Merge pull request #74 from savisxss/dsdsdsds

src/EterLib/GameThreadPool.cpp

@@ -0,0 +1,158 @@
+#include "StdAfx.h"
+#include "GameThreadPool.h"
+
+CGameThreadPool::CGameThreadPool()
+	: m_bShutdown(false)
+	, m_bInitialized(false)
+	, m_iNextWorkerIndex(0)
+{
+}
+
+CGameThreadPool::~CGameThreadPool()
+{
+	Destroy();
+}
+
+void CGameThreadPool::Initialize(int iWorkerCount)
+{
+	if (m_bInitialized)
+	{
+		TraceError("CGameThreadPool::Initialize - Already initialized!");
+		return;
+	}
+
+	// Determine worker count
+	if (iWorkerCount <= 0)
+	{
+		iWorkerCount = static_cast<int>(std::thread::hardware_concurrency());
+		if (iWorkerCount <= 0)
+			iWorkerCount = 4; // Fallback to 4 workers
+	}
+
+	// Clamp worker count to reasonable range
+	iWorkerCount = std::max(2, std::min(16, iWorkerCount));
+
+	Tracef("CGameThreadPool::Initialize - Creating %d worker threads\n", iWorkerCount);
+
+	m_bShutdown.store(false, std::memory_order_release);
+	m_workers.reserve(iWorkerCount);
+
+	// Initialize each worker
+	for (int i = 0; i < iWorkerCount; ++i)
+	{
+		std::unique_ptr<TWorkerThread> pWorker(new TWorkerThread());
+		pWorker->pTaskQueue.reset(new SPSCQueue<TTask>(QUEUE_SIZE));
+		pWorker->bBusy.store(false, std::memory_order_relaxed);
+		pWorker->uTaskCount.store(0, std::memory_order_relaxed);
+		pWorker->thread = std::thread(&CGameThreadPool::WorkerThreadProc, this, i);
+		m_workers.push_back(std::move(pWorker));
+	}
+
+	m_bInitialized.store(true, std::memory_order_release);
+}
+
+void CGameThreadPool::Destroy()
+{
+	if (!m_bInitialized)
+		return;
+
+	Tracef("CGameThreadPool::Destroy - Shutting down %d worker threads\n", GetWorkerCount());
+
+	m_bShutdown.store(true, std::memory_order_release);
+	m_bInitialized.store(false, std::memory_order_release);
+
+	// Join all worker threads
+	for (auto& pWorker : m_workers)
+	{
+		if (pWorker->thread.joinable())
+			pWorker->thread.join();
+	}
+
+	m_workers.clear();
+}
+
+void CGameThreadPool::WorkerThreadProc(int iWorkerIndex)
+{
+	TWorkerThread* pWorker = m_workers[iWorkerIndex].get();
+	int iIdleCount = 0;
+
+	while (!m_bShutdown.load(std::memory_order_acquire))
+	{
+		TTask task;
+		if (pWorker->pTaskQueue->Pop(task))
+		{
+			pWorker->bBusy.store(true, std::memory_order_relaxed);
+			iIdleCount = 0;
+
+			// Execute the task
+			try
+			{
+				task();
+			}
+			catch (const std::exception& e)
+			{
+				TraceError("CGameThreadPool::WorkerThreadProc - Exception in worker %d: %s", iWorkerIndex, e.what());
+			}
+			catch (...)
+			{
+				TraceError("CGameThreadPool::WorkerThreadProc - Unknown exception in worker %d", iWorkerIndex);
+			}
+
+			pWorker->uTaskCount.fetch_sub(1, std::memory_order_relaxed);
+			pWorker->bBusy.store(false, std::memory_order_relaxed);
+		}
+		else
+		{
+			// No work available - idle strategy
+			++iIdleCount;
+
+			if (iIdleCount < 100)
+			{
+				// Spin briefly for immediate work
+				std::this_thread::yield();
+			}
+			else if (iIdleCount < 1000)
+			{
+				// Short sleep for moderate idle
+				std::this_thread::sleep_for(std::chrono::microseconds(10));
+			}
+			else
+			{
+				// Longer sleep for extended idle
+				std::this_thread::sleep_for(std::chrono::milliseconds(1));
+			}
+		}
+	}
+}
+
+int CGameThreadPool::SelectLeastBusyWorker() const
+{
+	if (m_workers.empty())
+		return 0;
+
+	// Simple load balancing: find worker with least pending tasks
+	int iBestWorker = 0;
+	uint32_t uMinTasks = m_workers[0]->uTaskCount.load(std::memory_order_relaxed);
+
+	for (size_t i = 1; i < m_workers.size(); ++i)
+	{
+		uint32_t uTasks = m_workers[i]->uTaskCount.load(std::memory_order_relaxed);
+		if (uTasks < uMinTasks)
+		{
+			uMinTasks = uTasks;
+			iBestWorker = static_cast<int>(i);
+		}
+	}
+
+	return iBestWorker;
+}
+
+size_t CGameThreadPool::GetPendingTaskCount() const
+{
+	size_t uTotal = 0;
+	for (const auto& pWorker : m_workers)
+	{
+		uTotal += pWorker->uTaskCount.load(std::memory_order_relaxed);
+	}
+	return uTotal;
+}

src/EterLib/GameThreadPool.h

@@ -0,0 +1,132 @@
+#pragma once
+
+#include "SPSCQueue.h"
+#include "EterBase/Singleton.h"
+#include <thread>
+#include <vector>
+#include <functional>
+#include <future>
+#include <atomic>
+#include <memory>
+
+class CGameThreadPool : public CSingleton<CGameThreadPool>
+{
+public:
+	using TTask = std::function<void()>;
+
+	CGameThreadPool();
+	~CGameThreadPool();
+
+	// Initialize thread pool with specified worker count
+	// If count <= 0, uses hardware_concurrency
+	void Initialize(int iWorkerCount = -1);
+
+	// Shutdown and join all worker threads
+	void Destroy();
+
+	// Enqueue a task and get a future to track completion
+	template<typename TFunc>
+	std::future<void> Enqueue(TFunc&& func);
+
+	// Get number of active workers
+	int GetWorkerCount() const { return static_cast<int>(m_workers.size()); }
+
+	// Get approximate number of pending tasks across all queues
+	size_t GetPendingTaskCount() const;
+
+	// Check if pool is initialized
+	bool IsInitialized() const { return m_bInitialized; }
+
+private:
+	struct TWorkerThread
+	{
+		std::thread thread;
+		std::unique_ptr<SPSCQueue<TTask>> pTaskQueue;
+		std::atomic<bool> bBusy;
+		std::atomic<uint32_t> uTaskCount;
+
+		TWorkerThread()
+			: bBusy(false)
+			, uTaskCount(0)
+		{
+		}
+	};
+
+	void WorkerThreadProc(int iWorkerIndex);
+	int SelectLeastBusyWorker() const;
+
+	std::vector<std::unique_ptr<TWorkerThread>> m_workers;
+	std::atomic<bool> m_bShutdown;
+	std::atomic<bool> m_bInitialized;
+	std::atomic<int> m_iNextWorkerIndex; // For round-robin distribution
+
+	static const size_t QUEUE_SIZE = 8192;
+};
+
+// Template implementation
+template<typename TFunc>
+std::future<void> CGameThreadPool::Enqueue(TFunc&& func)
+{
+	if (!m_bInitialized)
+	{
+		// If not initialized, execute on calling thread
+		auto promise = std::make_shared<std::promise<void>>();
+		auto future = promise->get_future();
+		try
+		{
+			func();
+			promise->set_value();
+		}
+		catch (...)
+		{
+			promise->set_exception(std::current_exception());
+		}
+		return future;
+	}
+
+	// Create a promise to track task completion
+	auto promise = std::make_shared<std::promise<void>>();
+	auto future = promise->get_future();
+
+	// Wrap function in shared_ptr to avoid move issues with std::function
+	auto pFunc = std::make_shared<typename std::decay<TFunc>::type>(std::forward<TFunc>(func));
+
+	// Wrap the task with promise completion
+	TTask task = [promise, pFunc]()
+	{
+		try
+		{
+			(*pFunc)();
+			promise->set_value();
+		}
+		catch (...)
+		{
+			promise->set_exception(std::current_exception());
+		}
+	};
+
+	// Select worker with least load
+	int iWorkerIndex = SelectLeastBusyWorker();
+	TWorkerThread* pWorker = m_workers[iWorkerIndex].get();
+
+	// Try to enqueue the task
+	if (!pWorker->pTaskQueue->Push(std::move(task)))
+	{
+		// Queue is full, execute on calling thread as fallback
+		try
+		{
+			(*pFunc)();
+			promise->set_value();
+		}
+		catch (...)
+		{
+			promise->set_exception(std::current_exception());
+		}
+	}
+	else
+	{
+		pWorker->uTaskCount.fetch_add(1, std::memory_order_relaxed);
+	}
+
+	return future;
+}

src/EterLib/ResourceManager.cpp

@@ -252,6 +252,9 @@ void CResourceManager::RegisterResourceNewFunctionByTypePointer(int iType, CReso
 
 CResource * CResourceManager::InsertResourcePointer(DWORD dwFileCRC, CResource* pResource)
 {
+	// Thread-safe check and insert
+	std::lock_guard<std::mutex> lock(m_ResourceMapMutex);
+
 	TResourcePointerMap::iterator itor = m_pResMap.find(dwFileCRC);
 
 	if (m_pResMap.end() != itor)

src/EterLib/ResourceManager.h

@@ -7,6 +7,7 @@
 #include <set>
 #include <map>
 #include <string>
+#include <mutex>
 
 class CTextureCache;
 
@@ -76,6 +77,8 @@ class CResourceManager : public CSingleton<CResourceManager>
 		static CFileLoaderThread				ms_loadingThread;
 		CFileLoaderThreadPool*					m_pLoaderThreadPool;
 		CTextureCache*							m_pTextureCache;
+
+		mutable std::mutex						m_ResourceMapMutex;  // Thread-safe resource map access
 };
 
 extern int g_iLoadingDelayTime;

src/GameLib/ActorInstanceRender.cpp

@@ -32,6 +32,10 @@ void CActorInstance::SetMaterialAlpha(DWORD dwAlpha)
 
 void CActorInstance::OnRender()
 {
+	// Early out if race data is not loaded yet (async loading)
+	if (!m_pkCurRaceData)
+		return;
+
 	// MR-5: Fix effect rendering when actor is semi-transparent
 	// Credits to d1str4ught
 	if (GetAlphaValue() < 1.0f)

src/GameLib/AreaLoaderThread.cpp

@@ -5,6 +5,7 @@
 #include "StdAfx.h"
 
 #include "EterLib/ResourceManager.h"
+#include "EterLib/GameThreadPool.h"
 
 #include "AreaLoaderThread.h"
 #include "AreaTerrain.h"
@@ -14,213 +15,112 @@
 // Construction/Destruction
 //////////////////////////////////////////////////////////////////////
 
-TEMP_CAreaLoaderThread::TEMP_CAreaLoaderThread() : m_bShutdowned(false), m_pArg(NULL), m_hThread(NULL), m_uThreadID(0)
+TEMP_CAreaLoaderThread::TEMP_CAreaLoaderThread() : m_bShutdowned(false)
 {
 
 }
 
 TEMP_CAreaLoaderThread::~TEMP_CAreaLoaderThread()
 {
-	Destroy();
+	Shutdown();
 }
 
 bool TEMP_CAreaLoaderThread::Create(void * arg)
 {
-	Arg(arg);
-	m_hThread = (HANDLE) _beginthreadex(NULL, 0, EntryPoint, this, 0, &m_uThreadID);
-
-	if (!m_hThread)
-		return false;
-
-	SetThreadPriority(m_hThread, THREAD_PRIORITY_NORMAL);
+	// Modern implementation doesn't need explicit thread creation
+	// The global CGameThreadPool handles threading
+	m_bShutdowned = false;
 	return true;
 }
 
-UINT TEMP_CAreaLoaderThread::Run(void * arg)
-{
-	if (!Setup())
-		return 0;
-
-	return (Execute(arg));
-}
-
-/* Static */
-UINT CALLBACK TEMP_CAreaLoaderThread::EntryPoint(void * pThis)
-{
-	TEMP_CAreaLoaderThread * pThread = (TEMP_CAreaLoaderThread *) pThis;
-	return pThread->Run(pThread->Arg());
-}
-
-//////////////////////////////////////////////////////////////////////////
-void TEMP_CAreaLoaderThread::Destroy()
-{
-	if (m_hSemaphore)
-	{
-		CloseHandle(m_hSemaphore);
-		m_hSemaphore = NULL;
-	}
-
-/*
-	while(!m_pTerrainRequestDeque.empty())
-	{
-		CTerrain * pTerrain = m_pTerrainRequestDeque.front();
-		delete pTerrain;
-		pTerrain = NULL;
-		m_pTerrainRequestDeque.pop_front();
-	}
-
-	while(!m_pTerrainCompleteDeque.empty())
-	{
-		CTerrain * pTerrain = m_pTerrainCompleteDeque.front();
-		delete pTerrain;
-		pTerrain = NULL;
-		m_pTerrainCompleteDeque.pop_front();
-	}
-*/
-
-	/*stl_wipe(m_pTerrainRequestDeque);
-	stl_wipe(m_pTerrainCompleteDeque);
-	stl_wipe(m_pAreaRequestDeque);
-	stl_wipe(m_pAreaCompleteDeque);*/
-}
-
-UINT TEMP_CAreaLoaderThread::Setup()
-{
-	m_hSemaphore = CreateSemaphore(NULL,		// no security attributes
-								   0,			// initial count
-								   65535,		// maximum count
-								   NULL);		// unnamed semaphore
-	if (!m_hSemaphore)
-		return 0;
-
-	return 1;
-}
-
 void TEMP_CAreaLoaderThread::Shutdown()
 {
-	if (!m_hSemaphore)
-		return;
-
-	BOOL bRet;
-	
 	m_bShutdowned = true;
 
-	do
+	// Clear any pending completed items
 	{
-		bRet = ReleaseSemaphore(m_hSemaphore, 1, NULL);
-	}
-	while (!bRet);
-
-	WaitForSingleObject(m_hThread, 10000);	// 쓰레드가 종료 되기를 10초 기다림
-}
-
-UINT TEMP_CAreaLoaderThread::Execute(void * pvArg)
-{
-	bool bProcessTerrain = true;
-	while (!m_bShutdowned)
-	{
-		DWORD dwWaitResult; 
-
-		dwWaitResult = WaitForSingleObject(m_hSemaphore, INFINITE);
-
-		if (m_bShutdowned)
-			break;
-
-		switch (dwWaitResult) 
-		{ 
-			case WAIT_OBJECT_0:
-				if (bProcessTerrain)
-					ProcessTerrain();
-				else
-					ProcessArea();
-				break;
-			case WAIT_TIMEOUT:
-				TraceError("TEMP_CAreaLoaderThread::Execute: Timeout occured while time-out interval is INIFITE");
-				break;
-		}
+		std::lock_guard<std::mutex> lock(m_TerrainCompleteMutex);
+		m_pTerrainCompleteDeque.clear();
 	}
 
-	Destroy();
-	return 1;
+	{
+		std::lock_guard<std::mutex> lock(m_AreaCompleteMutex);
+		m_pAreaCompleteDeque.clear();
+	}
 }
 
-void TEMP_CAreaLoaderThread::Request(CTerrain * pTerrain)	// called in main thread
+void TEMP_CAreaLoaderThread::Request(CTerrain * pTerrain)
 {
-	m_TerrainRequestMutex.Lock();
-	m_pTerrainRequestDeque.push_back(pTerrain);
-	m_TerrainRequestMutex.Unlock();
+	if (m_bShutdowned)
+		return;
 
-	++m_iRestSemCount;
-
-	if (!ReleaseSemaphore(m_hSemaphore, m_iRestSemCount, NULL))
-		TraceError("TEMP_CAreaLoaderThread::Request: ReleaseSemaphore error");
-
-	--m_iRestSemCount;
+	// Enqueue terrain loading to the global thread pool
+	CGameThreadPool* pThreadPool = CGameThreadPool::InstancePtr();
+	if (pThreadPool)
+	{
+		pThreadPool->Enqueue([this, pTerrain]()
+		{
+			ProcessTerrain(pTerrain);
+		});
+	}
+	else
+	{
+		// Fallback to synchronous loading if thread pool not available
+		ProcessTerrain(pTerrain);
+	}
 }
 
-bool TEMP_CAreaLoaderThread::Fetch(CTerrain ** ppTerrain)	// called in main thread
+bool TEMP_CAreaLoaderThread::Fetch(CTerrain ** ppTerrain)
 {
-	m_TerrainCompleteMutex.Lock();
+	std::lock_guard<std::mutex> lock(m_TerrainCompleteMutex);
 
 	if (m_pTerrainCompleteDeque.empty())
-	{
-		m_TerrainCompleteMutex.Unlock();
 		return false;
-	}
 
 	*ppTerrain = m_pTerrainCompleteDeque.front();
 	m_pTerrainCompleteDeque.pop_front();
 
-	m_TerrainCompleteMutex.Unlock();
 	return true;
 }
 
-void TEMP_CAreaLoaderThread::Request(CArea * pArea)	// called in main thread
+void TEMP_CAreaLoaderThread::Request(CArea * pArea)
 {
-	m_AreaRequestMutex.Lock();
-	m_pAreaRequestDeque.push_back(pArea);
-	m_AreaRequestMutex.Unlock();
+	if (m_bShutdowned)
+		return;
 
-	++m_iRestSemCount;
-
-	if (!ReleaseSemaphore(m_hSemaphore, m_iRestSemCount, NULL))
-		TraceError("TEMP_CAreaLoaderThread::Request: ReleaseSemaphore error");
-
-	--m_iRestSemCount;
+	// Enqueue area loading to the global thread pool
+	CGameThreadPool* pThreadPool = CGameThreadPool::InstancePtr();
+	if (pThreadPool)
+	{
+		pThreadPool->Enqueue([this, pArea]()
+		{
+			ProcessArea(pArea);
+		});
+	}
+	else
+	{
+		// Fallback to synchronous loading if thread pool not available
+		ProcessArea(pArea);
+	}
 }
 
-bool TEMP_CAreaLoaderThread::Fetch(CArea ** ppArea)	// called in main thread
+bool TEMP_CAreaLoaderThread::Fetch(CArea ** ppArea)
 {
-	m_AreaCompleteMutex.Lock();
+	std::lock_guard<std::mutex> lock(m_AreaCompleteMutex);
 
 	if (m_pAreaCompleteDeque.empty())
-	{
-		m_AreaCompleteMutex.Unlock();
 		return false;
-	}
 
 	*ppArea = m_pAreaCompleteDeque.front();
 	m_pAreaCompleteDeque.pop_front();
 
-	m_AreaCompleteMutex.Unlock();
 	return true;
 }
 
-void TEMP_CAreaLoaderThread::ProcessArea()	// called in loader thread
+void TEMP_CAreaLoaderThread::ProcessArea(CArea * pArea)
 {
-	m_AreaRequestMutex.Lock();
-
-	if (m_pAreaRequestDeque.empty())
-	{
-		m_AreaRequestMutex.Unlock();
+	if (m_bShutdowned)
 		return;
-	}
-
-	CArea * pArea = m_pAreaRequestDeque.front();
-	m_pAreaRequestDeque.pop_front();
-
-	Tracef("TEMP_CAreaLoaderThread::ProcessArea() RequestDeque Size : %d\n", m_pAreaRequestDeque.size());
-	m_AreaRequestMutex.Unlock();
 
 	DWORD dwStartTime = ELTimer_GetMSec();
 
@@ -238,28 +138,17 @@ void TEMP_CAreaLoaderThread::ProcessArea()	// called in loader thread
 
 	Tracef("TEMP_CAreaLoaderThread::ProcessArea LoadArea : %d ms elapsed\n", ELTimer_GetMSec() - dwStartTime);
 
-	m_AreaCompleteMutex.Lock();
-	m_pAreaCompleteDeque.push_back(pArea);
-	m_AreaCompleteMutex.Unlock();
-
-	Sleep(g_iLoadingDelayTime);
+	// Add to completed queue
+	{
+		std::lock_guard<std::mutex> lock(m_AreaCompleteMutex);
+		m_pAreaCompleteDeque.push_back(pArea);
+	}
 }
 
-void TEMP_CAreaLoaderThread::ProcessTerrain()	// called in loader thread
+void TEMP_CAreaLoaderThread::ProcessTerrain(CTerrain * pTerrain)
 {
-	m_TerrainRequestMutex.Lock();
-
-	if (m_pTerrainRequestDeque.empty())
-	{
-		m_TerrainRequestMutex.Unlock();
+	if (m_bShutdowned)
 		return;
-	}
-
-	CTerrain * pTerrain = m_pTerrainRequestDeque.front();
-	m_pTerrainRequestDeque.pop_front();
-
-	Tracef("TEMP_CAreaLoaderThread::ProcessTerrain() RequestDeque Size : %d\n", m_pTerrainRequestDeque.size());
-	m_TerrainRequestMutex.Unlock();
 
 	DWORD dwStartTime = ELTimer_GetMSec();
 
@@ -271,26 +160,24 @@ void TEMP_CAreaLoaderThread::ProcessTerrain()	// called in loader thread
 	const std::string & c_rStrMapName = pTerrain->GetOwner()->GetName();
 	char filename[256];
 	sprintf(filename, "%s\\%06u\\AreaProperty.txt", c_rStrMapName.c_str(), dwID);
-	
+
 	CTokenVectorMap stTokenVectorMap;
-	
+
 	if (!LoadMultipleTextData(filename, stTokenVectorMap))
 		return;
-	
-	Sleep(g_iLoadingDelayTime);
 
 	if (stTokenVectorMap.end() == stTokenVectorMap.find("scripttype"))
 		return;
-	
+
 	if (stTokenVectorMap.end() == stTokenVectorMap.find("areaname"))
 		return;
-	
+
 	const std::string & c_rstrType = stTokenVectorMap["scripttype"][0];
 	const std::string & c_rstrAreaName = stTokenVectorMap["areaname"][0];
-	
+
 	if (c_rstrType != "AreaProperty")
 		return;
-	
+
 	char szRawHeightFieldname[64+1];
 	char szWaterMapName[64+1];
 	char szAttrMapName[64+1];
@@ -298,7 +185,7 @@ void TEMP_CAreaLoaderThread::ProcessTerrain()	// called in loader thread
 	char szShadowMapName[64+1];
 	char szMiniMapTexName[64+1];
 	char szSplatName[64+1];
-	
+
 	_snprintf(szRawHeightFieldname, sizeof(szRawHeightFieldname), "%s\\%06u\\height.raw", c_rStrMapName.c_str(), dwID);
 	_snprintf(szSplatName, sizeof(szSplatName), "%s\\%06u\\tile.raw", c_rStrMapName.c_str(), dwID);
 	_snprintf(szAttrMapName, sizeof(szAttrMapName), "%s\\%06u\\attr.atr", c_rStrMapName.c_str(), dwID);
@@ -306,35 +193,26 @@ void TEMP_CAreaLoaderThread::ProcessTerrain()	// called in loader thread
 	_snprintf(szShadowTexName, sizeof(szShadowTexName), "%s\\%06u\\shadowmap.dds", c_rStrMapName.c_str(), dwID);
 	_snprintf(szShadowMapName, sizeof(szShadowMapName), "%s\\%06u\\shadowmap.raw", c_rStrMapName.c_str(), dwID);
 	_snprintf(szMiniMapTexName,	sizeof(szMiniMapTexName), "%s\\%06u\\minimap.dds", c_rStrMapName.c_str(), dwID);
-	
+
 	pTerrain->CopySettingFromGlobalSetting();
 
 	pTerrain->LoadWaterMap(szWaterMapName);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->LoadHeightMap(szRawHeightFieldname);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->LoadAttrMap(szAttrMapName);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->RAW_LoadTileMap(szSplatName, true);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->LoadShadowTexture(szShadowTexName);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->LoadShadowMap(szShadowMapName);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->LoadMiniMapTexture(szMiniMapTexName);
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->SetName(c_rstrAreaName.c_str());
-	Sleep(g_iLoadingDelayTime);
 	pTerrain->CalculateTerrainPatch();
-	Sleep(g_iLoadingDelayTime);
 
 	pTerrain->SetReady();
 
 	Tracef("TEMP_CAreaLoaderThread::ProcessTerrain LoadTerrain : %d ms elapsed\n", ELTimer_GetMSec() - dwStartTime);
 
-	m_TerrainCompleteMutex.Lock();
-	m_pTerrainCompleteDeque.push_back(pTerrain);
-	m_TerrainCompleteMutex.Unlock();
-
-	Sleep(g_iLoadingDelayTime);
+	// Add to completed queue
+	{
+		std::lock_guard<std::mutex> lock(m_TerrainCompleteMutex);
+		m_pTerrainCompleteDeque.push_back(pTerrain);
+	}
 }

src/GameLib/AreaLoaderThread.h

@@ -2,19 +2,15 @@
 //
 //////////////////////////////////////////////////////////////////////
 
-#if !defined(AFX_AREALOADERTHREAD_H__E43FBE42_42F4_4F0E_B9DA_D7B7C5EA0753__INCLUDED_)
-#define AFX_AREALOADERTHREAD_H__E43FBE42_42F4_4F0E_B9DA_D7B7C5EA0753__INCLUDED_
-
-#if _MSC_VER > 1000
 #pragma once
-#endif // _MSC_VER > 1000
 
-#include "EterLib/Mutex.h"
+#include <deque>
+#include <mutex>
 
 class CTerrain;
 class CArea;
 
-class TEMP_CAreaLoaderThread  
+class TEMP_CAreaLoaderThread
 {
 public:
 	TEMP_CAreaLoaderThread();
@@ -24,50 +20,21 @@ public:
 	void						Shutdown();
 
 	void						Request(CTerrain * pTerrain);
-
 	bool						Fetch(CTerrain ** ppTerrian);
 
 	void						Request(CArea * pArea);
-	
 	bool						Fetch(CArea ** ppArea);
 
-protected:
-	static UINT CALLBACK		EntryPoint(void * pThis);
-	UINT						Run(void * arg);
-	
-	void *						Arg() const		{ return m_pArg; }
-	void						Arg(void * arg) { m_pArg = arg; }
-	
-	HANDLE						m_hThread;
-	
 private:
-	void *						m_pArg;
-	unsigned					m_uThreadID;
-	
-protected:
-	UINT						Setup();
-	UINT						Execute(void * pvArg);
-	void						Destroy();
-	void						ProcessTerrain();
-	void						ProcessArea();
-	
+	void						ProcessTerrain(CTerrain * pTerrain);
+	void						ProcessArea(CArea * pArea);
+
 private:
-	std::deque<CTerrain *>		m_pTerrainRequestDeque;
-	Mutex						m_TerrainRequestMutex;
-	
 	std::deque<CTerrain *>		m_pTerrainCompleteDeque;
-	Mutex						m_TerrainCompleteMutex;
-	
-	std::deque<CArea *>			m_pAreaRequestDeque;
-	Mutex						m_AreaRequestMutex;
-	
+	std::mutex					m_TerrainCompleteMutex;
+
 	std::deque<CArea *>			m_pAreaCompleteDeque;
-	Mutex						m_AreaCompleteMutex;
+	std::mutex					m_AreaCompleteMutex;
 
-	HANDLE						m_hSemaphore;
-	int							m_iRestSemCount;
 	bool						m_bShutdowned;
-
 };
-
-#endif // !defined(AFX_AREALOADERTHREAD_H__E43FBE42_42F4_4F0E_B9DA_D7B7C5EA0753__INCLUDED_)

src/GameLib/RaceManager.cpp

@@ -2,6 +2,7 @@
 #include "RaceManager.h"
 #include "RaceMotionData.h"
 #include "PackLib/PackManager.h"
+#include "EterLib/GameThreadPool.h"
 #include <future>
 #include <vector>
 #include <set>
@@ -374,25 +375,61 @@ CRaceData * CRaceManager::GetSelectedRaceDataPointer()
 
 BOOL CRaceManager::GetRaceDataPointer(DWORD dwRaceIndex, CRaceData ** ppRaceData)
 {
-	TRaceDataIterator itor = m_RaceDataMap.find(dwRaceIndex);
-
-	if (m_RaceDataMap.end() == itor)
+	// Thread-safe lookup
 	{
-		CRaceData* pRaceData = __LoadRaceData(dwRaceIndex);
+		std::lock_guard<std::mutex> lock(m_RaceDataMapMutex);
+		TRaceDataIterator itor = m_RaceDataMap.find(dwRaceIndex);
 
-		if (pRaceData)
+		if (m_RaceDataMap.end() != itor)
 		{
-			m_RaceDataMap.insert(TRaceDataMap::value_type(dwRaceIndex, pRaceData));
-			*ppRaceData = pRaceData;
+			*ppRaceData = itor->second;
 			return TRUE;
 		}
-
-		TraceError("CRaceManager::GetRaceDataPointer: cannot load data by dwRaceIndex %lu", dwRaceIndex);
-		return FALSE;
 	}
 
-	*ppRaceData = itor->second;
-	return TRUE;
+	// Check if already loading asynchronously
+	{
+		std::lock_guard<std::mutex> lock(m_LoadingRacesMutex);
+		if (m_LoadingRaces.find(dwRaceIndex) != m_LoadingRaces.end())
+		{
+			// Race is being loaded asynchronously
+			// Wait for it to complete by loading synchronously now (needed for immediate visibility)
+			Tracef("CRaceManager::GetRaceDataPointer: Race %lu is loading async, switching to sync load\n", dwRaceIndex);
+		}
+	}
+
+	// Not found - load synchronously to ensure immediate availability
+	CRaceData* pRaceData = __LoadRaceData(dwRaceIndex);
+
+	if (pRaceData)
+	{
+		std::lock_guard<std::mutex> lock(m_RaceDataMapMutex);
+		// Check again in case another thread loaded it
+		TRaceDataIterator itor = m_RaceDataMap.find(dwRaceIndex);
+		if (m_RaceDataMap.end() != itor)
+		{
+			// Already loaded by another thread, use that one
+			delete pRaceData;
+			*ppRaceData = itor->second;
+		}
+		else
+		{
+			// Insert our newly loaded data
+			m_RaceDataMap.insert(TRaceDataMap::value_type(dwRaceIndex, pRaceData));
+			*ppRaceData = pRaceData;
+		}
+
+		// Remove from loading set if present
+		{
+			std::lock_guard<std::mutex> lock2(m_LoadingRacesMutex);
+			m_LoadingRaces.erase(dwRaceIndex);
+		}
+
+		return TRUE;
+	}
+
+	*ppRaceData = NULL;
+	return FALSE;
 }
 
 void CRaceManager::SetPathName(const char * c_szPathName)
@@ -462,27 +499,19 @@ void CRaceManager::PreloadPlayerRaceMotions()
 
 	CRaceManager& rkRaceMgr = CRaceManager::Instance();
 
-	// Phase 1: Parallel Load Race Data (MSM)
-	std::vector<std::future<CRaceData*>> raceLoadFutures;
-
 	for (DWORD dwRace = 0; dwRace <= 7; ++dwRace)
 	{
 		TRaceDataIterator it = rkRaceMgr.m_RaceDataMap.find(dwRace);
-		if (it == rkRaceMgr.m_RaceDataMap.end()) {
-			raceLoadFutures.push_back(std::async(std::launch::async, [&rkRaceMgr, dwRace]() {
-				return rkRaceMgr.__LoadRaceData(dwRace);
-			}));
+		if (it == rkRaceMgr.m_RaceDataMap.end())
+		{
+			CRaceData* pRaceData = rkRaceMgr.__LoadRaceData(dwRace);
+			if (pRaceData)
+			{
+				rkRaceMgr.m_RaceDataMap.insert(TRaceDataMap::value_type(pRaceData->GetRaceIndex(), pRaceData));
+			}
 		}
 	}
 
-	for (auto& f : raceLoadFutures) {
-		CRaceData* pRaceData = f.get();
-		if (pRaceData) {
-			rkRaceMgr.m_RaceDataMap.insert(TRaceDataMap::value_type(pRaceData->GetRaceIndex(), pRaceData));
-		}
-	}
-
-	// Phase 2: Parallel Load Motions
 	std::set<CGraphicThing*> uniqueMotions;
 
 	for (DWORD dwRace = 0; dwRace <= 7; ++dwRace)
@@ -518,28 +547,117 @@ void CRaceManager::PreloadPlayerRaceMotions()
 	std::vector<CGraphicThing*> motionVec(uniqueMotions.begin(), uniqueMotions.end());
 	size_t total = motionVec.size();
 
-	if (total > 0) {
-		size_t threadCount = std::thread::hardware_concurrency();
-		if (threadCount == 0) threadCount = 4;
-		
-		size_t chunkSize = (total + threadCount - 1) / threadCount;
-		std::vector<std::future<void>> motionFutures;
+	if (total > 0)
+	{
+		CGameThreadPool* pThreadPool = CGameThreadPool::InstancePtr();
+		if (pThreadPool && pThreadPool->IsInitialized())
+		{
+			size_t workerCount = pThreadPool->GetWorkerCount();
+			size_t chunkSize = (total + workerCount - 1) / workerCount;
 
-		for (size_t i = 0; i < threadCount; ++i) {
-			size_t start = i * chunkSize;
-			size_t end = std::min(start + chunkSize, total);
-			
-			if (start < end) {
-				motionFutures.push_back(std::async(std::launch::async, [start, end, &motionVec]() {
-					for (size_t k = start; k < end; ++k) {
-						motionVec[k]->AddReference();
-					}
-				}));
+			std::vector<std::future<void>> futures;
+			futures.reserve(workerCount);
+
+			for (size_t i = 0; i < workerCount; ++i)
+			{
+				size_t start = i * chunkSize;
+				size_t end = std::min(start + chunkSize, total);
+
+				if (start < end)
+				{
+					// Copy values instead of capturing by reference
+					futures.push_back(pThreadPool->Enqueue([start, end, motionVec]() {
+						for (size_t k = start; k < end; ++k)
+						{
+							motionVec[k]->AddReference();
+						}
+					}));
+				}
+			}
+
+			// Wait for all tasks to complete
+			for (auto& f : futures)
+			{
+				f.wait();
+			}
+		}
+		else
+		{
+			// Fallback to sequential if thread pool not available
+			for (auto* pMotion : motionVec)
+			{
+				pMotion->AddReference();
 			}
 		}
-		
-		for (auto& f : motionFutures) f.get();
 	}
 
 	s_bPreloaded = true;
 }
+
+void CRaceManager::RequestAsyncRaceLoad(DWORD dwRaceIndex)
+{
+	// Mark as loading
+	{
+		std::lock_guard<std::mutex> lock(m_LoadingRacesMutex);
+		if (m_LoadingRaces.find(dwRaceIndex) != m_LoadingRaces.end())
+		{
+			// Already loading
+			return;
+		}
+		m_LoadingRaces.insert(dwRaceIndex);
+	}
+
+	// Enqueue async load to game thread pool
+	CGameThreadPool* pThreadPool = CGameThreadPool::InstancePtr();
+	if (pThreadPool)
+	{
+		pThreadPool->Enqueue([this, dwRaceIndex]()
+		{
+			CRaceData* pRaceData = __LoadRaceData(dwRaceIndex);
+
+			if (pRaceData)
+			{
+				// Thread-safe insertion
+				{
+					std::lock_guard<std::mutex> lock(m_RaceDataMapMutex);
+					m_RaceDataMap.insert(TRaceDataMap::value_type(dwRaceIndex, pRaceData));
+				}
+
+				Tracef("CRaceManager::RequestAsyncRaceLoad: Successfully loaded race %lu asynchronously\n", dwRaceIndex);
+			}
+			else
+			{
+				TraceError("CRaceManager::RequestAsyncRaceLoad: Failed to load race %lu", dwRaceIndex);
+			}
+
+			// Remove from loading set
+			{
+				std::lock_guard<std::mutex> lock(m_LoadingRacesMutex);
+				m_LoadingRaces.erase(dwRaceIndex);
+			}
+		});
+	}
+	else
+	{
+		// Fallback to synchronous loading if thread pool not available
+		CRaceData* pRaceData = __LoadRaceData(dwRaceIndex);
+
+		if (pRaceData)
+		{
+			std::lock_guard<std::mutex> lock(m_RaceDataMapMutex);
+			m_RaceDataMap.insert(TRaceDataMap::value_type(dwRaceIndex, pRaceData));
+		}
+
+		// Remove from loading set
+		{
+			std::lock_guard<std::mutex> lock(m_LoadingRacesMutex);
+			m_LoadingRaces.erase(dwRaceIndex);
+		}
+	}
+}
+
+bool CRaceManager::IsRaceLoading(DWORD dwRaceIndex) const
+{
+	std::lock_guard<std::mutex> lock(m_LoadingRacesMutex);
+	return m_LoadingRaces.find(dwRaceIndex) != m_LoadingRaces.end();
+}

src/GameLib/RaceManager.h

@@ -1,6 +1,8 @@
 #pragma once
 
 #include "RaceData.h"
+#include <mutex>
+#include <set>
 
 class CRaceManager : public CSingleton<CRaceManager>
 {
@@ -29,6 +31,10 @@ class CRaceManager : public CSingleton<CRaceManager>
 
 		BOOL GetRaceDataPointer(DWORD dwRaceIndex, CRaceData ** ppRaceData);
 
+		// Async race loading
+		void RequestAsyncRaceLoad(DWORD dwRaceIndex);
+		bool IsRaceLoading(DWORD dwRaceIndex) const;
+
 		// Race motion preloading
 		static void PreloadPlayerRaceMotions();
 		static bool IsPreloaded() { return s_bPreloaded; }
@@ -42,6 +48,10 @@ class CRaceManager : public CSingleton<CRaceManager>
 
 	protected:
 		TRaceDataMap					m_RaceDataMap;
+		mutable std::mutex				m_RaceDataMapMutex;
+
+		std::set<DWORD>					m_LoadingRaces;
+		mutable std::mutex				m_LoadingRacesMutex;
 
 		std::map<std::string, std::string> m_kMap_stRaceName_stSrcName;
 		std::map<DWORD, std::string>	m_kMap_dwRaceKey_stRaceName;

src/UserInterface/PythonApplication.cpp

@@ -939,6 +939,13 @@ unsigned __GetWindowMode(bool windowed)
 
 bool CPythonApplication::Create(PyObject * poSelf, const char * c_szName, int width, int height, int Windowed)
 {
+	// Initialize Game Thread Pool first - required by other systems
+	CGameThreadPool* pThreadPool = CGameThreadPool::InstancePtr();
+	if (pThreadPool)
+	{
+		pThreadPool->Initialize();
+	}
+
 	NANOBEGIN
 		Windowed = CPythonSystem::Instance().IsWindowed() ? 1 : 0;
 
@@ -1250,6 +1257,9 @@ void CPythonApplication::Destroy()
 	m_kEftMgr.Destroy();
 	m_LightManager.Destroy();
 
+	// Game Thread Pool
+	CGameThreadPool::Instance().Destroy();
+
 	// DEFAULT_FONT
 	DefaultFont_Cleanup();
 	// END_OF_DEFAULT_FONT

src/UserInterface/PythonApplication.h

@@ -6,6 +6,7 @@
 #include "eterLib/GrpDevice.h"
 #include "eterLib/NetDevice.h"
 #include "eterLib/GrpLightManager.h"
+#include "eterLib/GameThreadPool.h"
 #include "EffectLib/EffectManager.h"
 #include "gamelib/RaceManager.h"
 #include "gamelib/ItemManager.h"

src/UserInterface/UserInterface.cpp

@@ -10,6 +10,7 @@
 #endif
 
 #include "eterLib/Util.h"
+#include "EterLib/GameThreadPool.h"
 #include "EterBase/lzo.h"
 
 #include "PackLib/PackManager.h"
@@ -180,14 +181,16 @@ bool PackInitialize(const char * c_pszFolder)
 	const size_t packsPerThread = (packFiles.size() + numThreads - 1) / numThreads;
 
 	std::vector<std::thread> threads;
+	threads.reserve(numThreads);  // Pre-allocate to prevent reallocation
 	std::atomic<size_t> failedCount(0);
 
+	// Create all threads first (prevents vector reallocation during emplace_back)
 	for (size_t t = 0; t < numThreads; ++t)
 	{
-		threads.emplace_back([&, t]() {
-			size_t start = t * packsPerThread;
-			size_t end = std::min(start + packsPerThread, packFiles.size());
+		size_t start = t * packsPerThread;
+		size_t end = std::min(start + packsPerThread, packFiles.size());
 
+		threads.emplace_back([&failedCount, &packFiles, c_pszFolder, start, end]() {
 			for (size_t i = start; i < end; ++i)
 			{
 				std::string packPath = std::format("{}/{}.pck", c_pszFolder, packFiles[i]);
@@ -296,6 +299,9 @@ static bool Main(HINSTANCE hInstance, LPSTR lpCmdLine)
 		return false;
 	}
 
+	// Create game thread pool singleton before CPythonApplication
+	static CGameThreadPool gameThreadPool;
+
 	auto app = new CPythonApplication;
 	app->Initialize (hInstance);
 	CPythonLauncher pyLauncher;