v2: Initial clean branch with unit tests + CI/CD pipeline

- Started from ac75faa (initial E4B-MarkBase integration) - Kept Sources/ (all engine code) + Package.swift + .gitignore - Removed all ad-hoc tests, documentation, scripts, Python files - Added Tests/00_Unit/ (MathTest, TokenizerTest, SamplerTest) - Added .gitea/workflows/ci.yaml (build + unit tests + lint) - Added Scripts/check_resources.sh (memory-aware test runner) - Added Tests/Manifest.json (resource requirements for all tests) - Focus: 4-bit quantized models only
2026-07-05 13:29:25 +08:00
commit 8a66b9086a
90 changed files with 22252 additions and 0 deletions
@@ -0,0 +1,89 @@
+import Metal
+
+/// Buffer pool for reusing MTLBuffers to reduce allocation overhead
+/// 
+/// Metal buffer allocation can be expensive, especially when done frequently
+/// during inference. This pool caches and reuses buffers of common sizes.
+public final class BufferPool: @unchecked Sendable {
+    private let device: MTLDevice
+    private var availableBuffers: [Int: [MTLBuffer]] = [:]  // size -> [buffers]
+    private let lock = NSLock()
+    
+    /// Statistics
+    public private(set) var totalAllocations: Int = 0
+    public private(set) var totalReuses: Int = 0
+    public private(set) var peakBufferCount: Int = 0
+    
+    public init(device: MTLDevice) {
+        self.device = device
+    }
+    
+    /// Acquire a buffer of the specified size
+    /// Returns a reusable buffer if available, otherwise creates a new one
+    public func acquire(length: Int) -> MTLBuffer {
+        lock.lock()
+        defer { lock.unlock() }
+        
+        // Round up to nearest 256 bytes for alignment
+        let alignedLength = (length + 255) & ~255
+        
+        // Check for available buffer
+        if var buffers = availableBuffers[alignedLength], !buffers.isEmpty {
+            let buffer = buffers.removeLast()
+            availableBuffers[alignedLength] = buffers
+            totalReuses += 1
+            return buffer
+        }
+        
+        // Create new buffer
+        totalAllocations += 1
+        guard let buffer = device.makeBuffer(
+            length: alignedLength,
+            options: .storageModeShared
+        ) else {
+            fatalError("Failed to allocate Metal buffer of size \(alignedLength)")
+        }
+        
+        // Track peak
+        let currentCount = totalAllocations - totalReuses
+        if currentCount > peakBufferCount {
+            peakBufferCount = currentCount
+        }
+        
+        return buffer
+    }
+    
+    /// Release a buffer back to the pool for reuse
+    public func release(_ buffer: MTLBuffer) {
+        lock.lock()
+        defer { lock.unlock() }
+        
+        let length = buffer.length
+        if availableBuffers[length] == nil {
+            availableBuffers[length] = []
+        }
+        availableBuffers[length]?.append(buffer)
+    }
+    
+    /// Clear all cached buffers (useful for memory pressure situations)
+    public func clear() {
+        lock.lock()
+        defer { lock.unlock() }
+        availableBuffers.removeAll()
+    }
+    
+    /// Get pool statistics
+    public var stats: String {
+        lock.lock()
+        defer { lock.unlock() }
+        let totalBuffers = availableBuffers.values.reduce(0) { $0 + $1.count }
+        return """
+        BufferPool Stats:
+          Allocations: \(totalAllocations)
+          Reuses: \(totalReuses)
+          Available: \(totalBuffers)
+          Peak: \(peakBufferCount)
+          Hit Rate: \(totalAllocations > 0 ? String(format: "%.1f%%", Float(totalReuses) / Float(totalAllocations) * 100) : "0%")
+        """
+    }
+}