Merge pull request #65 from savisxss/main

A bit late, but Happy New Year!

src/EterLib/BufferPool.cpp

@@ -0,0 +1,103 @@
+#include "StdAfx.h"
+#include "BufferPool.h"
+#include <algorithm>
+
+CBufferPool::CBufferPool()
+	: m_totalAllocated(0)
+{
+}
+
+CBufferPool::~CBufferPool()
+{
+	Clear();
+}
+
+std::vector<uint8_t> CBufferPool::Acquire(size_t minSize)
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+
+	size_t bestIndex = SIZE_MAX;
+	size_t bestCapacity = SIZE_MAX;
+
+	for (size_t i = 0; i < m_pool.size(); ++i)
+	{
+		if (m_pool[i].capacity >= minSize && m_pool[i].capacity < bestCapacity)
+		{
+			bestIndex = i;
+			bestCapacity = m_pool[i].capacity;
+
+			if (bestCapacity == minSize)
+				break;
+		}
+	}
+
+	if (bestIndex != SIZE_MAX)
+	{
+		std::vector<uint8_t> result = std::move(m_pool[bestIndex].buffer);
+		m_pool.erase(m_pool.begin() + bestIndex);
+		result.clear();
+		return result;
+	}
+
+	std::vector<uint8_t> newBuffer;
+	newBuffer.reserve(minSize);
+	m_totalAllocated++;
+	return newBuffer;
+}
+
+void CBufferPool::Release(std::vector<uint8_t>&& buffer)
+{
+	size_t capacity = buffer.capacity();
+
+	if (capacity == 0 || capacity > MAX_BUFFER_SIZE)
+	{
+		return;
+	}
+
+	std::lock_guard<std::mutex> lock(m_mutex);
+
+	if (m_pool.size() >= MAX_POOL_SIZE)
+	{
+		auto smallest = std::min_element(m_pool.begin(), m_pool.end(),
+			[](const TPooledBuffer& a, const TPooledBuffer& b) {
+				return a.capacity < b.capacity;
+			});
+
+		if (smallest != m_pool.end() && smallest->capacity < capacity)
+		{
+			*smallest = TPooledBuffer(std::move(buffer));
+		}
+		return;
+	}
+
+	m_pool.emplace_back(std::move(buffer));
+}
+
+size_t CBufferPool::GetPoolSize() const
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+	return m_pool.size();
+}
+
+size_t CBufferPool::GetTotalAllocated() const
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+	return m_totalAllocated;
+}
+
+size_t CBufferPool::GetTotalMemoryPooled() const
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+	size_t total = 0;
+	for (const auto& buf : m_pool)
+	{
+		total += buf.capacity;
+	}
+	return total;
+}
+
+void CBufferPool::Clear()
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+	m_pool.clear();
+}

src/EterLib/BufferPool.h

@@ -0,0 +1,50 @@
+#ifndef __INC_ETERLIB_BUFFERPOOL_H__
+#define __INC_ETERLIB_BUFFERPOOL_H__
+
+#include <vector>
+#include <mutex>
+#include <cstdint>
+
+// Buffer pool for file I/O operations
+class CBufferPool
+{
+public:
+	CBufferPool();
+	~CBufferPool();
+
+	// Get buffer with minimum size
+	std::vector<uint8_t> Acquire(size_t minSize);
+
+	// Return buffer to pool
+	void Release(std::vector<uint8_t>&& buffer);
+
+	// Get statistics
+	size_t GetPoolSize() const;
+	size_t GetTotalAllocated() const;
+	size_t GetTotalMemoryPooled() const; // Total bytes held in pool
+
+	// Clear pool
+	void Clear();
+
+private:
+	struct TPooledBuffer
+	{
+		std::vector<uint8_t> buffer;
+		size_t capacity;
+
+		TPooledBuffer(std::vector<uint8_t>&& buf)
+			: buffer(std::move(buf))
+			, capacity(buffer.capacity())
+		{
+		}
+	};
+
+	std::vector<TPooledBuffer> m_pool;
+	mutable std::mutex m_mutex;
+	size_t m_totalAllocated;
+
+	static const size_t MAX_POOL_SIZE = 64;
+	static const size_t MAX_BUFFER_SIZE = 64 * 1024 * 1024;
+};
+
+#endif // __INC_ETERLIB_BUFFERPOOL_H__

src/EterLib/DecodedImageData.h

@@ -0,0 +1,59 @@
+#ifndef __INC_ETERLIB_DECODEDIMAGEDATA_H__
+#define __INC_ETERLIB_DECODEDIMAGEDATA_H__
+
+#include <vector>
+#include <cstdint>
+#include <d3d9.h>
+
+// Decoded image data for GPU upload
+struct TDecodedImageData
+{
+	enum EFormat
+	{
+		FORMAT_UNKNOWN = 0,
+		FORMAT_RGBA8,
+		FORMAT_RGB8,
+		FORMAT_DDS,
+	};
+
+	std::vector<uint8_t> pixels;
+	int width;
+	int height;
+	EFormat format;
+	D3DFORMAT d3dFormat;
+	bool isDDS;
+	int mipLevels;
+
+	TDecodedImageData()
+		: width(0)
+		, height(0)
+		, format(FORMAT_UNKNOWN)
+		, d3dFormat(D3DFMT_UNKNOWN)
+		, isDDS(false)
+		, mipLevels(1)
+	{
+	}
+
+	void Clear()
+	{
+		pixels.clear();
+		width = 0;
+		height = 0;
+		format = FORMAT_UNKNOWN;
+		d3dFormat = D3DFMT_UNKNOWN;
+		isDDS = false;
+		mipLevels = 1;
+	}
+
+	bool IsValid() const
+	{
+		return width > 0 && height > 0 && !pixels.empty();
+	}
+
+	size_t GetDataSize() const
+	{
+		return pixels.size();
+	}
+};
+
+#endif // __INC_ETERLIB_DECODEDIMAGEDATA_H__

src/EterLib/FileLoaderThreadPool.cpp

@@ -0,0 +1,270 @@
+#include "StdAfx.h"
+#include "FileLoaderThreadPool.h"
+#include "BufferPool.h"
+#include "ImageDecoder.h"
+#include "PackLib/PackManager.h"
+#include <algorithm>
+
+static const bool USE_STAGED_TEXTURE_LOADING = true;
+
+CFileLoaderThreadPool::CFileLoaderThreadPool()
+	: m_pCompletedQueue(nullptr)
+	, m_bShutdown(false)
+	, m_nextRequestID(0)
+	, m_activeTasks(0)
+	, m_threadCount(0)
+{
+}
+
+CFileLoaderThreadPool::~CFileLoaderThreadPool()
+{
+	Shutdown();
+}
+
+bool CFileLoaderThreadPool::Initialize(unsigned int threadCount)
+{
+	if (!m_workers.empty())
+	{
+		TraceError("CFileLoaderThreadPool::Initialize: Already initialized");
+		return false;
+	}
+
+	if (threadCount == 0)
+	{
+		threadCount = std::thread::hardware_concurrency();
+		if (threadCount == 0)
+			threadCount = 4;
+		else
+			threadCount = std::max(4u, threadCount / 2);
+	}
+
+	threadCount = std::max(4u, std::min(16u, threadCount));
+	m_threadCount = threadCount;
+
+	Tracenf("CFileLoaderThreadPool: Initializing with %u worker threads", threadCount);
+
+	m_pCompletedQueue = new SPSCQueue<TLoadResult>(COMPLETED_QUEUE_SIZE);
+
+	m_workers.reserve(threadCount);
+	for (unsigned int i = 0; i < threadCount; ++i)
+	{
+		TWorkerThread worker;
+		worker.pRequestQueue = new SPSCQueue<TLoadRequest>(REQUEST_QUEUE_SIZE);
+		worker.bBusy.store(false, std::memory_order_relaxed);
+
+		try
+		{
+			worker.thread = std::thread(&CFileLoaderThreadPool::WorkerThreadFunction, this, i);
+		}
+		catch (const std::exception& e)
+		{
+			TraceError("CFileLoaderThreadPool::Initialize: Failed to create thread %u: %s", i, e.what());
+			delete worker.pRequestQueue;
+			worker.pRequestQueue = nullptr;
+			Shutdown();
+			return false;
+		}
+
+		m_workers.push_back(std::move(worker));
+	}
+
+	return true;
+}
+
+void CFileLoaderThreadPool::Shutdown()
+{
+	if (m_workers.empty())
+		return;
+
+	// Signal shutdown
+	m_bShutdown.store(true, std::memory_order_release);
+
+	// Wait for all workers to finish
+	for (auto& worker : m_workers)
+	{
+		if (worker.thread.joinable())
+			worker.thread.join();
+
+		// Cleanup request queue
+		if (worker.pRequestQueue)
+		{
+			delete worker.pRequestQueue;
+			worker.pRequestQueue = nullptr;
+		}
+	}
+
+	m_workers.clear();
+
+	// Cleanup completed queue
+	if (m_pCompletedQueue)
+	{
+		delete m_pCompletedQueue;
+		m_pCompletedQueue = nullptr;
+	}
+
+	m_threadCount = 0;
+}
+
+bool CFileLoaderThreadPool::Request(const std::string& fileName)
+{
+	if (m_workers.empty())
+	{
+		TraceError("CFileLoaderThreadPool::Request: Thread pool not initialized");
+		return false;
+	}
+
+	TLoadRequest request;
+	request.stFileName = fileName;
+	request.requestID = m_nextRequestID.fetch_add(1, std::memory_order_relaxed);
+
+	request.decodeImage = false;
+	if (USE_STAGED_TEXTURE_LOADING)
+	{
+		size_t dotPos = fileName.find_last_of('.');
+		if (dotPos != std::string::npos && dotPos + 1 < fileName.size())
+		{
+			const char* ext = fileName.c_str() + dotPos;
+			size_t extLen = fileName.size() - dotPos;
+
+			if ((extLen == 4 && (_stricmp(ext, ".dds") == 0 || _stricmp(ext, ".png") == 0 ||
+				_stricmp(ext, ".jpg") == 0 || _stricmp(ext, ".tga") == 0 || _stricmp(ext, ".bmp") == 0)) ||
+				(extLen == 5 && _stricmp(ext, ".jpeg") == 0))
+			{
+				request.decodeImage = true;
+			}
+		}
+	}
+
+	unsigned int targetWorker = SelectLeastBusyWorker();
+
+	if (!m_workers[targetWorker].pRequestQueue->Push(request))
+	{
+		for (unsigned int i = 0; i < m_threadCount; ++i)
+		{
+			unsigned int workerIdx = (targetWorker + i) % m_threadCount;
+			if (m_workers[workerIdx].pRequestQueue->Push(request))
+			{
+				m_activeTasks.fetch_add(1, std::memory_order_relaxed);
+				return true;
+			}
+		}
+
+		TraceError("CFileLoaderThreadPool::Request: All worker queues full for file: %s", fileName.c_str());
+		return false;
+	}
+
+	m_activeTasks.fetch_add(1, std::memory_order_relaxed);
+	return true;
+}
+
+bool CFileLoaderThreadPool::Fetch(TLoadResult& result)
+{
+	if (!m_pCompletedQueue)
+		return false;
+
+	if (m_pCompletedQueue->Pop(result))
+	{
+		m_activeTasks.fetch_sub(1, std::memory_order_relaxed);
+		return true;
+	}
+	return false;
+}
+
+size_t CFileLoaderThreadPool::GetPendingCount() const
+{
+	size_t total = 0;
+	for (const auto& worker : m_workers)
+	{
+		if (worker.pRequestQueue)
+			total += worker.pRequestQueue->Size();
+	}
+	return total;
+}
+
+bool CFileLoaderThreadPool::IsIdle() const
+{
+	return m_activeTasks.load(std::memory_order_acquire) == 0;
+}
+
+unsigned int CFileLoaderThreadPool::SelectLeastBusyWorker() const
+{
+	unsigned int leastBusyIdx = 0;
+	size_t minSize = m_workers[0].pRequestQueue->Size();
+
+	for (unsigned int i = 1; i < m_threadCount; ++i)
+	{
+		size_t queueSize = m_workers[i].pRequestQueue->Size();
+		if (queueSize < minSize)
+		{
+			minSize = queueSize;
+			leastBusyIdx = i;
+		}
+	}
+
+	return leastBusyIdx;
+}
+
+void CFileLoaderThreadPool::WorkerThreadFunction(unsigned int workerIndex)
+{
+	TWorkerThread& worker = m_workers[workerIndex];
+	SPSCQueue<TLoadRequest>* pRequestQueue = worker.pRequestQueue;
+
+	CBufferPool* pBufferPool = CPackManager::instance().GetBufferPool();
+
+	Tracenf("CFileLoaderThreadPool: Worker thread %u started", workerIndex);
+
+	int idleCount = 0;
+
+	while (!m_bShutdown.load(std::memory_order_acquire))
+	{
+		TLoadRequest request;
+
+		if (pRequestQueue->Pop(request))
+		{
+			idleCount = 0;
+			worker.bBusy.store(true, std::memory_order_release);
+
+			TLoadResult result;
+			result.stFileName = request.stFileName;
+			result.requestID = request.requestID;
+			result.File.clear();
+			result.hasDecodedImage = false;
+
+			CPackManager::instance().GetFileWithPool(request.stFileName, result.File, pBufferPool);
+
+			if (request.decodeImage && !result.File.empty())
+			{
+				if (CImageDecoder::DecodeImage(result.File.data(), result.File.size(), result.decodedImage))
+				{
+					result.hasDecodedImage = true;
+					result.File.clear();
+				}
+			}
+
+			while (!m_pCompletedQueue->Push(result))
+			{
+				std::this_thread::yield();
+
+				if (m_bShutdown.load(std::memory_order_acquire))
+					break;
+			}
+
+			worker.bBusy.store(false, std::memory_order_release);
+		}
+		else
+		{
+			idleCount++;
+			if (idleCount > 1000)
+			{
+				Sleep(1);
+				idleCount = 0;
+			}
+			else if (idleCount > 10)
+			{
+				std::this_thread::yield();
+			}
+		}
+	}
+
+	Tracenf("CFileLoaderThreadPool: Worker thread %u stopped", workerIndex);
+}

src/EterLib/FileLoaderThreadPool.h

@@ -0,0 +1,90 @@
+#ifndef __INC_ETERLIB_FILELOADERTHREADPOOL_H__
+#define __INC_ETERLIB_FILELOADERTHREADPOOL_H__
+
+#include <vector>
+#include <thread>
+#include <atomic>
+#include "SPSCQueue.h"
+#include "PackLib/PackManager.h"
+#include "DecodedImageData.h"
+
+class CFileLoaderThreadPool
+{
+public:
+	struct TLoadRequest
+	{
+		std::string stFileName;
+		uint32_t requestID;
+		bool decodeImage;
+	};
+
+	struct TLoadResult
+	{
+		std::string stFileName;
+		TPackFile File;
+		uint32_t requestID;
+		TDecodedImageData decodedImage;
+		bool hasDecodedImage;
+	};
+
+public:
+	CFileLoaderThreadPool();
+	~CFileLoaderThreadPool();
+
+	bool Initialize(unsigned int threadCount = 0);
+	void Shutdown();
+	bool Request(const std::string& fileName);
+	bool Fetch(TLoadResult& result);
+	size_t GetPendingCount() const;
+	bool IsIdle() const;
+
+private:
+	struct TWorkerThread
+	{
+		std::thread thread;
+		SPSCQueue<TLoadRequest>* pRequestQueue;
+		std::atomic<bool> bBusy;
+
+		TWorkerThread() : pRequestQueue(nullptr), bBusy(false) {}
+
+		TWorkerThread(TWorkerThread&& other) noexcept
+			: thread(std::move(other.thread))
+			, pRequestQueue(other.pRequestQueue)
+			, bBusy(other.bBusy.load())
+		{
+			other.pRequestQueue = nullptr;
+		}
+
+		TWorkerThread& operator=(TWorkerThread&& other) noexcept
+		{
+			if (this != &other)
+			{
+				thread = std::move(other.thread);
+				pRequestQueue = other.pRequestQueue;
+				bBusy.store(other.bBusy.load());
+				other.pRequestQueue = nullptr;
+			}
+			return *this;
+		}
+
+		TWorkerThread(const TWorkerThread&) = delete;
+		TWorkerThread& operator=(const TWorkerThread&) = delete;
+	};
+
+	void WorkerThreadFunction(unsigned int workerIndex);
+	unsigned int SelectLeastBusyWorker() const;
+
+private:
+	std::vector<TWorkerThread> m_workers;
+	SPSCQueue<TLoadResult>* m_pCompletedQueue;
+
+	std::atomic<bool> m_bShutdown;
+	std::atomic<uint32_t> m_nextRequestID;
+	std::atomic<int> m_activeTasks; // Fast IsIdle check
+	unsigned int m_threadCount;
+
+	static const size_t REQUEST_QUEUE_SIZE = 16384;  // Doubled from 8192
+	static const size_t COMPLETED_QUEUE_SIZE = 32768; // Doubled from 16384
+};
+
+#endif // __INC_ETERLIB_FILELOADERTHREADPOOL_H__

src/EterLib/GrpImage.cpp

@@ -1,5 +1,6 @@
 #include "StdAfx.h"
 #include "GrpImage.h"
+#include "DecodedImageData.h"
 
 CGraphicImage::CGraphicImage(const char * c_szFileName, DWORD dwFilter) : 
 CResource(c_szFileName),
@@ -79,6 +80,23 @@ bool CGraphicImage::OnLoad(int iSize, const void * c_pvBuf)
 	return true;
 }
 
+bool CGraphicImage::OnLoadFromDecodedData(const TDecodedImageData& decodedImage)
+{
+	if (!decodedImage.IsValid())
+		return false;
+
+	m_imageTexture.SetFileName(CResource::GetFileName());
+
+	if (!m_imageTexture.CreateFromDecodedData(decodedImage, D3DFMT_UNKNOWN, m_dwFilter))
+		return false;
+
+	m_rect.left = 0;
+	m_rect.top = 0;
+	m_rect.right = m_imageTexture.GetWidth();
+	m_rect.bottom = m_imageTexture.GetHeight();
+	return true;
+}
+
 void CGraphicImage::OnClear()
 {
 //	Tracef("Image Destroy : %s\n", m_pszFileName);

src/EterLib/GrpImage.h

@@ -5,6 +5,8 @@
 #include "Resource.h"
 #include "GrpImageTexture.h"
 
+struct TDecodedImageData;
+
 class CGraphicImage : public CResource
 {
 	public:
@@ -28,6 +30,8 @@ class CGraphicImage : public CResource
 		const CGraphicTexture & GetTextureReference() const;
 		CGraphicTexture * GetTexturePointer();
 
+		bool OnLoadFromDecodedData(const TDecodedImageData& decodedImage);
+
 	protected:
 		bool OnLoad(int iSize, const void * c_pvBuf);
 		

src/EterLib/GrpImageTexture.cpp

@@ -2,9 +2,15 @@
 #include "PackLib/PackManager.h"
 #include "GrpImageTexture.h"
 #include "EterImageLib/DDSTextureLoader9.h"
+#include "DecodedImageData.h"
 
 #include <stb_image.h>
 
+#if defined(_M_IX86) || defined(_M_X64)
+#include <emmintrin.h> // SSE2
+#include <tmmintrin.h> // SSSE3 (for _mm_shuffle_epi8)
+#endif
+
 bool CGraphicImageTexture::Lock(int* pRetPitch, void** ppRetPixels, int level)
 {
 	D3DLOCKED_RECT lockedRect;
@@ -110,17 +116,41 @@ bool CGraphicImageTexture::CreateFromSTB(UINT bufSize, const void* c_pvBuf)
 	unsigned char* data = stbi_load_from_memory((stbi_uc*)c_pvBuf, bufSize, &width, &height, &channels, 4); // force RGBA
 	if (data) {
 		LPDIRECT3DTEXTURE9 texture;
-		if (SUCCEEDED(ms_lpd3dDevice->CreateTexture(width, height, 1, 0, channels == 4 ? D3DFMT_A8R8G8B8 : D3DFMT_X8R8G8B8, D3DPOOL_DEFAULT, &texture, nullptr))) {
+		if (SUCCEEDED(ms_lpd3dDevice->CreateTexture(width, height, 1, 0, channels == 4 ? D3DFMT_A8R8G8B8 : D3DFMT_X8R8G8B8, D3DPOOL_MANAGED, &texture, nullptr))) {
 			D3DLOCKED_RECT rect;
 			if (SUCCEEDED(texture->LockRect(0, &rect, nullptr, 0))) {
 				uint8_t* dstData = (uint8_t*)rect.pBits;
 				uint8_t* srcData = (uint8_t*)data;
-				for (size_t i = 0; i < width * height; ++i, dstData += 4, srcData += 4) {
+				size_t pixelCount = width * height;
-					dstData[0] = srcData[2];
+
-					dstData[1] = srcData[1];
+				#if defined(_M_IX86) || defined(_M_X64)
-					dstData[2] = srcData[0];
+				{
-					dstData[3] = srcData[3];
+					size_t simdPixels = pixelCount & ~3;
+					__m128i shuffle_mask = _mm_setr_epi8(2, 1, 0, 3, 6, 5, 4, 7, 10, 9, 8, 11, 14, 13, 12, 15);
+
+					for (size_t i = 0; i < simdPixels; i += 4) {
+						__m128i pixels = _mm_loadu_si128((__m128i*)(srcData + i * 4));
+						pixels = _mm_shuffle_epi8(pixels, shuffle_mask);
+						_mm_storeu_si128((__m128i*)(dstData + i * 4), pixels);
+					}
+
+					for (size_t i = simdPixels; i < pixelCount; ++i) {
+						size_t idx = i * 4;
+						dstData[idx + 0] = srcData[idx + 2];
+						dstData[idx + 1] = srcData[idx + 1];
+						dstData[idx + 2] = srcData[idx + 0];
+						dstData[idx + 3] = srcData[idx + 3];
+					}
 				}
+				#else
+				for (size_t i = 0; i < pixelCount; ++i) {
+					size_t idx = i * 4;
+					dstData[idx + 0] = srcData[idx + 2];
+					dstData[idx + 1] = srcData[idx + 1];
+					dstData[idx + 2] = srcData[idx + 0];
+					dstData[idx + 3] = srcData[idx + 3];
+				}
+				#endif
 
 				texture->UnlockRect(0);
 				m_width = width;
@@ -228,6 +258,98 @@ bool CGraphicImageTexture::CreateFromDiskFile(const char * c_szFileName, D3DFORM
 	return CreateDeviceObjects();
 }
 
+bool CGraphicImageTexture::CreateFromDecodedData(const TDecodedImageData& decodedImage, D3DFORMAT d3dFmt, DWORD dwFilter)
+{
+	assert(ms_lpd3dDevice != NULL);
+	assert(m_lpd3dTexture == NULL);
+
+	if (!decodedImage.IsValid())
+		return false;
+
+	m_bEmpty = true;
+
+	if (decodedImage.isDDS)
+	{
+		// DDS format - use DirectX loader
+		if (!CreateFromDDSTexture(decodedImage.pixels.size(), decodedImage.pixels.data()))
+			return false;
+	}
+	else if (decodedImage.format == TDecodedImageData::FORMAT_RGBA8)
+	{
+		LPDIRECT3DTEXTURE9 texture;
+		D3DFORMAT format = D3DFMT_A8R8G8B8;
+
+		if (FAILED(ms_lpd3dDevice->CreateTexture(
+			decodedImage.width,
+			decodedImage.height,
+			1,
+			0,
+			format,
+			D3DPOOL_MANAGED,
+			&texture,
+			nullptr)))
+		{
+			return false;
+		}
+
+		D3DLOCKED_RECT rect;
+		if (SUCCEEDED(texture->LockRect(0, &rect, nullptr, 0)))
+		{
+			uint8_t* dstData = (uint8_t*)rect.pBits;
+			const uint8_t* srcData = decodedImage.pixels.data();
+			size_t pixelCount = decodedImage.width * decodedImage.height;
+
+			#if defined(_M_IX86) || defined(_M_X64)
+			{
+				size_t simdPixels = pixelCount & ~3;
+				__m128i shuffle_mask = _mm_setr_epi8(2, 1, 0, 3, 6, 5, 4, 7, 10, 9, 8, 11, 14, 13, 12, 15);
+
+				for (size_t i = 0; i < simdPixels; i += 4) {
+					__m128i pixels = _mm_loadu_si128((__m128i*)(srcData + i * 4));
+					pixels = _mm_shuffle_epi8(pixels, shuffle_mask);
+					_mm_storeu_si128((__m128i*)(dstData + i * 4), pixels);
+				}
+
+				for (size_t i = simdPixels; i < pixelCount; ++i) {
+					size_t idx = i * 4;
+					dstData[idx + 0] = srcData[idx + 2];
+					dstData[idx + 1] = srcData[idx + 1];
+					dstData[idx + 2] = srcData[idx + 0];
+					dstData[idx + 3] = srcData[idx + 3];
+				}
+			}
+			#else
+			for (size_t i = 0; i < pixelCount; ++i) {
+				size_t idx = i * 4;
+				dstData[idx + 0] = srcData[idx + 2];
+				dstData[idx + 1] = srcData[idx + 1];
+				dstData[idx + 2] = srcData[idx + 0];
+				dstData[idx + 3] = srcData[idx + 3];
+			}
+			#endif
+
+			texture->UnlockRect(0);
+
+			m_width = decodedImage.width;
+			m_height = decodedImage.height;
+			m_lpd3dTexture = texture;
+			m_bEmpty = false;
+		}
+		else
+		{
+			texture->Release();
+			return false;
+		}
+	}
+	else
+	{
+		TraceError("CreateFromDecodedData: Unsupported decoded image format");
+		return false;
+	}
+
+	return !m_bEmpty;
+}
+
 CGraphicImageTexture::CGraphicImageTexture()
 {
 	Initialize();

src/EterLib/GrpImageTexture.h

@@ -2,6 +2,8 @@
 
 #include "GrpTexture.h"
 
+struct TDecodedImageData;
+
 class CGraphicImageTexture : public CGraphicTexture
 {
 	public:
@@ -18,6 +20,7 @@ class CGraphicImageTexture : public CGraphicTexture
 		bool		CreateFromMemoryFile(UINT bufSize, const void* c_pvBuf, D3DFORMAT d3dFmt, DWORD dwFilter = D3DX_FILTER_LINEAR);
 		bool		CreateFromDDSTexture(UINT bufSize, const void* c_pvBuf);
 		bool		CreateFromSTB(UINT bufSize, const void* c_pvBuf);
+		bool		CreateFromDecodedData(const TDecodedImageData& decodedImage, D3DFORMAT d3dFmt, DWORD dwFilter);
 
 		void		SetFileName(const char * c_szFileName);
 		

src/EterLib/ImageDecoder.cpp

@@ -0,0 +1,92 @@
+#include "StdAfx.h"
+#include "ImageDecoder.h"
+#include "EterImageLib/DDSTextureLoader9.h"
+#include <stb_image.h>
+
+bool CImageDecoder::DecodeImage(const void* pData, size_t dataSize, TDecodedImageData& outImage)
+{
+	if (!pData || dataSize == 0)
+		return false;
+
+	outImage.Clear();
+
+	if (DecodeDDS(pData, dataSize, outImage))
+		return true;
+
+	if (DecodeSTB(pData, dataSize, outImage))
+		return true;
+
+	return false;
+}
+
+bool CImageDecoder::DecodeDDS(const void* pData, size_t dataSize, TDecodedImageData& outImage)
+{
+	if (dataSize < 4)
+		return false;
+
+	const uint32_t DDS_MAGIC = 0x20534444;
+	uint32_t magic = *(const uint32_t*)pData;
+
+	if (magic != DDS_MAGIC)
+		return false;
+
+	if (dataSize < 128)
+		return false;
+
+	struct DDSHeader
+	{
+		uint32_t magic;
+		uint32_t size;
+		uint32_t flags;
+		uint32_t height;
+		uint32_t width;
+		uint32_t pitchOrLinearSize;
+		uint32_t depth;
+		uint32_t mipMapCount;
+		uint32_t reserved1[11];
+	};
+
+	const DDSHeader* header = (const DDSHeader*)pData;
+
+	outImage.width = header->width;
+	outImage.height = header->height;
+	outImage.mipLevels = (header->mipMapCount > 0) ? header->mipMapCount : 1;
+	outImage.isDDS = true;
+	outImage.format = TDecodedImageData::FORMAT_DDS;
+
+	outImage.pixels.resize(dataSize);
+	memcpy(outImage.pixels.data(), pData, dataSize);
+
+	return true;
+}
+
+bool CImageDecoder::DecodeSTB(const void* pData, size_t dataSize, TDecodedImageData& outImage)
+{
+	int width, height, channels;
+
+	unsigned char* imageData = stbi_load_from_memory(
+		(const stbi_uc*)pData,
+		(int)dataSize,
+		&width,
+		&height,
+		&channels,
+		4
+	);
+
+	if (!imageData)
+		return false;
+
+	outImage.width = width;
+	outImage.height = height;
+	outImage.format = TDecodedImageData::FORMAT_RGBA8;
+	outImage.isDDS = false;
+	outImage.mipLevels = 1;
+
+	size_t pixelDataSize = width * height * 4;
+	outImage.pixels.resize(pixelDataSize);
+	memcpy(outImage.pixels.data(), imageData, pixelDataSize);
+
+	stbi_image_free(imageData);
+
+	return true;
+}

src/EterLib/ImageDecoder.h

@@ -0,0 +1,18 @@
+#ifndef __INC_ETERLIB_IMAGEDECODER_H__
+#define __INC_ETERLIB_IMAGEDECODER_H__
+
+#include "DecodedImageData.h"
+
+// Image decoder for worker threads
+class CImageDecoder
+{
+public:
+	// Decode image from memory (DDS, PNG, JPG, TGA, BMP)
+	static bool DecodeImage(const void* pData, size_t dataSize, TDecodedImageData& outImage);
+
+private:
+	static bool DecodeDDS(const void* pData, size_t dataSize, TDecodedImageData& outImage);
+	static bool DecodeSTB(const void* pData, size_t dataSize, TDecodedImageData& outImage);
+};
+
+#endif // __INC_ETERLIB_IMAGEDECODER_H__

src/EterLib/ResourceManager.cpp

@@ -7,6 +7,8 @@
 
 #include "ResourceManager.h"
 #include "GrpImage.h"
+#include "TextureCache.h"
+#include "DecodedImageData.h"
 
 int g_iLoadingDelayTime = 1;  // Reduced from 20ms to 1ms for faster async loading
 
@@ -68,7 +70,16 @@ void CResourceManager::ProcessBackgroundLoading()
 		}
 
 		//printf("REQ %s\n", stFileName.c_str());
-		ms_loadingThread.Request(stFileName);
+
+			if (m_pLoaderThreadPool)
+		{
+			m_pLoaderThreadPool->Request(stFileName);
+		}
+		else
+		{
+			ms_loadingThread.Request(stFileName);
+		}
+
 		m_WaitingMap.insert(TResourceRequestMap::value_type(dwFileCRC, stFileName));
 		itor = m_RequestMap.erase(itor);
 		//break; // NOTE: 여기서 break 하면 천천히 로딩 된다.
@@ -76,6 +87,44 @@ void CResourceManager::ProcessBackgroundLoading()
 
 	DWORD dwCurrentTime = ELTimer_GetMSec();
 
+	if (m_pLoaderThreadPool)
+	{
+		CFileLoaderThreadPool::TLoadResult result;
+		while (m_pLoaderThreadPool->Fetch(result))
+		{
+			CResource * pResource = GetResourcePointer(result.stFileName.c_str());
+
+			if (pResource)
+			{
+				if (pResource->IsEmpty())
+				{
+						if (result.hasDecodedImage)
+					{
+						CGraphicImage* pImage = dynamic_cast<CGraphicImage*>(pResource);
+						if (pImage)
+						{
+							pImage->OnLoadFromDecodedData(result.decodedImage);
+						}
+						else
+						{
+									pResource->OnLoad(result.File.size(), result.File.data());
+						}
+					}
+					else
+					{
+							pResource->OnLoad(result.File.size(), result.File.data());
+					}
+
+					pResource->AddReferenceOnly();
+					m_pResRefDecreaseWaitingMap.insert(TResourceRefDecreaseWaitingMap::value_type(dwCurrentTime, pResource));
+				}
+			}
+
+			m_WaitingMap.erase(GetCRC32(result.stFileName.c_str(), result.stFileName.size()));
+		}
+	}
+
+	// Process old thread results
 	CFileLoaderThread::TData * pData;
 	while (ms_loadingThread.Fetch(&pData))
 	{
@@ -528,12 +577,36 @@ void CResourceManager::ReserveDeletingResource(CResource * pResource)
 }
 
 CResourceManager::CResourceManager()
+	: m_pLoaderThreadPool(nullptr)
+	, m_pTextureCache(nullptr)
 {
 	ms_loadingThread.Create(0);
+
+	m_pLoaderThreadPool = new CFileLoaderThreadPool();
+	if (!m_pLoaderThreadPool->Initialize())
+	{
+		TraceError("CResourceManager: Failed to initialize FileLoaderThreadPool");
+		delete m_pLoaderThreadPool;
+		m_pLoaderThreadPool = nullptr;
+	}
+
+	m_pTextureCache = new CTextureCache(512);
 }
 
 CResourceManager::~CResourceManager()
 {
 	Destroy();
 	ms_loadingThread.Shutdown();
+
+	if (m_pLoaderThreadPool)
+	{
+		delete m_pLoaderThreadPool;
+		m_pLoaderThreadPool = nullptr;
+	}
+
+	if (m_pTextureCache)
+	{
+		delete m_pTextureCache;
+		m_pTextureCache = nullptr;
+	}
 }

src/EterLib/ResourceManager.h

@@ -2,11 +2,14 @@
 
 #include "Resource.h"
 #include "FileLoaderThread.h"
+#include "FileLoaderThreadPool.h"
 
 #include <set>
 #include <map>
 #include <string>
 
+class CTextureCache;
+
 class CResourceManager : public CSingleton<CResourceManager>
 {
 	public:
@@ -42,6 +45,9 @@ class CResourceManager : public CSingleton<CResourceManager>
 		void		ProcessBackgroundLoading();
 		void		PushBackgroundLoadingSet(std::set<std::string> & LoadingSet);
 
+		CTextureCache* GetTextureCache() { return m_pTextureCache; }
+		CFileLoaderThreadPool* GetLoaderThreadPool() { return m_pLoaderThreadPool; }
+
 	protected:
 		void		__DestroyDeletingResourceMap();
 		void		__DestroyResourceMap();
@@ -68,6 +74,8 @@ class CResourceManager : public CSingleton<CResourceManager>
 		TResourceRefDecreaseWaitingMap			m_pResRefDecreaseWaitingMap;
 
 		static CFileLoaderThread				ms_loadingThread;
+		CFileLoaderThreadPool*					m_pLoaderThreadPool;
+		CTextureCache*							m_pTextureCache;
 };
 
 extern int g_iLoadingDelayTime;

src/EterLib/SPSCQueue.h

@@ -0,0 +1,79 @@
+#ifndef __INC_ETERLIB_SPSCQUEUE_H__
+#define __INC_ETERLIB_SPSCQUEUE_H__
+
+#include <atomic>
+#include <vector>
+#include <cassert>
+
+// Lock-free queue for single producer/consumer pairs
+template<typename T>
+class SPSCQueue
+{
+public:
+	explicit SPSCQueue(size_t capacity)
+		: m_capacity(capacity + 1) // +1 to distinguish full from empty
+		, m_buffer(m_capacity)
+		, m_head(0)
+		, m_tail(0)
+	{
+		assert(capacity > 0);
+	}
+
+	~SPSCQueue()
+	{
+	}
+
+	// Push item (returns false if full)
+	bool Push(const T& item)
+	{
+		const size_t head = m_head.load(std::memory_order_relaxed);
+		const size_t next_head = (head + 1) % m_capacity;
+
+		if (next_head == m_tail.load(std::memory_order_acquire))
+			return false; // Queue is full
+
+		m_buffer[head] = item;
+		m_head.store(next_head, std::memory_order_release);
+		return true;
+	}
+
+	// Pop item (returns false if empty)
+	bool Pop(T& item)
+	{
+		const size_t tail = m_tail.load(std::memory_order_relaxed);
+
+		if (tail == m_head.load(std::memory_order_acquire))
+			return false; // Queue is empty
+
+		item = m_buffer[tail];
+		m_tail.store((tail + 1) % m_capacity, std::memory_order_release);
+		return true;
+	}
+
+	// Check if empty
+	bool IsEmpty() const
+	{
+		return m_tail.load(std::memory_order_acquire) == m_head.load(std::memory_order_acquire);
+	}
+
+	// Get queue size
+	size_t Size() const
+	{
+		const size_t head = m_head.load(std::memory_order_acquire);
+		const size_t tail = m_tail.load(std::memory_order_acquire);
+
+		if (head >= tail)
+			return head - tail;
+		else
+			return m_capacity - tail + head;
+	}
+
+private:
+	const size_t m_capacity;
+	std::vector<T> m_buffer;
+
+	alignas(64) std::atomic<size_t> m_head;
+	alignas(64) std::atomic<size_t> m_tail;
+};
+
+#endif // __INC_ETERLIB_SPSCQUEUE_H__

src/EterLib/TextureCache.cpp

@@ -0,0 +1,109 @@
+#include "StdAfx.h"
+#include "TextureCache.h"
+
+CTextureCache::CTextureCache(size_t maxMemoryMB)
+	: m_maxMemory(maxMemoryMB * 1024 * 1024)
+	, m_currentMemory(0)
+	, m_hits(0)
+	, m_misses(0)
+{
+}
+
+CTextureCache::~CTextureCache()
+{
+	Clear();
+}
+
+bool CTextureCache::Get(const std::string& filename, TCachedTexture& outTexture)
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+
+	auto it = m_cache.find(filename);
+	if (it == m_cache.end())
+	{
+		m_misses.fetch_add(1);
+		return false;
+	}
+
+	// Move to back of LRU (most recently used)
+	m_lruList.erase(it->second.second);
+	m_lruList.push_back(filename);
+	it->second.second = std::prev(m_lruList.end());
+
+	// Copy texture data
+	outTexture = it->second.first;
+
+	m_hits.fetch_add(1);
+	return true;
+}
+
+void CTextureCache::Put(const std::string& filename, const TCachedTexture& texture)
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+
+	// Check if already cached
+	auto it = m_cache.find(filename);
+	if (it != m_cache.end())
+	{
+		// Update existing entry
+		m_currentMemory -= it->second.first.memorySize;
+		m_lruList.erase(it->second.second);
+		m_cache.erase(it);
+	}
+
+	// Evict if needed
+	while (m_currentMemory + texture.memorySize > m_maxMemory && !m_cache.empty())
+	{
+		Evict();
+	}
+
+	// Don't cache if too large
+	if (texture.memorySize > m_maxMemory / 4)
+	{
+		return; // Skip caching huge textures
+	}
+
+	// Add to cache
+	m_lruList.push_back(filename);
+	auto lruIt = std::prev(m_lruList.end());
+	m_cache[filename] = {texture, lruIt};
+	m_currentMemory += texture.memorySize;
+}
+
+void CTextureCache::Clear()
+{
+	std::lock_guard<std::mutex> lock(m_mutex);
+	m_cache.clear();
+	m_lruList.clear();
+	m_currentMemory = 0;
+}
+
+float CTextureCache::GetHitRate() const
+{
+	size_t hits = m_hits.load();
+	size_t misses = m_misses.load();
+	size_t total = hits + misses;
+
+	if (total == 0)
+		return 0.0f;
+
+	return (float)hits / (float)total;
+}
+
+void CTextureCache::Evict()
+{
+	// Remove least recently used (front of list)
+	if (m_lruList.empty())
+		return;
+
+	const std::string& filename = m_lruList.front();
+	auto it = m_cache.find(filename);
+
+	if (it != m_cache.end())
+	{
+		m_currentMemory -= it->second.first.memorySize;
+		m_cache.erase(it);
+	}
+
+	m_lruList.pop_front();
+}

src/EterLib/TextureCache.h

@@ -0,0 +1,55 @@
+#ifndef __INC_ETERLIB_TEXTURECACHE_H__
+#define __INC_ETERLIB_TEXTURECACHE_H__
+
+#include <unordered_map>
+#include <list>
+#include <string>
+#include <mutex>
+
+// LRU cache for decoded textures
+class CTextureCache
+{
+public:
+	struct TCachedTexture
+	{
+		std::vector<uint8_t> pixels;
+		int width;
+		int height;
+		size_t memorySize;
+		std::string filename;
+	};
+
+	CTextureCache(size_t maxMemoryMB = 256);
+	~CTextureCache();
+
+	// Get cached texture
+	bool Get(const std::string& filename, TCachedTexture& outTexture);
+
+	// Add texture to cache
+	void Put(const std::string& filename, const TCachedTexture& texture);
+
+	// Clear cache
+	void Clear();
+
+	// Get statistics
+	size_t GetMemoryUsage() const { return m_currentMemory; }
+	size_t GetMaxMemory() const { return m_maxMemory; }
+	size_t GetCachedCount() const { return m_cache.size(); }
+	float GetHitRate() const;
+
+private:
+	void Evict();
+
+private:
+	size_t m_maxMemory;
+	size_t m_currentMemory;
+
+	std::list<std::string> m_lruList;
+	std::unordered_map<std::string, std::pair<TCachedTexture, std::list<std::string>::iterator>> m_cache;
+
+	mutable std::mutex m_mutex;
+	std::atomic<size_t> m_hits;
+	std::atomic<size_t> m_misses;
+};
+
+#endif // __INC_ETERLIB_TEXTURECACHE_H__

src/PRTerrainLib/TextureSet.cpp

@@ -56,6 +56,27 @@ bool CTextureSet::Load(const char * c_szTextureSetFileName, float fTerrainTexCoo
 
 	m_Textures.resize(lCount + 1);
 
+	std::vector<std::string> textureFiles;
+	textureFiles.reserve(lCount);
+
+	for (long i = 0; i < lCount; ++i)
+	{
+		_snprintf(szTextureName, sizeof(szTextureName), "texture%03d", i + 1);
+
+		if (stTokenVectorMap.end() == stTokenVectorMap.find(szTextureName))
+			continue;
+
+		const CTokenVector & rVector = stTokenVectorMap[szTextureName];
+		const std::string & c_rstrFileName = rVector[0].c_str();
+
+		textureFiles.push_back(c_rstrFileName);
+	}
+
+	for (const auto& filename : textureFiles)
+	{
+		CResourceManager::Instance().GetResourcePointer(filename.c_str());
+	}
+
 	for (long i = 0; i < lCount; ++i)
 	{
 		_snprintf(szTextureName, sizeof(szTextureName), "texture%03d", i + 1);

src/PackLib/Pack.cpp

@@ -1,6 +1,18 @@
 #include "Pack.h"
+#include "EterLib/BufferPool.h"
 #include <zstd.h>
 
+static thread_local ZSTD_DCtx* g_zstdDCtx = nullptr;
+
+static ZSTD_DCtx* GetThreadLocalZSTDContext()
+{
+	if (!g_zstdDCtx)
+	{
+		g_zstdDCtx = ZSTD_createDCtx();
+	}
+	return g_zstdDCtx;
+}
+
 bool CPack::Open(const std::string& path, TPackFileMap& entries)
 {
 	std::error_code ec;
@@ -38,27 +50,44 @@ bool CPack::Open(const std::string& path, TPackFileMap& entries)
 }
 
 bool CPack::GetFile(const TPackFileEntry& entry, TPackFile& result)
+{
+	return GetFileWithPool(entry, result, nullptr);
+}
+
+bool CPack::GetFileWithPool(const TPackFileEntry& entry, TPackFile& result, CBufferPool* pPool)
 {
 	result.resize(entry.file_size);
 
 	size_t offset = m_header.data_begin + entry.offset;
+	ZSTD_DCtx* dctx = GetThreadLocalZSTDContext();
+
 	switch (entry.encryption)
 	{
 		case 0: {
-			size_t decompressed_size = ZSTD_decompress(result.data(), result.size(), m_file.data() + offset, entry.compressed_size);
+			size_t decompressed_size = ZSTD_decompressDCtx(dctx, result.data(), result.size(), m_file.data() + offset, entry.compressed_size);
 			if (decompressed_size != entry.file_size) {
 				return false;
 			}
 		} break;
 
 		case 1: {
-			std::vector<uint8_t> compressed_data(entry.compressed_size);
+			std::vector<uint8_t> compressed_data;
+			if (pPool) {
+				compressed_data = pPool->Acquire(entry.compressed_size);
+			}
+			compressed_data.resize(entry.compressed_size);
+
 			memcpy(compressed_data.data(), m_file.data() + offset, entry.compressed_size);
 
 			m_decryption.Resynchronize(entry.iv, sizeof(entry.iv));
 			m_decryption.ProcessData(compressed_data.data(), compressed_data.data(), entry.compressed_size);
 
-			size_t decompressed_size = ZSTD_decompress(result.data(), result.size(), compressed_data.data(), compressed_data.size());
+			size_t decompressed_size = ZSTD_decompressDCtx(dctx, result.data(), result.size(), compressed_data.data(), compressed_data.size());
+
+			if (pPool) {
+				pPool->Release(std::move(compressed_data));
+			}
+
 			if (decompressed_size != entry.file_size) {
 				return false;
 			}

src/PackLib/Pack.h

@@ -4,6 +4,8 @@
 
 #include "config.h"
 
+class CBufferPool;
+
 class CPack : public std::enable_shared_from_this<CPack>
 {
 public:
@@ -12,6 +14,7 @@ public:
 
 	bool Open(const std::string& path, TPackFileMap& entries);
 	bool GetFile(const TPackFileEntry& entry, TPackFile& result);
+	bool GetFileWithPool(const TPackFileEntry& entry, TPackFile& result, CBufferPool* pPool);
 
 private:
 	TPackFileHeader m_header;

src/PackLib/PackManager.cpp

@@ -1,14 +1,38 @@
 #include "PackManager.h"
+#include "EterLib/BufferPool.h"
 #include <fstream>
 #include <filesystem>
 
+CPackManager::CPackManager()
+	: m_load_from_pack(true)
+	, m_pBufferPool(nullptr)
+{
+	m_pBufferPool = new CBufferPool();
+}
+
+CPackManager::~CPackManager()
+{
+	if (m_pBufferPool)
+	{
+		delete m_pBufferPool;
+		m_pBufferPool = nullptr;
+	}
+}
+
 bool CPackManager::AddPack(const std::string& path)
 {
 	std::shared_ptr<CPack> pack = std::make_shared<CPack>();
+
+	std::lock_guard<std::mutex> lock(m_mutex);
 	return pack->Open(path, m_entries);
 }
 
 bool CPackManager::GetFile(std::string_view path, TPackFile& result)
+{
+	return GetFileWithPool(path, result, m_pBufferPool);
+}
+
+bool CPackManager::GetFileWithPool(std::string_view path, TPackFile& result, CBufferPool* pPool)
 {
 	thread_local std::string buf;
 	NormalizePath(path, buf);
@@ -16,7 +40,7 @@ bool CPackManager::GetFile(std::string_view path, TPackFile& result)
 	if (m_load_from_pack) {
 		auto it = m_entries.find(buf);
 		if (it != m_entries.end()) {
-			return it->second.first->GetFile(it->second.second, result);
+			return it->second.first->GetFileWithPool(it->second.second, result, pPool);
 		}
 	}
 	else {
@@ -25,7 +49,14 @@ bool CPackManager::GetFile(std::string_view path, TPackFile& result)
 			ifs.seekg(0, std::ios::end);
 			size_t size = ifs.tellg();
 			ifs.seekg(0, std::ios::beg);
-			result.resize(size);
+
+			if (pPool) {
+				result = pPool->Acquire(size);
+				result.resize(size);
+			} else {
+				result.resize(size);
+			}
+
 			if (ifs.read((char*)result.data(), size)) {
 				return true;
 			}

src/PackLib/PackManager.h

@@ -1,26 +1,34 @@
 #pragma once
 #include <unordered_map>
+#include <mutex>
 
 #include "EterBase/Singleton.h"
 #include "Pack.h"
 
+class CBufferPool;
+
 class CPackManager : public CSingleton<CPackManager>
 {
 public:
-	CPackManager() = default;
+	CPackManager();
-	virtual ~CPackManager() = default;
+	virtual ~CPackManager();
 
 	bool AddPack(const std::string& path);
 	bool GetFile(std::string_view path, TPackFile& result);
+	bool GetFileWithPool(std::string_view path, TPackFile& result, CBufferPool* pPool);
 	bool IsExist(std::string_view path) const;
 
 	void SetPackLoadMode() { m_load_from_pack = true; }
 	void SetFileLoadMode() { m_load_from_pack = false; }
 
+	CBufferPool* GetBufferPool() { return m_pBufferPool; }
+
 private:
 	void NormalizePath(std::string_view in, std::string& out) const;
 
 private:
 	bool m_load_from_pack = true;
 	TPackFileMap m_entries;
+	CBufferPool* m_pBufferPool;
+	mutable std::mutex m_mutex;  // Thread safety for parallel pack loading
 };

src/UserInterface/UserInterface.cpp

@@ -16,6 +16,8 @@
 
 #include <filesystem>
 #include <format>
+#include <thread>
+#include <atomic>
 
 #include <stdlib.h>
 #include <utf8.h>
@@ -166,11 +168,45 @@ bool PackInitialize(const char * c_pszFolder)
 		"uiloading",
 	};
 
-	CPackManager::instance().AddPack(std::format("{}/root.pck", c_pszFolder));
+	Tracef("PackInitialize: Loading root.pck...");
-	for (const std::string& packFileName : packFiles) {
+	if (!CPackManager::instance().AddPack(std::format("{}/root.pck", c_pszFolder)))
-		CPackManager::instance().AddPack(std::format("{}/{}.pck", c_pszFolder, packFileName));
+	{
+		TraceError("Failed to load root.pck");
+		return false;
 	}
 
+	Tracef("PackInitialize: Loading %d pack files in parallel...", packFiles.size());
+	const size_t numThreads = std::min<size_t>(std::thread::hardware_concurrency(), packFiles.size());
+	const size_t packsPerThread = (packFiles.size() + numThreads - 1) / numThreads;
+
+	std::vector<std::thread> threads;
+	std::atomic<size_t> failedCount(0);
+
+	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());
+
+			for (size_t i = start; i < end; ++i)
+			{
+				std::string packPath = std::format("{}/{}.pck", c_pszFolder, packFiles[i]);
+				if (!CPackManager::instance().AddPack(packPath))
+				{
+					TraceError("Failed to load %s", packPath.c_str());
+					failedCount++;
+				}
+			}
+		});
+	}
+
+	// Wait for all threads to complete
+	for (auto& thread : threads)
+	{
+		thread.join();
+	}
+
+	Tracef("PackInitialize: Completed! Failed: %d / %d", failedCount.load(), packFiles.size());
 	return true;
 }