Heap Allocator now commits and releases pages dynamically.

Implemented Heap::free
Heap::reset() now clears m_usedSize

.github/workflows/cmake-multi-platform.yml

@@ -76,7 +76,8 @@ jobs:
       run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }}
 
     - name: Test
-      working-directory: ${{ steps.strings.outputs.build-output-dir }}
+      working-directory: ${{ steps.strings.outputs.build-output-dir }}/tests
       # Execute tests defined by the CMake configuration. Note that --build-config is needed because the default Windows generator is a multi-config generator (Visual Studio generator).
       # See https://cmake.org/cmake/help/latest/manual/ctest.1.html for more detail
-      run: ctest --build-config ${{ matrix.build_type }}
+      run: ctest --output-on-failure --no-tests=error --build-config ${{ matrix.build_type }}
+

.gitignore

@@ -1 +1,2 @@
+.vs
 build

CMakeLists.txt

@@ -9,13 +9,17 @@ set(PUBLIC_HEADERS
   include/mimir.h
   include/arena.h
   include/block.h
+  include/heap.h
+  include/region.h
   include/util.h
 )
 
 set(SRCS
   src/arena.cpp
   src/block.cpp
+  src/heap.cpp
   src/mimir.cpp
+  src/region.cpp
   src/util.cpp
 )
 
@@ -37,3 +41,5 @@ install(
     PUBLIC_HEADER
       DESTINATION include/mimir
 )
+
+add_subdirectory(tests)

include/arena.h

@@ -31,29 +31,47 @@
 
 #pragma once
 
-#include <string>
 #include <cstddef>
 
+#include "region.h"
+
 namespace mimir {
+
+// Arena memory allocator, baed by a mimir::Region
 class Arena
 {
 public:
-  Arena(size_t minSize = 1ULL << 24, size_t maxSize = 1ULL << 32);
+  Arena();
   ~Arena();
+  bool init(size_t minSize = 1ULL << 24, size_t maxSize = 1ULL << 32);
 
+  // Allocate `size` bytes
   std::byte* alloc(size_t size);
+
+  // Allocate `size` bytes, aligned to 16 bytes and padded to next 16-byte offset.
   std::byte* allocA16(size_t size);
+
+  // Allocate `size` bytes, aligned to 64 bytes and padded to next 64-byte offset.
   std::byte* allocA64(size_t size);
+
+  // Reset the arena, potentially also releasing committed pages
   void reset();
+
+  // Get the actual used size.
+  // This may differ from the sum of allocations due to alignment requirements.
+  size_t getUsed() const;
+
+  // Get the actual maximum size.
+  // This may be greater than the maxSize passed into init, due to page size alignment.
+  size_t getMaxSize() const;
 private:
-  std::byte* m_data;
+  Region m_region;
   size_t m_minSize;
-  size_t m_maxSize;
-  size_t m_committedSize;
   size_t m_usedSize;
 
   static const size_t kWatermarkLen = 8;
   size_t m_watermark[kWatermarkLen];
+
 }; // class Arena
 
 } // namespace mimir

include/block.h

@@ -35,6 +35,7 @@
 #include <stdint.h>
 
 #if defined(_WIN32) || defined(_WIN64)
+#define NOMINMAX
 #include <Windows.h>
 #endif
 

include/heap.h

@@ -0,0 +1,101 @@
+//
+//  heap.h
+//  Kraken Engine / Mimir
+//
+//  Copyright 2026 Kearwood Gilbert. All rights reserved.
+//
+//  Redistribution and use in source and binary forms, with or without modification, are
+//  permitted provided that the following conditions are met:
+//
+//  1. Redistributions of source code must retain the above copyright notice, this list of
+//  conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright notice, this list
+//  of conditions and the following disclaimer in the documentation and/or other materials
+//  provided with the distribution.
+//
+//  THIS SOFTWARE IS PROVIDED BY KEARWOOD GILBERT ''AS IS'' AND ANY EXPRESS OR IMPLIED
+//  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+//  FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL KEARWOOD GILBERT OR
+//  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+//  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+//  SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+//  ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+//  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+//  ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+//  The views and conclusions contained in the software and documentation are those of the
+//  authors and should not be interpreted as representing official policies, either expressed
+//  or implied, of Kearwood Gilbert.
+//
+
+#pragma once
+
+#include <cstddef>
+
+#include "region.h"
+
+namespace mimir {
+
+struct TLSFBlock;
+
+// Heap implementation based on Two-Level Segregated Fit (TLSF) memory allocation
+class Heap
+{
+public:
+  Heap();
+  ~Heap();
+
+  bool init(size_t minSize = 1ULL << 24, size_t maxSize = 1ULL << 32);
+
+  // Allocate `size` bytes
+  std::byte* alloc(size_t size);
+
+  // Allocate `size` bytes, aligned to 16 bytes and padded to next 16-byte offset.
+  std::byte* allocA16(size_t size);
+
+  // Allocate `size` bytes, aligned to 64 bytes and padded to next 64-byte offset.
+  std::byte* allocA64(size_t size);
+
+  // Free the allocation at `address`
+  void free(std::byte* address);
+
+  // Reset the heap, freeing all memory and potentially releasing comitted pages.
+  void reset();
+
+  // Get the actual used size.
+  // This may differ from the sum of allocations due to alignment requirements.
+  size_t getUsed() const;
+
+  // Get the actual maximum size.
+  // This may be greater than the maxSize passed into init, due to page size alignment.
+  size_t getMaxSize() const;
+
+private:
+  Region m_region;
+  size_t m_minSize;
+  size_t m_usedSize;
+
+  // Add a free block to the index
+  void insertFreeBlock(TLSFBlock* block);
+
+  // Remove a free block from the index
+  void removeFreeBlock(TLSFBlock* block);
+
+  // Find a free block that can hold at least size bytes
+  TLSFBlock* findFreeBlock(size_t size) const;
+
+  // Get the usable size of a block
+  size_t getBlockUsableSize(const TLSFBlock* block) const;
+
+  // Get the next block, in physical order
+  TLSFBlock* getNextPhysBlock(TLSFBlock* block);
+
+  // Returns true if block is a free block
+  bool isBlockFree(const TLSFBlock* block) const;
+
+  // Return true if block is the last block, in physical order
+  bool isBlockLast(const TLSFBlock* block) const;
+};
+
+} // namespace mimir

include/mimir.h

@@ -31,8 +31,10 @@
 
 #pragma once
 
-#include "block.h"
 #include "arena.h"
+#include "block.h"
+#include "heap.h"
+#include "region.h"
 #include "util.h"
 
 namespace mimir {

include/region.h

@@ -0,0 +1,62 @@
+//
+//  region.h
+//  Kraken Engine / Mimir
+//
+//  Copyright 2026 Kearwood Gilbert. All rights reserved.
+//
+//  Redistribution and use in source and binary forms, with or without modification, are
+//  permitted provided that the following conditions are met:
+//
+//  1. Redistributions of source code must retain the above copyright notice, this list of
+//  conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright notice, this list
+//  of conditions and the following disclaimer in the documentation and/or other materials
+//  provided with the distribution.
+//
+//  THIS SOFTWARE IS PROVIDED BY KEARWOOD GILBERT ''AS IS'' AND ANY EXPRESS OR IMPLIED
+//  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+//  FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL KEARWOOD GILBERT OR
+//  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+//  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+//  SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+//  ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+//  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+//  ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+//  The views and conclusions contained in the software and documentation are those of the
+//  authors and should not be interpreted as representing official policies, either expressed
+//  or implied, of Kearwood Gilbert.
+//
+
+#pragma once
+
+#include <cstddef>
+
+namespace mimir {
+
+// Dynamically growable memory region with a contiguous address range.
+// Provides OS and hardware agnostic reservation and committing of memory pages.
+class Region
+{
+public:
+  Region();
+  ~Region();
+
+  bool init(size_t maxSize = 1ULL << 32);
+  bool resize(size_t size);
+  size_t getSize() const;
+  size_t getMaxSize() const;
+  std::byte* getAddress() const;
+
+  // Helper functions for allocators
+  bool maybeGrow(size_t minSize);
+  void maybeShrink(size_t targetSize);
+
+private:
+  std::byte* m_data;
+  size_t m_committedSize;
+  size_t m_maxSize;
+};
+
+} // namespace mimir

src/arena.cpp

@@ -32,43 +32,16 @@
 #include "../include/mimir.h"
 #include "mimir_impl.h"
 
-#if defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
-#include <unistd.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#endif
-
-#if defined(_WIN32) || defined(_WIN64)
-#include <Windows.h>
-#include <memoryapi.h>
-#endif
-
-#include <errno.h>
 #include <cassert>
 #include <cstring>
 #include <bit>
 
 namespace mimir {
 
-struct ArenaControlBlock
+Arena::Arena()
-{
+  : m_minSize(0)
-  size_t allocated;
-  size_t comitted;
-  size_t next;
-};
-
-static_assert(sizeof(ArenaControlBlock) < 4096);
-
-Arena::Arena(size_t minSize, size_t maxSize)
-  : m_minSize(minSize)
-  , m_maxSize(maxSize)
-  , m_data(nullptr)
   , m_usedSize(0)
-  , m_committedSize(0)
 {
-  assert(maxSize >= minSize);
-
   for (size_t i = 0; i < kWatermarkLen; i++) {
     m_watermark[i] = 0;
   }
@@ -76,88 +49,63 @@ Arena::Arena(size_t minSize, size_t maxSize)
 
 Arena::~Arena()
 {
-  if (m_data) {
+}
-#if defined(_WIN32) || defined(_WIN64)
+
-    if (!VirtualFree((void*)m_data, 0, MEM_RELEASE)) {
+bool Arena::init(size_t minSize, size_t maxSize)
-      ReportWindowsLastError("VirtualAlloc");
+{
-    }
+  assert(maxSize >= minSize);
-#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+  assert(m_usedSize == 0);
-    munmap(m_data, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize));
+
-#else
+  m_minSize = KRAKEN_MEM_ROUND_UP_PAGE(minSize);
-    static_assert(false, "Not Implemented");
+  return m_region.init(maxSize);
-#endif
+}
-  }
+
+// Get the actual used size.
+// This may differ from the sum of allocations due to alignment requirements.
+size_t Arena::getUsed() const
+{
+  return m_usedSize;
+}
+
+// Get the actual maximum size.
+// This may be greater than the maxSize passed into init, due to page size alignment.
+size_t Arena::getMaxSize() const
+{
+  return m_region.getMaxSize();
 }
 
 std::byte* Arena::alloc(size_t size)
 {
   size_t neededSize = m_usedSize + size;
-
+  if (!m_region.maybeGrow(std::max(neededSize, m_minSize))) {
-  if (m_data == nullptr) {
-#if defined(_WIN32) || defined(_WIN64)
-    m_data = (std::byte*)VirtualAlloc(NULL, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize), MEM_RESERVE, PAGE_NOACCESS);
-    if (m_data == nullptr) {
-      ReportWindowsLastError("VirtualAlloc");
     return nullptr;
   }
-#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
-    m_data = (std::byte*)mmap(NULL, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize), PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-    if (m_data == MAP_FAILED) {
-      return nullptr;
-    }
-#else
-    static_assert(false, "Not Implemented");
-#endif
-  }
 
-  if (neededSize > m_committedSize)
+  std::byte* ret = m_region.getAddress() + m_usedSize;
-  {
-    size_t newSize = std::bit_ceil(neededSize) << 1;
-    if (newSize < m_minSize) {
-      newSize = m_minSize;
-    }
-    newSize = KRAKEN_MEM_ROUND_UP_PAGE(newSize);
-
-#if defined(_WIN32) || defined(_WIN64)
-    if(VirtualAlloc(m_data + m_committedSize, newSize - m_committedSize, MEM_COMMIT, PAGE_READWRITE) == nullptr) {
-      ReportWindowsLastError("VirtualAlloc");
-      return nullptr;
-    }
-#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
-    if (mprotect(m_data + m_committedSize, newSize - m_committedSize, PROT_READ | PROT_WRITE) != 0) {
-      return nullptr;
-    }
-#else
-    static_assert(false, "Not Implemented");
-#endif
-    m_committedSize = newSize;
-  } // if (neededSize > m_committedSize)
-
-  std::byte* ret = m_data + m_usedSize;
   m_usedSize = neededSize;
   return ret;
 }
 
 std::byte* Arena::allocA16(size_t size)
 {
-  uint32_t nextByte = (m_usedSize + 15) & ~15;
+  size_t nextByte = (m_usedSize + 15) & ~15;
-  uint32_t roundedSize = (size + 15) & ~15;
+  size_t roundedSize = (size + 15) & ~15;
-  if (alloc(size) == nullptr)     {
+  if (alloc(roundedSize + nextByte - m_usedSize) == nullptr)     {
     return nullptr;
   }
 
-  return m_data + nextByte;
+  return m_region.getAddress() + nextByte;
 }
 
 std::byte* Arena::allocA64(size_t size)
 {
-  uint32_t nextByte = (m_usedSize + 63) & ~63;
+  size_t nextByte = (m_usedSize + 63) & ~63;
-  uint32_t roundedSize = (size + 63) & ~63;
+  size_t roundedSize = (size + 63) & ~63;
-  if (alloc(size) == nullptr) {
+  if (alloc(roundedSize + nextByte - m_usedSize) == nullptr) {
     return nullptr;
   }
 
-  return m_data + nextByte;
+  return m_region.getAddress() + nextByte;
 }
 
 void Arena::reset()
@@ -172,33 +120,7 @@ void Arena::reset()
   m_watermark[kWatermarkLen - 1] = m_usedSize;
 
   size_t targetSize = std::bit_ceil(highWatermark) << 1;
-  if (targetSize < m_minSize) {
+  m_region.maybeShrink(std::max(m_minSize, targetSize));
-    targetSize = m_minSize;
-  }
-  size_t thresholdSize = targetSize << 1; // The threshold for shrinking is greater than the target to implement hysteresis
-  if (m_committedSize > thresholdSize) {
-    size_t newSize = KRAKEN_MEM_ROUND_UP_PAGE(targetSize);
-
-    std::byte* rangeStart = m_data + newSize;
-    size_t rangeLen = m_committedSize - newSize;
-#if defined(_WIN32) || defined(_WIN64)
-    if (!VirtualAlloc(rangeStart, rangeLen, MEM_RESET, PAGE_NOACCESS)) {
-      ReportWindowsLastError("VirtualAlloc");
-    } else {
-      m_committedSize = newSize;
-    }
-#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
-    if (madvise(rangeStart, rangeLen, MADV_DONTNEED) != 0) {
-      // TODO - Log Error or debug build assert
-    }
-    if (mprotect(rangeStart, rangeLen, PROT_NONE) != 0) {
-      // TODO - Log Error or debug build assert
-    }
-#else
-    static_assert(false, "Not Implemented");
-#endif
-    m_committedSize = newSize;
-  }
 
   m_usedSize = 0;
 }

src/block.cpp

@@ -32,16 +32,9 @@
 #include "../include/mimir.h"
 #include "mimir_impl.h"
 
-#if defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
-#include <unistd.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#endif
-
-#include <errno.h>
 #include <cassert>
 #include <cstring>
+#include <bit>
 
 #define KRENGINE_MIN_MMAP 32768
 

src/heap.cpp

@@ -0,0 +1,417 @@
+//
+//  heap.cpp
+//  Kraken Engine
+//
+//  Copyright 2026 Kearwood Gilbert. All rights reserved.
+//
+//  Redistribution and use in source and binary forms, with or without modification, are
+//  permitted provided that the following conditions are met:
+//
+//  1. Redistributions of source code must retain the above copyright notice, this list of
+//  conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright notice, this list
+//  of conditions and the following disclaimer in the documentation and/or other materials
+//  provided with the distribution.
+//
+//  THIS SOFTWARE IS PROVIDED BY KEARWOOD GILBERT ''AS IS'' AND ANY EXPRESS OR IMPLIED
+//  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+//  FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL KEARWOOD GILBERT OR
+//  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+//  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+//  SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+//  ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+//  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+//  ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+//  The views and conclusions contained in the software and documentation are those of the
+//  authors and should not be interpreted as representing official policies, either expressed
+//  or implied, of Kearwood Gilbert.
+//
+
+#include "../include/mimir.h"
+#include "mimir_impl.h"
+
+#include <algorithm>
+#include <cassert>
+#include <cstring>
+#include <bit>
+
+namespace mimir {
+
+// Two-Level Segregated Fit (TLSF): http://www.gii.upv.es/tlsf/files/papers/ecrts04_tlsf.pdf
+
+const size_t kMinBlockSize = 16;
+
+struct TLSFBlock
+{
+  size_t size; // LSB: T, F: T = Last physical block, F = Free block
+  TLSFBlock* prevPhys;
+  TLSFBlock* prevFree;
+  TLSFBlock* nextFree;
+};
+static_assert(sizeof(TLSFBlock) == 32);
+
+struct TLSFIndex
+{
+  uint64_t firstLevelFreeBitMap;
+  uint8_t secondLevelFreeBitMap[60];
+  TLSFBlock* secondLevelFreeBlocks[60][16];
+};
+
+// First Level Ranges:
+// 0:  0b00000000010000  (    1 ~   16 )
+// 1:  0b00000000100000  (   17 ~   32 )
+// 2:  0b00000001000000  (   33 ~   64 )
+// 3:  0b00000010000000  (   65 ~  128 )
+// 4:  0b00000100000000  (  129 ~  256 )
+// 5:  0b00001000000000  (  257 ~  512 )
+// 6:  0b00010000000000  (  513 ~ 1024 )
+// 7:  0b00100000000000  ( 1025 ~ 2048 )
+//
+// bit_width(15): 4    bit_floor(15): 0b000001000
+// bit_width(16): 5    bit_floor(16): 0b000010000
+// bit_width(31): 5    bit_floor(31): 0b000010000
+// bit_width(32): 6    bit_floor(32): 0b000100000
+// bit_width(63): 6    bit_floor(63): 0b000100000
+// bit_width(64): 7    bit_floor(64): 0b001000000
+
+/*
+uint64_t blockSizeToIndex(uint64_t size)
+{
+  // MSB 60 bits are the first level index
+  // LSB 4 bits is the second level index
+
+  // size              0b0000001101011010000
+  // bit_width         13
+  // bit_width - 5     8
+  // size >> 8         0b0000000000000011010
+  //   & 0b1111        0b1010
+
+  size_t firstLevelIndex = std::bit_width(size) - 5;
+  size_t secondLevelIndex = (size >> firstLevelIndex) & 0b1111;
+  return (firstLevelIndex << 4) | secondLevelIndex;
+}
+*/
+
+Heap::Heap()
+  : m_minSize(0)
+  , m_usedSize(0)
+{
+}
+
+Heap::~Heap()
+{
+}
+
+bool Heap::init(size_t minSize, size_t maxSize)
+{
+  assert(maxSize >= minSize);
+  assert(m_usedSize == 0);
+
+  m_minSize = sizeof(TLSFIndex) + 16;
+  m_minSize = std::max(m_minSize, minSize);
+  m_minSize = KRAKEN_MEM_ROUND_UP_PAGE(m_minSize);
+  if (!m_region.init(maxSize)) {
+    return false;
+  }
+  if (!m_region.resize(m_minSize)) {
+    return false;
+  }
+  reset();
+  return true;
+}
+
+// Get the actual maximum size.
+// This may be greater than the maxSize passed into init, due to page size alignment.
+size_t Heap::getMaxSize() const
+{
+  return m_region.getSize();
+}
+
+// Get the actual used size.
+// This may differ from the sum of allocations due to alignment requirements.
+size_t Heap::getUsed() const
+{
+  return m_usedSize;
+}
+
+size_t Heap::getBlockUsableSize(const TLSFBlock* block) const
+{
+  return (block->size & ~0b11) - 16ULL;
+}
+
+TLSFBlock* Heap::getNextPhysBlock(TLSFBlock* block)
+{
+  if (isBlockLast(block)) {
+    return nullptr;
+  }
+
+  return (TLSFBlock*)((std::byte*)block + block->size);
+}
+
+// Return true if block is the last block in physical order
+bool Heap::isBlockLast(const TLSFBlock* block) const
+{
+  return (block->size & 0b10) == 0b10;
+}
+
+// Returns true if block is a free block
+bool Heap::isBlockFree(const TLSFBlock* block) const
+{
+  return (block->size & 0b01) == 0b01;
+}
+
+void Heap::insertFreeBlock(TLSFBlock* block)
+{
+  // Add the block to the index
+
+  TLSFIndex* index = (TLSFIndex*)m_region.getAddress();
+
+  size_t usableSize = getBlockUsableSize(block);
+  size_t firstLevel = std::bit_width(usableSize);
+  size_t firstLevelIndex = firstLevel - 5;
+  size_t secondLevelIndex = (usableSize >> firstLevelIndex) & 0b1111ULL;
+
+  // Update free bitmaps
+  index->firstLevelFreeBitMap |= std::bit_floor(usableSize);
+  index->secondLevelFreeBitMap[firstLevelIndex] |= secondLevelIndex;
+
+  TLSFBlock* prevFirstFreeBlock = index->secondLevelFreeBlocks[firstLevelIndex][secondLevelIndex];
+  index->secondLevelFreeBlocks[firstLevelIndex][secondLevelIndex] = block;
+
+  block->prevFree = nullptr;
+  if (prevFirstFreeBlock) {
+    block->nextFree = prevFirstFreeBlock;
+    prevFirstFreeBlock->prevFree = block;
+  }
+
+  // Mark block as free
+  block->size |= 0b01; // T=1: Last Block, F=1: Free Block
+}
+
+void Heap::removeFreeBlock(TLSFBlock* block)
+{
+  // Remove the block from the index
+  TLSFIndex* index = (TLSFIndex*)m_region.getAddress();
+
+  size_t usableSize = getBlockUsableSize(block);
+  size_t firstLevel = std::bit_width(usableSize);
+  size_t firstLevelIndex = firstLevel - 5;
+  size_t secondLevelIndex = (usableSize >> (firstLevelIndex)) & 0b1111ULL;
+
+  if (block->prevFree != nullptr) {
+    // Link the neighboring free blocks together
+    block->prevFree->nextFree = block->nextFree;
+    if (block->nextFree) {
+      block->nextFree->prevFree = block->prevFree;
+    }
+  } else {
+    // This block was the first for this level
+    index->secondLevelFreeBlocks[firstLevelIndex][secondLevelIndex] = block->nextFree;
+
+    if (block->nextFree != nullptr) {
+      block->nextFree->prevFree = nullptr;
+    } else {
+      // We have removed all the free blocks at this level.
+
+      // Update second level bitmask.
+      index->secondLevelFreeBitMap[firstLevelIndex] &= ~(1ULL << secondLevelIndex);
+
+      // Check if there are any remaining free blocks within the first level
+      if (index->secondLevelFreeBitMap[firstLevelIndex] == 0) {
+
+        // This was the last one.  Clear the first level bit as well.
+        index->firstLevelFreeBitMap &= ~(1ULL << firstLevel);
+      }
+    }
+  }
+
+  // Mark block as non-free
+  block->size &= ~0b01; // T=1: Last Block, F=1: Free Block
+}
+
+// Find a free block that can hold at least size bytes
+TLSFBlock* Heap::findFreeBlock(size_t size) const
+{
+  if (size < kMinBlockSize) {
+    size = kMinBlockSize;
+  }
+  TLSFIndex* index = (TLSFIndex*)m_region.getAddress();
+
+  // size:           0b0000000000010000 (16)
+  // bit_ceil(size): 0b0000000000010000
+  // - 1 ...         0b0000000000001111
+  // ~ ...           0b1111111111110000
+
+  // size:           0b0000001110011101 (925)
+  // bit_ceil(size): 0b0000010000000000
+  // - 1 ...         0b0000001111111111
+  // ~ ...           0b1111110000000000
+
+  uint64_t possibleFirstLevels = ~(std::bit_floor(size) - 1);
+  uint64_t freeFirstLevels = index->firstLevelFreeBitMap & possibleFirstLevels;
+  uint64_t selectedFirstLevel = std::countr_zero(freeFirstLevels);
+  if (selectedFirstLevel == 64) {
+    // No first level with free blocks of sufficient size.
+    return nullptr;
+  }
+
+  uint64_t firstLevelBufferMinSize = 1ULL << selectedFirstLevel;
+  uint64_t secondLevelMask = 0b1111ULL;
+  if (size > firstLevelBufferMinSize) {
+    uint64_t delta = size - firstLevelBufferMinSize;
+    secondLevelMask = (delta >> (selectedFirstLevel - 4)) & 0b1111ULL;
+  }
+  uint64_t possibleSecondLevels = secondLevelMask & index->secondLevelFreeBitMap[selectedFirstLevel - 4];
+  uint64_t selectedSecondLevel = std::countr_zero(possibleSecondLevels);
+  if (selectedSecondLevel == 64) {
+    // No free blocks available at this first level.
+
+    // Select the next highest available first level.
+    selectedFirstLevel = std::countr_zero(freeFirstLevels & ~(1ULL << selectedFirstLevel));
+    if (selectedFirstLevel == 64) {
+      // No other first level with free blocks of sufficient size.
+      return nullptr;
+    }
+
+    // Any buffer at the second level will fit this allocation. Pick the smallest buffer.
+    selectedSecondLevel = std::countr_zero(index->secondLevelFreeBitMap[selectedFirstLevel - 4]);
+  }
+
+  // Take the first free block
+  TLSFBlock* block = index->secondLevelFreeBlocks[selectedFirstLevel - 4][selectedSecondLevel];
+
+  return block;
+}
+
+// Allocate `size` bytes
+std::byte* Heap::alloc(size_t size)
+{
+  if (size < kMinBlockSize) {
+    size = kMinBlockSize;
+  }
+
+  TLSFBlock* block = findFreeBlock(size);
+  if (block == nullptr) {
+    // No free block found
+    return nullptr;
+  }
+
+  // Ensure that we can commit pages to back the new block
+  // as well as a trailing free block header
+  size_t neededCommit = ((std::byte*)block - m_region.getAddress()) + size + 16ULL + 32ULL;
+  if (!m_region.maybeGrow(std::max(m_minSize, neededCommit))) {
+    // Out of system memory
+    return nullptr;
+  }
+
+  removeFreeBlock(block);
+
+  size_t usableSize = getBlockUsableSize(block);
+
+  if (size - usableSize >= sizeof(TLSFBlock)) {
+    // Shrink the block to size; allocate a new free block for the remainder
+
+    TLSFBlock* freeBlock = (TLSFBlock*)((std::byte*)block + size + 16ULL);
+    freeBlock->size = size - usableSize;
+    freeBlock->prevPhys = block;
+    freeBlock->prevFree = nullptr;
+    freeBlock->nextFree = nullptr;
+    if (isBlockLast(block))
+    { // LSB: T, F: T = Last physical block, F = Free block
+      // Block was the last physical block
+      block->size &= ~0b10;
+      freeBlock->size &= 0b10;
+    }
+    block->size -= freeBlock->size & ~0b11;
+    insertFreeBlock(freeBlock);
+  }
+
+  m_usedSize += getBlockUsableSize(block);
+  return (std::byte*)block + 16;
+}
+
+// Allocate `size` bytes, aligned to 16 bytes and padded to next 16-byte offset.
+std::byte* Heap::allocA16(size_t size)
+{
+  return nullptr; // not implemented
+}
+
+// Allocate `size` bytes, aligned to 64 bytes and padded to next 64-byte offset.
+std::byte* Heap::allocA64(size_t size)
+{
+  return nullptr; // not implemented
+}
+
+// Free the allocation at `address`
+void Heap::free(std::byte* address)
+{
+  TLSFBlock* block = (TLSFBlock*)(address - 16ULL);
+  m_usedSize -= getBlockUsableSize(block);
+
+  TLSFBlock* prevPhys = block->prevPhys;
+  if (prevPhys && isBlockFree(prevPhys)) {
+    // The previous block in physical order is a free block.
+
+    // Remove it from the index
+    removeFreeBlock(prevPhys);
+
+    // Resize it to glob together with the block to free
+    prevPhys->size += (block->size & ~0b11);
+    block = prevPhys;
+  }
+
+  // Reset the free block flag and the last block flag
+  block->size &= ~0b11;
+  
+  TLSFBlock* nextPhys = getNextPhysBlock(block);  
+  if (nextPhys == nullptr) {
+    // This is the last block
+    block->size &= 0b10;
+  } else if (isBlockFree(nextPhys)) {
+    // The next block in physical order is a free block.
+
+    // remove it from the index
+    removeFreeBlock(nextPhys);
+
+    // Resize free'd block to glob together with the nextPhys block
+    block->size += (nextPhys->size & ~0b11);
+    if (isBlockLast(nextPhys)) {
+      block->size &= 0b10;
+    }
+  }
+
+  // Add the free'd block to the index
+  insertFreeBlock(block);
+
+  if (isBlockLast(block)) {
+    // Maybe free pages
+    size_t neededSize = ((std::byte*)block - m_region.getAddress()) + 32ULL;
+    m_region.maybeShrink(neededSize);
+  }
+}
+
+// Reset (or initialize) the heap, freeing all memory and potentially releasing comitted pages.
+void Heap::reset()
+{
+  // Clear the index, representing no free blocks
+  TLSFIndex* index = (TLSFIndex*)m_region.getAddress();
+  memset(index, 0, sizeof(TLSFIndex));
+  m_usedSize = 0;
+
+  // Free comitted pages except minimum required
+  m_region.resize(m_minSize);
+
+  // Start with one free block, filling the entire Region
+  TLSFBlock* block = (TLSFBlock*)(m_region.getAddress() + sizeof(TLSFIndex));
+  block->prevPhys = nullptr;
+  block->nextFree = nullptr;
+  block->prevFree = nullptr;
+  block->size = m_region.getMaxSize();
+  block->size |= 0b10; // T=1: Last Block, F=1: Free Block
+
+  insertFreeBlock(block);
+}
+
+} // namespace mimir

src/mimir.cpp

@@ -39,7 +39,6 @@
 #include <fcntl.h>
 #endif
 
-#include <errno.h>
 #include <cassert>
 #include <cstring>
 

src/mimir_impl.h

@@ -31,8 +31,17 @@
 
 #pragma once
 
+#if defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+#include <unistd.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#endif
+
 #if defined(_WIN32) || defined(_WIN64)
+#define NOMINMAX
 #include <Windows.h>
+#include <memoryapi.h>
 #endif
 
 namespace mimir {

src/region.cpp

@@ -0,0 +1,185 @@
+//
+//  region.cpp
+//  Kraken Engine
+//
+//  Copyright 2026 Kearwood Gilbert. All rights reserved.
+//
+//  Redistribution and use in source and binary forms, with or without modification, are
+//  permitted provided that the following conditions are met:
+//
+//  1. Redistributions of source code must retain the above copyright notice, this list of
+//  conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright notice, this list
+//  of conditions and the following disclaimer in the documentation and/or other materials
+//  provided with the distribution.
+//
+//  THIS SOFTWARE IS PROVIDED BY KEARWOOD GILBERT ''AS IS'' AND ANY EXPRESS OR IMPLIED
+//  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+//  FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL KEARWOOD GILBERT OR
+//  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+//  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+//  SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+//  ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+//  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+//  ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+//  The views and conclusions contained in the software and documentation are those of the
+//  authors and should not be interpreted as representing official policies, either expressed
+//  or implied, of Kearwood Gilbert.
+//
+
+#include "../include/mimir.h"
+#include "mimir_impl.h"
+
+#include <cassert>
+#include <cstring>
+#include <bit>
+
+namespace mimir {
+
+Region::Region()
+  : m_data(nullptr)
+  , m_committedSize(0)
+  , m_maxSize(0)
+{
+
+}
+
+Region::~Region()
+{
+  if (m_data) {
+#if defined(_WIN32) || defined(_WIN64)
+    if (!VirtualFree((void*)m_data, 0, MEM_RELEASE)) {
+      ReportWindowsLastError("VirtualAlloc");
+    }
+#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+    munmap(m_data, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize));
+#else
+    static_assert(false, "Not Implemented");
+#endif
+  }
+}
+
+bool Region::init(size_t maxSize)
+{
+  assert(m_data == nullptr);
+  assert(m_committedSize == 0);
+
+  m_maxSize = KRAKEN_MEM_ROUND_UP_PAGE(maxSize);
+
+#if defined(_WIN32) || defined(_WIN64)
+  m_data = (std::byte*)VirtualAlloc(NULL, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize), MEM_RESERVE, PAGE_NOACCESS);
+  if (m_data == nullptr) {
+    ReportWindowsLastError("VirtualAlloc");
+    return false;
+  }
+#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+  m_data = (std::byte*)mmap(NULL, KRAKEN_MEM_ROUND_UP_PAGE(m_maxSize), PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  if (m_data == MAP_FAILED) {
+    return false;
+  }
+#else
+  static_assert(false, "Not Implemented");
+#endif
+  return true;
+}
+ 
+bool Region::resize(size_t size)
+{
+  size_t newSize = KRAKEN_MEM_ROUND_UP_PAGE(size);
+
+  if (newSize == m_committedSize) {
+    return true;
+  }
+
+  if (newSize > m_maxSize) {
+    return false;
+  }
+
+  if (newSize > m_committedSize) {
+    // ---- Growing ----
+#if defined(_WIN32) || defined(_WIN64)
+    if (VirtualAlloc(m_data + m_committedSize, newSize - m_committedSize, MEM_COMMIT, PAGE_READWRITE) == nullptr) {
+      ReportWindowsLastError("VirtualAlloc");
+      return false;
+    }
+#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+    if (mprotect(m_data + m_committedSize, newSize - m_committedSize, PROT_READ | PROT_WRITE) != 0) {
+      return false;
+    }
+#else
+    static_assert(false, "Not Implemented");
+#endif
+    m_committedSize = newSize;
+    return true;
+  } // if (newSize > m_committedSize)
+
+  // ---- Shrinking ----
+  std::byte* rangeStart = m_data + newSize;
+  size_t rangeLen = m_committedSize - newSize;
+#if defined(_WIN32) || defined(_WIN64)
+  if (!VirtualAlloc(rangeStart, rangeLen, MEM_RESET, PAGE_NOACCESS)) {
+    ReportWindowsLastError("VirtualAlloc");
+  } else {
+    m_committedSize = newSize;
+  }
+#elif defined(__unix__) || defined(__APPLE__) || defined(ANDROID)
+  if (madvise(rangeStart, rangeLen, MADV_DONTNEED) != 0) {
+    // TODO - Log Error or debug build assert
+  }
+  if (mprotect(rangeStart, rangeLen, PROT_NONE) != 0) {
+    // TODO - Log Error or debug build assert
+  }
+#else
+  static_assert(false, "Not Implemented");
+#endif
+  m_committedSize = newSize;
+
+  return true;
+}
+
+size_t Region::getSize() const
+{
+  return m_committedSize;
+}
+
+size_t Region::getMaxSize() const
+{
+  return m_maxSize;
+}
+
+std::byte* Region::getAddress() const
+{
+  return m_data;
+}
+
+bool Region::maybeGrow(size_t minSize)
+{
+  if (minSize < m_committedSize) {
+    // Already big enough
+    return true;
+  }
+
+  if (minSize > m_maxSize) {
+    // We can't grow this big
+    return false;
+  }
+
+  // We need to grow.
+  // Increase the size exponentially, capped to m_maxSize
+  size_t newSize = std::bit_ceil(minSize) << 1;
+  newSize = std::min(m_maxSize, newSize);
+  
+  return resize(newSize);
+}
+
+void Region::maybeShrink(size_t targetSize)
+{
+  size_t thresholdSize = targetSize << 1; // The threshold for shrinking is greater than the target to implement hysteresis
+  if (getSize() > thresholdSize) {
+    resize(targetSize);
+  }
+}
+
+} // namespace mimir

tests/CMakeLists.txt

@@ -0,0 +1,30 @@
+# ----====---- GoogleTest ----====---
+include(FetchContent)
+FetchContent_Declare(
+  googletest
+  URL https://github.com/google/googletest/archive/03597a01ee50ed33e9dfd640b249b4be3799d395.zip
+)
+# For Windows: Prevent overriding the parent project's compiler/linker settings
+set(gtest_force_shared_crt ON CACHE BOOL "" FORCE)
+FetchContent_MakeAvailable(googletest)
+
+enable_testing()
+
+# ----====---- End: GoogleTest ----====---
+
+add_executable(
+  mimir_test
+  "arena_test.cpp"
+  "heap_test.cpp"
+  "region_test.cpp"
+  "test_util.cpp")
+target_link_libraries(
+  mimir_test
+  GTest::gtest_main
+  mimir
+)
+
+target_include_directories(mimir_test PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../include)
+
+include(GoogleTest)
+gtest_discover_tests(mimir_test)

tests/arena_test.cpp

@@ -0,0 +1,171 @@
+#include <gtest/gtest.h>
+#include <random>
+#include "mimir.h"
+
+#include "test_util.h"
+
+namespace mimir_tests {
+
+TEST(Arena, InitialState)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init());
+  EXPECT_EQ(arena.getUsed(), 0);
+}
+
+TEST(Arena, OutOfMemory)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init(1ULL << 24, 1ULL << 46));
+  EXPECT_TRUE(arena.alloc(1ULL << 49) == nullptr);
+}
+
+TEST(Arena, Overflow)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init(128, 1024));
+  EXPECT_TRUE(arena.alloc(512) != nullptr);
+  EXPECT_TRUE(arena.alloc(512) != nullptr);
+  EXPECT_TRUE(arena.alloc(arena.getMaxSize() - 1024 + 1) == nullptr);
+}
+
+TEST(Arena, Alignment16)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init());
+
+  for (int i = 0; i < 128; i++) {
+    for (int j = 1; j < 128; j++) {
+      EXPECT_TRUE(arena.alloc(i) != nullptr);
+      for (int k = 0; k < 16; k++) {
+        std::byte* addr = arena.allocA16(j);
+        EXPECT_TRUE(addr != nullptr);
+        EXPECT_EQ((size_t)addr & 0b1111, 0);
+        EXPECT_EQ(arena.getUsed() & 0b1111, 0);
+      }
+      arena.reset();
+    }
+  }
+}
+
+TEST(Arena, Alignment64)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init());
+
+  for (int i = 0; i < 128; i++) {
+    for (int j = 1; j < 128; j++) {
+      EXPECT_TRUE(arena.alloc(i) != nullptr);
+      for (int k = 0; k < 16; k++) {
+        std::byte* addr = arena.allocA64(j);
+        EXPECT_TRUE(addr != nullptr);
+        EXPECT_EQ((size_t)addr & 0b111111, 0);
+        EXPECT_EQ(arena.getUsed() & 0b111111, 0);
+      }
+      arena.reset();
+    }
+  }
+}
+
+TEST(Arena, OverflowByAlignmentA16)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init(128, 1024));
+
+  // Controlled test - first try with no alignment. This should pass.
+  EXPECT_TRUE(arena.alloc(alignDown(arena.getMaxSize(), 16) - 15) != nullptr);
+  EXPECT_TRUE(arena.alloc(arena.getMaxSize() - arena.getUsed()) != nullptr);
+  arena.reset();
+
+  // Now try with allocA16. This should fail.
+  EXPECT_TRUE(arena.alloc(alignDown(arena.getMaxSize(), 16) - 15) != nullptr);
+  EXPECT_TRUE(arena.allocA16(arena.getMaxSize() - arena.getUsed()) == nullptr);
+}
+
+TEST(Arena, OverflowByAlignmentA64)
+{
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init(128, 1024));
+
+  // Controlled test - first try with no alignment. This should pass.
+  EXPECT_TRUE(arena.alloc(alignDown(arena.getMaxSize(), 64) - 63) != nullptr);
+  EXPECT_TRUE(arena.alloc(arena.getMaxSize() - arena.getUsed()) != nullptr);
+  arena.reset();
+
+  // Now try with allocA64. This should fail.
+  EXPECT_TRUE(arena.alloc(alignDown(arena.getMaxSize(), 64) - 63) != nullptr);
+  EXPECT_TRUE(arena.allocA64(arena.getMaxSize() - arena.getUsed()) == nullptr);
+}
+
+TEST(Arena, FillAndReset)
+{
+  // We purposefully want to have the same random sequence each time
+
+  const int maxAllocCount = 128;
+  const int cycleCount = 128;
+  std::mt19937_64 rgen(1234);
+  std::uniform_int_distribution<int> count_dist(0, maxAllocCount);
+  std::uniform_int_distribution<size_t> alloc_dist1(1, 256);
+  std::uniform_int_distribution<size_t> alloc_dist2(1, 1024 * 256);
+  std::uniform_int_distribution<size_t> alloc_dist3(1, 1024 * 1024 * 256);
+
+  mimir::init();
+  mimir::Arena arena;
+  EXPECT_TRUE(arena.init(0, 1ULL << 32)); // Use up to 4GB
+  size_t allocSize[maxAllocCount];
+  std::byte* allocAddr[maxAllocCount];
+
+  for (int i = 0; i < cycleCount; i++) {
+    int allocCount = 0;
+    if (i != 5) {
+      // Make sure at least one cycle has 0 allocations
+      allocCount = count_dist(rgen);
+    }
+    for (int j = 0; j < allocCount; j++) {
+      allocSize[j] = alloc_dist1(rgen);
+    }
+    for (int j = 0; j < 32; j++) {
+      int k = count_dist(rgen);
+      if (k < allocCount) {
+        allocSize[k] = alloc_dist2(rgen);
+      }
+    }
+    
+    for (int j = 0; j < 3; j++) {
+      int k = count_dist(rgen);
+      if (k < allocCount) {
+        allocSize[k] = alloc_dist3(rgen);
+      }
+    }
+
+    size_t totalSize = 0;
+    for (int j = 0; j < allocCount; j++) {
+      totalSize += allocSize[j];
+      allocAddr[j] = arena.alloc(allocSize[j]);
+      EXPECT_TRUE(allocAddr[j] != nullptr);
+      if (allocAddr[j] != nullptr) {
+        writeData(allocAddr[j], allocSize[j], j);
+      }
+    }
+
+    for (int j = 0; j < allocCount; j++) {
+      if (allocAddr[j] != nullptr) {
+        EXPECT_TRUE(checkData(allocAddr[j], allocSize[j], j));
+      }
+    }
+
+    EXPECT_EQ(arena.getUsed(), totalSize);
+
+    arena.reset();
+    EXPECT_EQ(arena.getUsed(), 0);
+  }
+}
+
+}; // namespace mimir_tests

tests/heap_test.cpp

@@ -0,0 +1,23 @@
+#include <gtest/gtest.h>
+#include "mimir.h"
+
+#include "test_util.h"
+
+namespace mimir_tests {
+
+TEST(Heap, InitialState)
+{
+  mimir::init();
+  mimir::Heap heap;
+  EXPECT_EQ(heap.getUsed(), 0);
+}
+
+TEST(Heap, OutOfMemory)
+{
+  mimir::init();
+  mimir::Heap heap;
+  EXPECT_TRUE(heap.init(1ULL << 24, 1ULL << 46));
+  EXPECT_TRUE(heap.alloc(1ULL << 49) == nullptr);
+}
+
+}; // namespace mimir_tests

tests/region_test.cpp

@@ -0,0 +1,86 @@
+#include <gtest/gtest.h>
+#include "mimir.h"
+
+#include "test_util.h"
+
+namespace mimir_tests {
+
+TEST(Region, InitialState)
+{
+  mimir::init();
+  mimir::Region region;
+  EXPECT_TRUE(region.getAddress() == nullptr);
+  EXPECT_TRUE(region.getMaxSize() == 0);
+  EXPECT_TRUE(region.getSize() == 0);
+}
+
+TEST(Region, TooBig)
+{
+  mimir::init();
+  mimir::Region region;
+  EXPECT_FALSE(region.init(1ULL << 50)); // Attempt to reserve too many pages
+}
+
+TEST(Region, OutOfMemory)
+{
+  mimir::init();
+  mimir::Region region;
+  EXPECT_TRUE(region.init(1ULL << 46)); // Attempt to reserve 64TB of pages
+  EXPECT_FALSE(region.resize(1ULL << 60)); // Attempt to commit an extremely large number of pages
+}
+
+TEST(Region, Min1GB)
+{
+  mimir::init();
+  mimir::Region region;
+  EXPECT_TRUE(region.init(1ULL << 30)); // Allocate 1GB of pages
+}
+
+TEST(Region, Resizing)
+{
+  mimir::init();
+  mimir::Region region;
+
+  EXPECT_TRUE(region.init(1 << 10ULL));
+  EXPECT_NE(region.getAddress(), nullptr);
+  EXPECT_GT(region.getMaxSize(), 0);
+  EXPECT_EQ(region.getSize(), 0);
+  EXPECT_TRUE(region.resize(0));
+  EXPECT_EQ(region.getSize(), 0);
+
+  EXPECT_TRUE(region.resize(1ULL << 4));
+  EXPECT_TRUE(region.resize(1ULL << 10));
+  EXPECT_TRUE(region.resize(1ULL << 5));
+  EXPECT_TRUE(region.resize(1ULL << 0));
+  EXPECT_TRUE(region.resize(1ULL << 9));
+}
+
+TEST(Region, Oversized)
+{
+  mimir::init();
+  mimir::Region region;
+
+  EXPECT_TRUE(region.init(1ULL << 4));
+  size_t prevSize = region.getSize();
+
+  EXPECT_FALSE(region.resize(1ULL << 20));
+  EXPECT_EQ(region.getSize(), prevSize);
+}
+
+TEST(Region, WriteData)
+{
+  mimir::init();
+  mimir::Region region;
+
+  EXPECT_TRUE(region.init());
+  EXPECT_TRUE(region.resize(100));
+  EXPECT_GE(region.getSize(), 100);
+  writeData(region.getAddress(), 100, 1);
+  EXPECT_TRUE(checkData(region.getAddress(), 100, 1));
+  EXPECT_TRUE(region.resize(1 << 9ULL));
+  writeData(region.getAddress() + 100, 900, 2);
+  EXPECT_TRUE(checkData(region.getAddress(), 100, 1));
+  EXPECT_TRUE(checkData(region.getAddress() + 100, 900, 2));
+}
+
+}; // namespace mimir_tests

tests/test_util.cpp

@@ -0,0 +1,50 @@
+#include "test_util.h"
+
+#include <algorithm>
+#include <cstring>
+
+namespace mimir_tests {
+
+const size_t kPatternLen = 32;
+
+void getPattern(std::byte* pattern, int seed)
+{
+  for (int i = 0; i < kPatternLen; i++) {
+    pattern[i] = std::byte((seed * 3162 + i * 63163) % 0xff);
+  }
+}
+
+void writeData(std::byte* dest, size_t count, int seed)
+{
+  std::byte pattern[kPatternLen];
+  getPattern(pattern, seed);
+
+  for (int i = 0; i < count; i += kPatternLen) {
+    memcpy(dest + i, pattern, std::min(count - i, kPatternLen));
+  }
+}
+
+bool checkData(std::byte* src, size_t count, int seed)
+{
+  std::byte pattern[kPatternLen];
+  getPattern(pattern, seed);
+
+  for (int i = 0; i < count; i += kPatternLen) {
+    if (memcmp(src + i, pattern, std::min(count - i, kPatternLen)) != 0) {
+      return false;
+    }
+  }
+  return true;
+}
+
+size_t alignUp(size_t val, size_t alignment)
+{
+  return alignDown(val - 1, alignment) + alignment;
+}
+
+size_t alignDown(size_t val, size_t alignment)
+{
+  return val & ~(alignment - 1);
+}
+
+}; // namespace mimir_tests

tests/test_util.h

@@ -0,0 +1,12 @@
+#pragma once
+
+#include <cstddef>
+
+namespace mimir_tests {
+
+void writeData(std::byte* dest, size_t count, int seed);
+bool checkData(std::byte* src, size_t count, int seed);
+size_t alignUp(size_t val, size_t alignment);
+size_t alignDown(size_t val, size_t alignment);
+
+}; // namespace mimir_tests