwitches();
        UVoxelMarchingCubesMesher::InitializeConsoleSwitches();
    }
    FStandaloneStreamingGenerator(const FStandaloneStreamingGenerator&)=delete;
    FStandaloneStreamingGenerator& operator=(const FStandaloneStreamingGenerator&)=delete;
    ~FStandaloneStreamingGenerator(){Stop();}

    FStandaloneStreamingUpdate Start(const FVector& CameraPositionVoxels,
                                     const std::vector<FVoxelTileKey>& ResidentKeys={},
                                     uint64 CrossingIndex=0)
    {
        Stop();
        {
            std::lock_guard<std::mutex> Lock(Mutex);
            bStop=false;
        }
        for(int32 I=0;I<WorkerCount;++I)Workers.emplace_back([this](){WorkerLoop();});
        return UpdateDesired(CameraPositionVoxels,ResidentKeys,CrossingIndex);
    }

    void Stop()
    {
        {
            std::lock_guard<std::mutex> Lock(Mutex);
            bStop=true;
            Queue.clear();
        }
        Wake.notify_all();
        for(std::thread& Worker:Workers)if(Worker.joinable())Worker.join();
        Workers.clear();
        std::lock_guard<std::mutex> Lock(Mutex);
        Ready.clear();Outstanding.clear();Desired.clear();
    }

    FStandaloneStreamingUpdate UpdateDesired(
        const FVector& CameraPositionVoxels,const std::vector<FVoxelTileKey>& ResidentKeys,
        uint64 CrossingIndex=0)
    {
        FStandaloneDesiredTileSet NewDesired=StandaloneWorldMesher::BuildDesiredSet(*World,CameraPositionVoxels);
        FStandaloneStreamingUpdate Stats;
        Stats.CenterChunk=NewDesired.CenterChunk;
        Stats.DesiredTiles=static_cast<int32>(NewDesired.Tiles.size());
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> Resident;
        Resident.reserve(ResidentKeys.size()*2+1);
        for(const FVoxelTileKey& Key:ResidentKeys)Resident.insert(Key);
        for(const FVoxelTileKey& Key:NewDesired.Tiles)
        {
            if(Key.Level<World->FullResClipLevels)++Stats.FullResTiles;
            if(Resident.contains(Key))++Stats.RetainedResidentTiles;
        }

        std::lock_guard<std::mutex> Lock(Mutex);
        Desired.clear();Desired.reserve(NewDesired.Tiles.size()*2+1);
        for(const FVoxelTileKey& Key:NewDesired.Tiles)Desired.insert(Key);
        FullResTileCount=Stats.FullResTiles;
        std::unordered_map<uint64,uint64> CancelledByCrossing;
        auto QueueEnd=std::remove_if(Queue.begin(),Queue.end(),[&](const FStandaloneTileJob& Job)
        {
            if(Desired.contains(Job.Key))return false;
            Outstanding.erase(Job.Key);++Stats.CancelledQueuedJobs;
            ++CancelledByCrossing[Job.RequestedCrossing];return true;
        });
        Queue.erase(QueueEnd,Queue.end());
        Stats.CancelledByCrossing.reserve(CancelledByCrossing.size());
        for(const auto& Pair:CancelledByCrossing)Stats.CancelledByCrossing.push_back(Pair);

        for(FStandaloneTileJob& Job:Queue)
            Job.PriorityDistanceSquared=StandaloneWorldMesher::TileDistanceSquaredChunks(
                Job.Key,NewDesired.CenterChunk);

        const int32 BandMin=World->StrateContentCutMinLevel<9&&NewDesired.bHasStrate
            ? NewDesired.StrateBottomChunkZ*CHUNK_SIZE:MIN_int32;
        const int32 BandMax=World->StrateContentCutMinLevel<9&&NewDesired.bHasStrate
            ? (NewDesired.StrateTopChunkZ+1)*CHUNK_SIZE:MAX_int32;
        for(const FVoxelTileKey& Key:NewDesired.Tiles)
        {
            if(Resident.contains(Key)||Outstanding.contains(Key))continue;
            Queue.push_back({Key,BandMin,BandMax,CrossingIndex,
                StandaloneWorldMesher::TileDistanceSquaredChunks(Key,NewDesired.CenterChunk)});
            Outstanding.insert(Key);++Stats.NewJobs;
        }
        CurrentDesired=std::move(NewDesired);
        std::make_heap(Queue.begin(),Queue.end(),FStandaloneTileJobHeapOrder{});
        Wake.notify_all();
        return Stats;
    }

    std::vector<FStandaloneCpuTile> TakeReady()
    {
        std::vector<FStandaloneCpuTile> Out;
        std::lock_guard<std::mutex> Lock(Mutex);
        Out.swap(Ready);
        for(const FStandaloneCpuTile& Tile:Out)Outstanding.erase(Tile.Key);
        return Out;
    }

    bool IsDesired(const FVoxelTileKey& Key)const
    {std::lock_guard<std::mutex> Lock(Mutex);return Desired.contains(Key);}
    int32 DesiredCount()const
    {std::lock_guard<std::mutex> Lock(Mutex);return static_cast<int32>(Desired.size());}
    int32 FullResDesiredCount()const
    {std::lock_guard<std::mutex> Lock(Mutex);return FullResTileCount;}
    FIntVector CenterChunk()const
    {std::lock_guard<std::mutex> Lock(Mutex);return CurrentDesired.CenterChunk;}
    int32 GetWorkerCount()const{return WorkerCount;}

private:
    FStandaloneCpuTile Generate(const FStandaloneTileJob& Job)
    {
        const auto StartTime=FClock::now();
        FStandaloneCpuTile Out;Out.Key=Job.Key;Out.RequestedCrossing=Job.RequestedCrossing;
        UVoxelMarchingCubesMesher Mesher;
        Mesher.SetGenerator(World->Generator.get());
        Mesher.bGenerateSkirts=World->bGenerateSkirts;
        Mesher.SkirtCells=static_cast<float>(World->SkirtCells);
        Mesher.LODOctaveDrop=World->LODOctaveDrop;
        const int32 Level=Job.Key.Level;
        const int32 Extent=Job.Key.ExtentVoxels();
        const int32 Cells=Level<World->FullResClipLevels?CHUNK_SIZE:FMath::Max(1,World->CoarseTileCells);
        const int32 Step=FMath::Max(1,Extent/Cells);
        int64 SampleCount=0;
        const bool bProfile=VoxelDensityProfile::GetMode()!=VoxelDensityProfile::EMode::Disabled;
        const VoxelDensityProfile::FSnapshot Before=bProfile
            ? VoxelDensityProfile::SnapshotCurrentThread():VoxelDensityProfile::FSnapshot{};
        FVoxelMeshData Mesh=Mesher.GenerateMesh(Job.Key.OriginVoxels(),Step,Cells,nullptr,
            Job.BandZMin,Job.BandZMax,&SampleCount);
        if(bProfile)
        {
            const VoxelDensityProfile::FSnapshot After=VoxelDensityProfile::SnapshotCurrentThread();
            for(size_t I=0;I<Out.ProfileCycles.size();++I)
            {
                const size_t Bucket=I;
                Out.ProfileCycles[I]=After.Cycles[Bucket]>=Before.Cycles[Bucket]
                    ?After.Cycles[Bucket]-Before.Cycles[Bucket]:0;
            }
        }
        Out.DensitySamples=static_cast<uint64>(FMath::Max<int64>(0,SampleCount));
        Out.Vertices.reserve(static_cast<size_t>(Mesh.Vertices.Num()));
        for(int32 I=0;I<Mesh.Vertices.Num();++I)
        {
            const FVector& P=Mesh.Vertices[I];const FVector& N=Mesh.Normals[I];
            Out.Vertices.push_back({
                {static_cast<float>(P.X*0.01),static_cast<float>(P.Y*0.01),static_cast<float>(P.Z*0.01)},
                {static_cast<float>(N.X),static_cast<float>(N.Y),static_cast<float>(N.Z)}});
        }
        Out.Indices.reserve(static_cast<size_t>(Mesh.Triangles.Num()));
        for(const int32 Index:Mesh.Triangles)Out.Indices.push_back(static_cast<uint32>(Index));
        Out.WorkerSeconds=std::chrono::duration<double>(FClock::now()-StartTime).count();
        return Out;
    }

    void WorkerLoop()
    {
        for(;;)
        {
            FStandaloneTileJob Job;
            {
                std::unique_lock<std::mutex> Lock(Mutex);
                Wake.wait(Lock,[&](){return bStop||!Queue.empty();});
                if(bStop)return;
                std::pop_heap(Queue.begin(),Queue.end(),FStandaloneTileJobHeapOrder{});
                Job=Queue.back();Queue.pop_back();
            }
            FStandaloneCpuTile Tile=Generate(Job);
            {
                std::lock_guard<std::mutex> Lock(Mutex);
                if(bStop)return;
                Ready.push_back(std::move(Tile));
            }
        }
    }

    using FClock=std::chrono::steady_clock;
    FStandaloneOwnerWorld* World=nullptr;
    int32 WorkerCount=6;
    mutable std::mutex Mutex;
    std::condition_variable Wake;
    bool bStop=true;
    std::vector<std::thread> Workers;
    std::vector<FStandaloneTileJob> Queue;
    std::vector<FStandaloneCpuTile> Ready;
    std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> Desired;
    std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> Outstanding;
    FStandaloneDesiredTileSet CurrentDesired;
    int32 FullResTileCount=0;
};