build(windows): add Nvidia and AMD GPU high performance exports

Replace std::shared_mutex and std::unordered_set with tbb::concurrent_hash_map for chunk storage, tbb::concurrent_queue for pending uploads, and absl::flat_hash_set for player chunk positions. This improves thread safety and reduces locking overhead. Refactor receive_chunk and request_chunk to work with the new concurrent model.

include/Cubed/gameplay/client_player.hpp

@@ -6,6 +6,7 @@
 #include "Cubed/gameplay/game_mode.hpp"
 #include "Cubed/input.hpp"
 
+#include <absl/container/flat_hash_set.h>
 #include <glm/glm.hpp>
 #include <optional>
 #include <shared_mutex>
@@ -14,8 +15,14 @@ enum class Gait { WALK = 0, RUN };
 class ClientWorld;
 class ClientPlayer {
 public:
+    using ChunkPosSet = absl::flat_hash_set<ChunkPos, ChunkPos::Hash>;
     ClientPlayer(ClientWorld& world);
     ~ClientPlayer();
+
+    void update_chunk_set(const ChunkPosSet& set);
+    const ChunkPosSet& get_chunk_pos_set() const;
+    ChunkPosSet& get_chunk_pos_set();
+
     AABB get_aabb(const glm::vec3& pos) const;
     const glm::vec3& get_front() const;
     const Gait& get_gait() const;
@@ -101,7 +108,8 @@ private:
     ClientWorld& m_world;
 
     mutable std::shared_mutex m_player_pos_mutex;
-
+    mutable std::shared_mutex m_chunk_pos_mutex;
+    ChunkPosSet m_player_chunk_pos_set;
     bool ray_cast(const glm::vec3& start, const glm::vec3& dir,
                   glm::ivec3& block_pos, glm::vec3& normal,
                   float distance = 4.0f);

include/Cubed/gameplay/client_world.hpp

@@ -7,7 +7,10 @@
 #include "Cubed/gameplay/network_client.hpp"
 #include "Cubed/tools/thread_pool.hpp"
 
+#include <absl/container/flat_hash_set.h>
 #include <deque>
+#include <tbb/concurrent_hash_map.h>
+#include <tbb/concurrent_queue.h>
 #include <tbb/concurrent_unordered_map.h>
 namespace Cubed {
 
@@ -64,10 +67,10 @@ public:
     const std::vector<ChunkRenderSnapshot>& render_snapshots() const;
     const std::vector<RemotePlayerRenderData>& render_player_data() const;
     glm::vec3 sunlight_dir() const;
-    void receive_chunk(ChunkDataRsp data);
+    void receive_chunk(std::vector<uint8_t> data, PacketHeader header);
     void request_exit();
     bool is_receive_exit();
-
+    int chunk_size() const;
     template <typename Fn>
     void register_timer(std::string_view id, TickType threshold, Fn&& f) {
         m_timers.emplace(std::piecewise_construct,
@@ -78,12 +81,14 @@ public:
 private:
     enum class ChunkLoadStyle { RANDOM, CENTER };
     using ChunkHashMap =
-        tbb::concurrent_unordered_map<ChunkPos, ClientChunk, ChunkPos::Hash>;
+        tbb::concurrent_hash_map<ChunkPos, std::shared_ptr<ClientChunk>,
-    using ChunkPosSet = std::unordered_set<ChunkPos, ChunkPos::Hash>;
+                                 ChunkPos::TBBHash>;
+    using ChunkPosSet = absl::flat_hash_set<ChunkPos, ChunkPos::Hash>;
     using ChunkPosVector = std::vector<ChunkPos>;
     using OtherPlayerHashMap =
         std::unordered_map<std::string, RemotePlayerInfo>;
-
+    using chunk_acc = ChunkHashMap::accessor;
+    using chunk_cacc = ChunkHashMap::const_accessor;
     static constexpr int WORLD_EXIT_TIMEOUT = 200;
 
     ClientPlayer m_player;
@@ -92,13 +97,11 @@ private:
     std::vector<glm::vec4> m_planes;
     std::jthread m_client_thread;
 
-    mutable std::shared_mutex m_chunks_mutex;
     std::mutex m_delete_vbo_mutex;
     std::mutex m_delete_vao_mutex;
-    std::mutex m_pending_upload_queue_mutex;
     std::mutex m_other_players_mutex;
 
-    std::deque<ClientChunk> m_pending_upload_queue;
+    tbb::concurrent_queue<std::unique_ptr<ClientChunk>> m_pending_upload_queue;
 
     std::vector<GLuint> m_pending_delete_vbo;
     std::vector<GLuint> m_pending_delete_vao;
@@ -127,5 +130,7 @@ private:
     void report_player_pos();
 
     void set_block(const glm::ivec3& pos, unsigned id);
+
+    void update_chunk(const ChunkPosSet& old, const ChunkPosSet& now);
 };
 } // namespace Cubed

src/dev_panel.cpp

@@ -576,6 +576,7 @@ void DevPanel::show_client_world_table_bar() {
         m_player->set_player_pos({0.0f, 255.0f, 0.0f});
     }
     ImGui::Text("Chunk Task Id %d", m_app.client_world().get_chunk_task_id());
+    ImGui::Text("Client World Chunk %d", m_app.client_world().chunk_size());
 }
 
 void DevPanel::show_player_tab_item() {

src/gameplay/client_player.cpp

@@ -542,6 +542,22 @@ void ClientPlayer::update_scroll(double yoffset) {
     }
 }
 
+void ClientPlayer::update_chunk_set(const ChunkPosSet& set) {
+    std::lock_guard lock(m_chunk_pos_mutex);
+    m_player_chunk_pos_set.clear();
+    m_player_chunk_pos_set.insert(set.begin(), set.end());
+}
+
+const ClientPlayer::ChunkPosSet& ClientPlayer::get_chunk_pos_set() const {
+    std::shared_lock lock(m_chunk_pos_mutex);
+    return m_player_chunk_pos_set;
+}
+
+ClientPlayer::ChunkPosSet& ClientPlayer::get_chunk_pos_set() {
+    std::lock_guard lock(m_chunk_pos_mutex);
+    return m_player_chunk_pos_set;
+}
+
 float& ClientPlayer::max_walk_speed() { return m_max_walk_speed; }
 float& ClientPlayer::max_run_speed() { return m_max_run_speed; }
 float& ClientPlayer::max_speed() { return m_max_speed; }

src/gameplay/client_world.cpp

@@ -23,10 +23,9 @@ ClientWorld::ClientWorld() : m_player(*this) {}
 ClientWorld::~ClientWorld() {
     stop_client_thread();
     stop_thread_pool();
-    {
+
-        std::lock_guard lock(m_chunks_mutex);
     m_chunks.clear();
-    }
+
     {
         std::lock_guard lk(m_delete_vbo_mutex);
         for (auto x : m_pending_delete_vbo) {
@@ -53,14 +52,13 @@ ClientPlayer& ClientWorld::get_player() { return m_player; }
 
 int ClientWorld::get_block(const glm::ivec3& block_pos) const {
     auto [chunk_x, chunk_z] = get_chunk_pos(block_pos.x, block_pos.z);
-    std::shared_lock lk(m_chunks_mutex);
+    chunk_cacc cacc;
-    auto it = m_chunks.find(ChunkPos{chunk_x, chunk_z});
 
-    if (it == m_chunks.end()) {
+    if (!m_chunks.find(cacc, ChunkPos{chunk_x, chunk_z})) {
         return 0;
     }
 
-    const auto& chunk_blocks = it->second.get_chunk_blocks();
+    const auto& chunk_blocks = cacc->second->get_chunk_blocks();
     auto [x, y, z] = ClientChunk::world_to_block(block_pos, {chunk_x, chunk_z});
     if (x < 0 || y < 0 || z < 0 || x >= CHUNK_SIZE || y >= WORLD_SIZE_Y ||
         z >= CHUNK_SIZE) {
@@ -70,13 +68,12 @@ int ClientWorld::get_block(const glm::ivec3& block_pos) const {
 }
 bool ClientWorld::is_solid(const glm::ivec3& block_pos) const {
     auto [chunk_x, chunk_z] = get_chunk_pos(block_pos.x, block_pos.z);
-    std::shared_lock lk(m_chunks_mutex);
+    chunk_cacc cacc;
-    auto it = m_chunks.find(ChunkPos{chunk_x, chunk_z});
 
-    if (it == m_chunks.end()) {
+    if (!m_chunks.find(cacc, ChunkPos{chunk_x, chunk_z})) {
         return false;
     }
-    const auto& chunk_blocks = it->second.get_chunk_blocks();
+    const auto& chunk_blocks = cacc->second->get_chunk_blocks();
     auto [x, y, z] = ClientChunk::world_to_block(block_pos, {chunk_x, chunk_z});
     if (x < 0 || y < 0 || z < 0 || x >= CHUNK_SIZE || y >= WORLD_SIZE_Y ||
         z >= CHUNK_SIZE) {
@@ -91,13 +88,12 @@ bool ClientWorld::is_solid(const glm::ivec3& block_pos) const {
 }
 bool ClientWorld::can_pass_block(const glm::ivec3& block_pos) const {
     auto [chunk_x, chunk_z] = get_chunk_pos(block_pos.x, block_pos.z);
-    std::shared_lock lk(m_chunks_mutex);
+    chunk_cacc cacc;
-    auto it = m_chunks.find(ChunkPos{chunk_x, chunk_z});
 
-    if (it == m_chunks.end()) {
+    if (!m_chunks.find(cacc, ChunkPos{chunk_x, chunk_z})) {
         return true;
     }
-    const auto& chunk_blocks = it->second.get_chunk_blocks();
+    const auto& chunk_blocks = cacc->second->get_chunk_blocks();
     auto [x, y, z] = ClientChunk::world_to_block(block_pos, {chunk_x, chunk_z});
     if (x < 0 || y < 0 || z < 0 || x >= CHUNK_SIZE || y >= WORLD_SIZE_Y ||
         z >= CHUNK_SIZE) {
@@ -111,31 +107,31 @@ void ClientWorld::rebuild_world() {
     if (m_is_rebuilding.exchange(true)) {
         return;
     }
+
+    stop_client_thread();
     stop_thread_pool();
-    {
+
-        std::lock_guard lk(m_chunks_mutex);
     m_chunks.clear();
-    }
+
-    {
-        std::lock_guard lock(m_pending_upload_queue_mutex);
     m_pending_upload_queue.clear();
-    }
+
     start_thread_pool();
+    start_client_thread(m_player.get_uuid());
     request_chunk();
     m_is_rebuilding = false;
 }
 
 BlockType ClientWorld::get_block_tpye(const glm::ivec3& block_pos) const {
     auto [chunk_x, chunk_z] = get_chunk_pos(block_pos.x, block_pos.z);
-    std::shared_lock lk(m_chunks_mutex);
+    chunk_cacc cacc;
-    auto it = m_chunks.find(ChunkPos{chunk_x, chunk_z});
+    ;
 
-    if (it == m_chunks.end()) {
+    if (!m_chunks.find(cacc, ChunkPos{chunk_x, chunk_z})) {
         // Logger::error("Can't Find Block {} {} {}", block_pos.x, block_pos.y,
         //               block_pos.z);
         return 0;
     }
-    const auto& chunk_blocks = it->second.get_chunk_blocks();
+    const auto& chunk_blocks = cacc->second->get_chunk_blocks();
     auto [x, y, z] = ClientChunk::world_to_block(block_pos, {chunk_x, chunk_z});
     if (x < 0 || y < 0 || z < 0 || x >= CHUNK_SIZE || y >= WORLD_SIZE_Y ||
         z >= CHUNK_SIZE) {
@@ -152,10 +148,10 @@ void ClientWorld::set_block(const glm::ivec3& block_pos, unsigned id) {
     world_z = block_pos.z;
 
     auto [chunk_x, chunk_z] = get_chunk_pos(world_x, world_z);
-    std::lock_guard lk(m_chunks_mutex);
+    {
-    auto it = m_chunks.find(ChunkPos{chunk_x, chunk_z});
+        chunk_acc acc;
 
-    if (it == m_chunks.end()) {
+        if (!m_chunks.find(acc, ChunkPos{chunk_x, chunk_z})) {
             return;
         }
 
@@ -166,7 +162,8 @@ void ClientWorld::set_block(const glm::ivec3& block_pos, unsigned id) {
             return;
         }
 
-    it->second.set_chunk_block(ClientChunk::index(x, y, z), id);
+        acc->second->set_chunk_block(ClientChunk::index(x, y, z), id);
+    }
 
     static const glm::ivec3 NEIGHBOR_DIRS[] = {
         {1, 0, 0}, {-1, 0, 0}, {0, 0, -1}, {0, 0, 1}};
@@ -175,9 +172,11 @@ void ClientWorld::set_block(const glm::ivec3& block_pos, unsigned id) {
         glm::ivec3 neighbor = block_pos + dir;
 
         auto [cx, cz] = get_chunk_pos(neighbor.x, neighbor.z);
-        auto it = m_chunks.find({cx, cz});
+        {
-        if (it != m_chunks.end()) {
+            chunk_acc acc;
-            it->second.mark_dirty();
+            if (m_chunks.find(acc, {cx, cz})) {
+                acc->second->mark_dirty();
+            }
         }
     }
 }
@@ -251,7 +250,6 @@ void ClientWorld::receive_player_logout(const LogoutRsp& rsp) {
 
 void ClientWorld::init(std::string_view player_name,
                        std::shared_ptr<NetworkClient> client) {
-    m_chunks.reserve(MAX_DISTANCE * MAX_DISTANCE * 4);
     m_player.init(player_name);
     m_client = client;
     // timer
@@ -361,6 +359,26 @@ void ClientWorld::report_player_pos() {
     m_client->send(make_packet(*pos));
 }
 
+void ClientWorld::update_chunk(const ChunkPosSet& old, const ChunkPosSet& now) {
+
+    // Elements in the old set that are not contained in now are not needed by
+    // the current player.
+
+    for (auto& pos : old) {
+        if (!now.contains(pos)) {
+
+            chunk_acc acc;
+            if (!m_chunks.find(acc, pos)) {
+                Logger::warn("Update Ref Count Error, can't Find old pos "
+                             "in m_chunks");
+                continue;
+            }
+
+            m_chunks.erase(acc);
+        }
+    }
+}
+
 void ClientWorld::request_chunk() {
     if (m_requesting_chunk.exchange(true)) {
         Logger::warn("It is requesting new chunk!");
@@ -386,24 +404,20 @@ void ClientWorld::request_chunk() {
         }
     }
 
+    ChunkPosSet old = std::move(m_player.get_chunk_pos_set());
+    m_player.update_chunk_set(required_chunks);
+
     ChunkPosVector need_send_pos;
-    {
-        std::lock_guard lk(m_chunks_mutex);
-        for (auto it = m_chunks.begin(); it != m_chunks.end();) {
-            if (required_chunks.find(it->first) == required_chunks.end()) {
-                it = m_chunks.unsafe_erase(it);
-            } else {
-                ++it;
-            }
-        }
 
     for (auto pos : required_chunks) {
-            auto it = m_chunks.find(pos);
+        chunk_cacc cacc;
-            if (it == m_chunks.end()) {
+        if (!m_chunks.find(cacc, pos)) {
             need_send_pos.emplace_back(pos);
         }
     }
-    }
+
+    update_chunk(old, required_chunks);
+
     if (need_send_pos.empty()) {
         m_requesting_chunk = false;
         return;
@@ -416,9 +430,8 @@ void ClientWorld::request_chunk() {
     case CENTER: {
 
         glm::vec3 player_pos = m_player.get_player_pos();
-        auto dist2 = [player_pos](ChunkPos chunk_pos) {
+        ChunkPos player_chunk_pos = get_chunk_pos(player_pos.x, player_pos.z);
-            ChunkPos player_chunk_pos =
+        auto dist2 = [player_chunk_pos](ChunkPos chunk_pos) {
-                get_chunk_pos(player_pos.x, player_pos.z);
             float dx = player_chunk_pos.x - chunk_pos.x;
             float dz = player_chunk_pos.z - chunk_pos.z;
             return dx * dx + dz * dz;
@@ -446,36 +459,48 @@ void ClientWorld::request_chunk() {
     m_requesting_chunk = false;
 }
 
-void ClientWorld::receive_chunk(ChunkDataRsp data) {
+void ClientWorld::receive_chunk(std::vector<uint8_t> raw_data,
-    if (data.task_id() < m_chunk_task_id) {
+                                PacketHeader header) {
-        return;
-    }
 
-    {
-        std::lock_guard lock(m_chunks_mutex);
-        ChunkPos pos{data.pos().x(), data.pos().z()};
-        if (m_chunks.find(pos) != m_chunks.end()) {
-            Logger::warn("Chunk {} {} has already in client world", pos.x,
-                         pos.z);
-            return;
-        }
-    }
     // vertex data will genrator in  client thread pool instead of net thread;
     auto pool = m_thread_pool.load();
     if (!pool) {
         Logger::error("Client Thread Pool is nullptr");
         return;
     }
-    pool->enqueue([this, data = std::move(data)]() {
+    pool->enqueue(
-        ClientChunk chunk{*this};
+        [this, raw_data = std::move(raw_data), header = std::move(header)]() {
-        chunk.receive_chunk(data);
+            Arena arena;
-        {
+            auto* data = Arena::Create<ChunkDataRsp>(&arena);
-            std::lock_guard lock(m_pending_upload_queue_mutex);
+            if (!decode_packet(*data, raw_data, header)) {
-            m_pending_upload_queue.emplace_back(std::move(chunk));
+                return;
             }
+
+            if (data->task_id() < m_chunk_task_id) {
+                return;
+            }
+
+            {
+                chunk_cacc cacc;
+                ChunkPos pos{data->pos().x(), data->pos().z()};
+                if (m_chunks.find(cacc, pos)) {
+                    Logger::warn("Chunk {} {} has already in client world",
+                                 pos.x, pos.z);
+                    return;
+                }
+            }
+
+            std::unique_ptr<ClientChunk> chunk =
+                std::make_unique<ClientChunk>(*this);
+            chunk->receive_chunk(*data);
+
+            m_pending_upload_queue.emplace(std::move(chunk));
         });
 }
 bool ClientWorld::is_receive_exit() { return m_receive_exit; }
+
+int ClientWorld::chunk_size() const { return m_chunks.size(); }
+
 void ClientWorld::request_exit() {
     if (m_receive_exit) {
         return;
@@ -512,51 +537,60 @@ void ClientWorld::update(float delta_time) {
         }
         m_pending_delete_vao.clear();
     }
-    std::vector<ClientChunk> new_chunks;
+    std::vector<std::unique_ptr<ClientChunk>> new_chunks;
     {
-        std::lock_guard lock(m_pending_upload_queue_mutex);
+        std::unique_ptr<ClientChunk> chunk;
-        for (auto& c : m_pending_upload_queue) {
+        while (m_pending_upload_queue.try_pop(chunk)) {
-            // Logger::info("{} {}", c.get_chunk_pos().x, c.get_chunk_pos().z);
+            new_chunks.emplace_back(std::move(chunk));
-            new_chunks.emplace_back(std::move(c));
         }
-        m_pending_upload_queue.clear();
     }
     for (auto& c : new_chunks) {
-        c.upload_to_gpu();
+        c->upload_to_gpu();
     }
     {
-        std::lock_guard lock(m_chunks_mutex);
+
         for (auto& c : new_chunks) {
-            m_chunks.emplace(c.get_chunk_pos(), std::move(c));
+            m_chunks.emplace(c->get_chunk_pos(), std::move(c));
         }
         m_render_snapshots.clear();
-        for (auto& [pos, chunk] : m_chunks) {
+        auto chunk_pos_set = m_player.get_chunk_pos_set();
-            if (chunk.is_dirty()) {
+        for (auto& pos : chunk_pos_set) {
+            std::shared_ptr<ClientChunk> chunk;
+            {
+                chunk_acc acc;
+                if (m_chunks.find(acc, pos)) {
+                    chunk = acc->second;
+                }
+            }
+            if (!chunk) {
+                continue;
+            }
+            if (chunk->is_dirty()) {
                 // the curial fator influence
                 OptionalBlockVectorArray neighbor_block;
                 for (int i = 0; i < 4; i++) {
-                    auto it = m_chunks.find(pos + CHUNK_DIR[i]);
+                    chunk_cacc cacc;
-                    if (it != m_chunks.end()) {
+                    if (m_chunks.find(cacc, pos + CHUNK_DIR[i])) {
-                        neighbor_block[i] = (it->second.get_chunk_blocks());
+                        neighbor_block[i] = (cacc->second->get_chunk_blocks());
                     } else {
                         neighbor_block[i] = std::nullopt;
                     }
                 }
-                chunk.gen_vertex_data(neighbor_block);
+                chunk->gen_vertex_data(neighbor_block);
-                chunk.upload_to_gpu();
+                chunk->upload_to_gpu();
             }
-            if (!chunk.is_dirty()) {
+            if (!chunk->is_dirty()) {
-                if (chunk.is_need_upload()) {
+                if (chunk->is_need_upload()) {
-                    chunk.upload_to_gpu();
+                    chunk->upload_to_gpu();
                 }
                 m_render_snapshots.push_back(
-                    {chunk.get_normal_vao(), chunk.get_normal_vertices_sum(),
+                    {chunk->get_normal_vao(), chunk->get_normal_vertices_sum(),
-                     chunk.get_cross_vao(), chunk.get_cross_vertices_sum(),
+                     chunk->get_cross_vao(), chunk->get_cross_vertices_sum(),
-                     chunk.get_normal_discard_vao(),
+                     chunk->get_normal_discard_vao(),
-                     chunk.get_normal_discard_vertices_sum(),
+                     chunk->get_normal_discard_vertices_sum(),
-                     chunk.get_normal_blend_vao(),
+                     chunk->get_normal_blend_vao(),
-                     chunk.get_normal_blend_vertices_sum(),
+                     chunk->get_normal_blend_vertices_sum(),
-                     chunk.get_water_vao(), chunk.get_water_vertices_sum(),
+                     chunk->get_water_vao(), chunk->get_water_vertices_sum(),
                      glm::vec3(static_cast<float>(pos.x * CHUNK_SIZE) +
                                    static_cast<float>(CHUNK_SIZE / 2),
                                static_cast<float>(WORLD_SIZE_Y / 2),

src/gameplay/network_client.cpp

@@ -59,6 +59,7 @@ asio::awaitable<void> NetworkClient::read_loop() {
 
                 throw std::runtime_error("invalid packet length");
             }
+            // maybe move, don't use it after switch!
             std::vector<uint8_t> body_data(header.compressed_size);
             if (header.compressed_size > 0) {
                 co_await asio::async_read(m_socket, asio::buffer(body_data),
@@ -80,12 +81,9 @@ asio::awaitable<void> NetworkClient::read_loop() {
                 }
             } break;
             case std::to_underlying(PacketEnum::CHUNK_DATA_RSP): {
-                ChunkDataRsp rsp;
                 // Logger::info("Client: Receive Chunk Data rsp, size {}mb",
                 //              body_data.size() / 1024.0f / 1024);
-                if (decode_packet(rsp, body_data, header)) {
+                m_world.receive_chunk(std::move(body_data), header);
-                    m_world.receive_chunk(std::move(rsp));
-                }
             } break;
             case std::to_underlying(PacketEnum::BLOCK_CHANGE_RSP): {
                 auto* rsp = Arena::Create<BlockChangeRsp>(&arena);

src/main.cpp

@@ -1,5 +1,14 @@
 #include "Cubed/app.hpp"
 
+#ifdef _WIN32
+
+extern "C" {
+__declspec(dllexport) unsigned long NvOptimusEnablement = 1;
+__declspec(dllexport) int AmdPowerXpressRequestHighPerformance = 1;
+}
+
+#endif
+
 int main(int argc, char** argv) {
 
     static_assert(sizeof(int) == sizeof(int32_t));