Merge commit 'a51b88cccfbc2e51fbf5ce64b564002000548ec5' into Prompt/Allocator

include/nd-wfc/wfc.hpp

@@ -11,7 +11,8 @@
 #include <algorithm>
 #include <concepts>
 #include <bit>
-#include <iostream>
+
+#include "wfc_allocator.hpp"
 
 namespace WFC {
 
@@ -144,7 +145,7 @@ template <typename MaskType>
 class Wave {
 public:
     Wave() = default;
-    Wave(size_t size, size_t variableAmount) : m_data(size) 
+    Wave(size_t size, size_t variableAmount, WFCStackAllocator& allocator) : m_data(size, WFCStackAllocatorAdapter<MaskType>(allocator)) 
     {
         for (auto& wave : m_data) wave = (1 << variableAmount) - 1;
     }
@@ -163,7 +164,7 @@ public:
     MaskType GetMask(size_t index) const { return m_data[index]; }
 
 private:
-    std::vector<MaskType> m_data;
+    WFCVector<MaskType> m_data;
 };
 
 /**
@@ -175,7 +176,7 @@ public:
     using MaskType = typename VariableIDMapT::MaskType;
 
 public:
-    Constrainer(Wave<MaskType>& wave, std::queue<size_t>& propagationQueue)
+    Constrainer(Wave<MaskType>& wave, WFCQueue<size_t>& propagationQueue)
         : m_wave(wave)
         , m_propagationQueue(propagationQueue)
     {}
@@ -224,7 +225,7 @@ private:
 
 private:
     Wave<MaskType>& m_wave;
-    std::queue<size_t>& m_propagationQueue;
+    WFCQueue<size_t>& m_propagationQueue;
 };
 
 /**
@@ -255,14 +256,17 @@ public:
 public:
     struct SolverState {
         WorldT& world;
-        std::queue<size_t> propagationQueue{};
-        Wave<MaskType> wave{};
+        WFCQueue<size_t> propagationQueue;
+        Wave<MaskType> wave;
         std::mt19937& rng;
+        WFCStackAllocator& allocator;
 
-        SolverState(WorldT& world, size_t variableAmount, std::mt19937& rng)
+        SolverState(WorldT& world, size_t variableAmount, std::mt19937& rng, WFCStackAllocator& allocator)
             : world(world)
-            , wave(world.size(), variableAmount)
+            , propagationQueue{ WFCStackAllocatorAdapter<size_t>(allocator) }
+            , wave{ world.size(), variableAmount, allocator }
             , rng(rng)
+            , allocator(allocator)
         {}
 
         SolverState(const SolverState& other) = default;
@@ -276,10 +280,9 @@ public:
 public:
     bool Run(WorldT& world, bool propagateInitialValues = false)
     {
-        //auto seed = std::random_device{}();
-        auto seed = 1844803044ul;
-        std::mt19937 random{ seed };
-        SolverState state(world, m_variables.size(), random);
+        WFCStackAllocator allocator{};
+        std::mt19937 random{ std::random_device{}() };
+        SolverState state(world, m_variables.size(), random, allocator);
         return Run(state, propagateInitialValues);
     }
 
@@ -388,6 +391,7 @@ private:
             
             {
                 // copy the state and branch out
+                auto stackFrame = state.allocator.createFrame();
                 SolverState newState(state);
                 newState.wave.Collapse(minEntropyCell, 1 << selectedValue);
                 newState.propagationQueue.push(minEntropyCell);

include/nd-wfc/wfc_allocator.hpp

@@ -0,0 +1,245 @@
+#pragma once
+
+#include <cstddef>
+#include <memory>
+#include <cassert>
+#include <vector>
+#include <queue>
+#include <cstring>
+
+namespace WFC {
+
+/**
+ * @brief Stack allocator specifically designed for WFC branching operations
+ * 
+ * This allocator uses a stack-based approach where memory is allocated in chunks
+ * and automatically deallocated when branches go out of scope. It's optimized
+ * for the recursive branching pattern used in the WFC algorithm.
+ */
+class WFCStackAllocator {
+private:
+    struct Block {
+        void* ptr;
+        size_t size;
+        size_t alignment;
+        
+        Block() : ptr(nullptr), size(0), alignment(0) {}
+        Block(void* p, size_t s, size_t a) : ptr(p), size(s), alignment(a) {}
+    };
+
+public:
+    /**
+     * @brief Construct allocator with initial capacity
+     * @param initialCapacity Initial memory pool size in bytes
+     */
+    explicit WFCStackAllocator(size_t initialCapacity = 1024 * 1024) // 1MB default
+        : m_capacity(initialCapacity)
+        , m_used(0)
+        , m_pool(nullptr)
+    {
+        m_pool = std::aligned_alloc(64, m_capacity); // 64-byte alignment for cache efficiency
+        if (!m_pool) {
+            throw std::bad_alloc();
+        }
+    }
+
+    ~WFCStackAllocator() {
+        if (m_pool) {
+            std::free(m_pool);
+        }
+    }
+
+    // Non-copyable, non-movable for safety
+    WFCStackAllocator(const WFCStackAllocator&) = delete;
+    WFCStackAllocator& operator=(const WFCStackAllocator&) = delete;
+    WFCStackAllocator(WFCStackAllocator&&) = delete;
+    WFCStackAllocator& operator=(WFCStackAllocator&&) = delete;
+
+    /**
+     * @brief Allocate memory from the stack
+     * @param size Number of bytes to allocate
+     * @param alignment Memory alignment requirement (default 8)
+     * @return Pointer to allocated memory
+     */
+    void* allocate(size_t size, size_t alignment = 8) {
+        // Align the current position
+        size_t alignedUsed = alignUp(m_used, alignment);
+        
+        // Check if we have enough space
+        if (alignedUsed + size > m_capacity) {
+            // Grow the pool if needed
+            growPool(alignedUsed + size);
+        }
+        
+        void* ptr = static_cast<char*>(m_pool) + alignedUsed;
+        m_used = alignedUsed + size;
+        
+        // Track this allocation for proper cleanup
+        m_allocations.emplace_back(ptr, size, alignment);
+        
+        return ptr;
+    }
+
+    /**
+     * @brief Deallocate memory (no-op as requested - memory freed when branch goes out of scope)
+     * @param ptr Pointer to deallocate
+     */
+    void deallocate(void*) {
+        // No-op as requested - deallocation happens when branch goes out of scope
+        // The stack nature ensures automatic cleanup
+    }
+
+    /**
+     * @brief Create a stack frame marker for RAII-based cleanup
+     * @return StackFrame object that will cleanup on destruction
+     */
+    class StackFrame {
+    private:
+        WFCStackAllocator& m_allocator;
+        size_t m_savedUsed;
+        size_t m_savedAllocCount;
+
+    public:
+        StackFrame(WFCStackAllocator& allocator) 
+            : m_allocator(allocator)
+            , m_savedUsed(allocator.m_used)
+            , m_savedAllocCount(allocator.m_allocations.size())
+        {}
+
+        ~StackFrame() {
+            // Restore the stack to the previous state
+            m_allocator.m_used = m_savedUsed;
+            m_allocator.m_allocations.resize(m_savedAllocCount);
+        }
+
+        // Non-copyable, movable
+        StackFrame(const StackFrame&) = delete;
+        StackFrame& operator=(const StackFrame&) = delete;
+        StackFrame(StackFrame&&) = default;
+        StackFrame& operator=(StackFrame&&) = default;
+    };
+
+    /**
+     * @brief Create a new stack frame for a branch
+     */
+    StackFrame createFrame() {
+        return StackFrame(*this);
+    }
+
+    /**
+     * @brief Get current memory usage
+     */
+    size_t getUsed() const { return m_used; }
+
+    /**
+     * @brief Get total capacity
+     */
+    size_t getCapacity() const { return m_capacity; }
+
+    /**
+     * @brief Get allocation count
+     */
+    size_t getAllocationCount() const { return m_allocations.size(); }
+
+    /**
+     * @brief Reset the allocator (useful for reusing between WFC runs)
+     */
+    void reset() {
+        m_used = 0;
+        m_allocations.clear();
+    }
+
+private:
+    size_t alignUp(size_t value, size_t alignment) const {
+        return (value + alignment - 1) & ~(alignment - 1);
+    }
+
+    void growPool(size_t requiredSize) {
+        size_t newCapacity = std::max(m_capacity * 2, requiredSize);
+        
+        void* newPool = std::aligned_alloc(64, newCapacity);
+        if (!newPool) {
+            throw std::bad_alloc();
+        }
+        
+        // Copy existing data
+        std::memcpy(newPool, m_pool, m_used);
+        
+        // Update all existing pointers in allocations
+        ptrdiff_t offset = static_cast<char*>(newPool) - static_cast<char*>(m_pool);
+        for (auto& block : m_allocations) {
+            block.ptr = static_cast<char*>(block.ptr) + offset;
+        }
+        
+        std::free(m_pool);
+        m_pool = newPool;
+        m_capacity = newCapacity;
+    }
+
+private:
+    size_t m_capacity;
+    size_t m_used;
+    void* m_pool;
+    std::vector<Block> m_allocations;
+};
+
+/**
+ * @brief Custom allocator adapter for STL containers using WFCStackAllocator
+ */
+template<typename T>
+class WFCStackAllocatorAdapter {
+public:
+    using value_type = T;
+    using pointer = T*;
+    using const_pointer = const T*;
+    using reference = T&;
+    using const_reference = const T&;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    template<typename U>
+    struct rebind {
+        using other = WFCStackAllocatorAdapter<U>;
+    };
+
+    WFCStackAllocatorAdapter(WFCStackAllocator& allocator) 
+        : m_allocator(&allocator) {}
+
+    template<typename U>
+    WFCStackAllocatorAdapter(const WFCStackAllocatorAdapter<U>& other) 
+        : m_allocator(other.m_allocator) {}
+
+    pointer allocate(size_type n) {
+        return static_cast<pointer>(m_allocator->allocate(n * sizeof(T), alignof(T)));
+    }
+
+    void deallocate(pointer ptr, size_type) {
+        m_allocator->deallocate(ptr);
+    }
+
+    template<typename U>
+    bool operator==(const WFCStackAllocatorAdapter<U>& other) const {
+        return m_allocator == other.m_allocator;
+    }
+
+    template<typename U>
+    bool operator!=(const WFCStackAllocatorAdapter<U>& other) const {
+        return !(*this == other);
+    }
+
+    WFCStackAllocator* m_allocator;
+};
+
+/**
+ * @brief Stack-allocated vector using WFCStackAllocator
+ */
+template<typename T>
+using WFCVector = std::vector<T, WFCStackAllocatorAdapter<T>>;
+
+/**
+ * @brief Stack-allocated queue using WFCStackAllocator
+ */
+template<typename T>
+using WFCQueue = std::queue<T, std::deque<T, WFCStackAllocatorAdapter<T>>>;
+
+} // namespace WFC