diff options
Diffstat (limited to 'src/3rdparty')
20 files changed, 23810 insertions, 9 deletions
diff --git a/src/3rdparty/VulkanMemoryAllocator.pri b/src/3rdparty/VulkanMemoryAllocator.pri new file mode 100644 index 0000000000..7466200dfc --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator.pri @@ -0,0 +1 @@ +INCLUDEPATH += $$PWD/VulkanMemoryAllocator diff --git a/src/3rdparty/VulkanMemoryAllocator/LICENSE.txt b/src/3rdparty/VulkanMemoryAllocator/LICENSE.txt new file mode 100644 index 0000000000..dbfe253391 --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/LICENSE.txt @@ -0,0 +1,19 @@ +Copyright (c) 2017-2018 Advanced Micro Devices, Inc. All rights reserved. + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. diff --git a/src/3rdparty/VulkanMemoryAllocator/patches/0001-Avoid-compiler-warnings.patch b/src/3rdparty/VulkanMemoryAllocator/patches/0001-Avoid-compiler-warnings.patch new file mode 100644 index 0000000000..f459db6c7a --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/patches/0001-Avoid-compiler-warnings.patch @@ -0,0 +1,402 @@ +diff --git a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +index a2f7a1b..fbe6f9e 100644 +--- a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h ++++ b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +@@ -3661,7 +3661,7 @@ static void VmaWriteMagicValue(void* pData, VkDeviceSize offset) + { + uint32_t* pDst = (uint32_t*)((char*)pData + offset); + const size_t numberCount = VMA_DEBUG_MARGIN / sizeof(uint32_t); +- for(size_t i = 0; i < numberCount; ++i, ++pDst) ++ for(size_t i = 0; i != numberCount; ++i, ++pDst) + { + *pDst = VMA_CORRUPTION_DETECTION_MAGIC_VALUE; + } +@@ -3671,7 +3671,7 @@ static bool VmaValidateMagicValue(const void* pData, VkDeviceSize offset) + { + const uint32_t* pSrc = (const uint32_t*)((const char*)pData + offset); + const size_t numberCount = VMA_DEBUG_MARGIN / sizeof(uint32_t); +- for(size_t i = 0; i < numberCount; ++i, ++pSrc) ++ for(size_t i = 0; i != numberCount; ++i, ++pSrc) + { + if(*pSrc != VMA_CORRUPTION_DETECTION_MAGIC_VALUE) + { +@@ -3866,7 +3866,7 @@ public: + template<typename U> VmaStlAllocator(const VmaStlAllocator<U>& src) : m_pCallbacks(src.m_pCallbacks) { } + + T* allocate(size_t n) { return VmaAllocateArray<T>(m_pCallbacks, n); } +- void deallocate(T* p, size_t n) { VmaFree(m_pCallbacks, p); } ++ void deallocate(T* p, size_t /*n*/) { VmaFree(m_pCallbacks, p); } + + template<typename U> + bool operator==(const VmaStlAllocator<U>& rhs) const +@@ -5214,7 +5214,7 @@ public: + virtual void FreeAtOffset(VkDeviceSize offset) = 0; + + // Tries to resize (grow or shrink) space for given allocation, in place. +- virtual bool ResizeAllocation(const VmaAllocation alloc, VkDeviceSize newSize) { return false; } ++ virtual bool ResizeAllocation(const VmaAllocation /*alloc*/, VkDeviceSize /*newSize*/) { return false; } + + protected: + const VkAllocationCallbacks* GetAllocationCallbacks() const { return m_pAllocationCallbacks; } +@@ -5574,7 +5574,7 @@ public: + + virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); + +- virtual VkResult CheckCorruption(const void* pBlockData) { return VK_ERROR_FEATURE_NOT_PRESENT; } ++ virtual VkResult CheckCorruption(const void* /*pBlockData*/) { return VK_ERROR_FEATURE_NOT_PRESENT; } + + virtual void Alloc( + const VmaAllocationRequest& request, +@@ -6133,7 +6133,7 @@ public: + bool overlappingMoveSupported); + virtual ~VmaDefragmentationAlgorithm_Fast(); + +- virtual void AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged) { ++m_AllocationCount; } ++ virtual void AddAllocation(VmaAllocation /*hAlloc*/, VkBool32* /*pChanged*/) { ++m_AllocationCount; } + virtual void AddAll() { m_AllAllocations = true; } + + virtual VkResult Defragment( +@@ -6318,7 +6318,7 @@ private: + // Redundant, for convenience not to fetch from m_hCustomPool->m_BlockVector or m_hAllocator->m_pBlockVectors. + VmaBlockVector* const m_pBlockVector; + const uint32_t m_CurrFrameIndex; +- const uint32_t m_AlgorithmFlags; ++ /*const uint32_t m_AlgorithmFlags;*/ + // Owner of this object. + VmaDefragmentationAlgorithm* m_pAlgorithm; + +@@ -7073,6 +7073,7 @@ void VmaJsonWriter::BeginValue(bool isString) + if(currItem.type == COLLECTION_TYPE_OBJECT && + currItem.valueCount % 2 == 0) + { ++ (void) isString; + VMA_ASSERT(isString); + } + +@@ -7660,7 +7661,9 @@ bool VmaBlockMetadata_Generic::Validate() const + } + + // Margin required between allocations - every free space must be at least that large. ++#if VMA_DEBUG_MARGIN + VMA_VALIDATE(subAlloc.size >= VMA_DEBUG_MARGIN); ++#endif + } + else + { +@@ -7806,6 +7809,7 @@ bool VmaBlockMetadata_Generic::CreateAllocationRequest( + { + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(!upperAddress); ++ (void) upperAddress; + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(pAllocationRequest != VMA_NULL); + VMA_HEAVY_ASSERT(Validate()); +@@ -8033,6 +8037,7 @@ void VmaBlockMetadata_Generic::Alloc( + VmaAllocation hAllocation) + { + VMA_ASSERT(!upperAddress); ++ (void) upperAddress; + VMA_ASSERT(request.item != m_Suballocations.end()); + VmaSuballocation& suballoc = *request.item; + // Given suballocation is a free block. +@@ -9609,7 +9614,7 @@ bool VmaBlockMetadata_Linear::CreateAllocationRequest( + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, +- uint32_t strategy, ++ uint32_t /*strategy*/, + VmaAllocationRequest* pAllocationRequest) + { + VMA_ASSERT(allocSize > 0); +@@ -9651,10 +9656,12 @@ bool VmaBlockMetadata_Linear::CreateAllocationRequest( + // Apply VMA_DEBUG_MARGIN at the end. + if(VMA_DEBUG_MARGIN > 0) + { ++#if VMA_DEBUG_MARGIN + if(resultOffset < VMA_DEBUG_MARGIN) + { + return false; + } ++#endif + resultOffset -= VMA_DEBUG_MARGIN; + } + +@@ -10542,18 +10549,19 @@ void VmaBlockMetadata_Buddy::PrintDetailedMap(class VmaJsonWriter& json) const + #endif // #if VMA_STATS_STRING_ENABLED + + bool VmaBlockMetadata_Buddy::CreateAllocationRequest( +- uint32_t currentFrameIndex, +- uint32_t frameInUseCount, ++ uint32_t /*currentFrameIndex*/, ++ uint32_t /*frameInUseCount*/, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, +- bool canMakeOtherLost, +- uint32_t strategy, ++ bool /*canMakeOtherLost*/, ++ uint32_t /*strategy*/, + VmaAllocationRequest* pAllocationRequest) + { + VMA_ASSERT(!upperAddress && "VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT can be used only with linear algorithm."); ++ (void) upperAddress; + + // Simple way to respect bufferImageGranularity. May be optimized some day. + // Whenever it might be an OPTIMAL image... +@@ -10593,8 +10601,8 @@ bool VmaBlockMetadata_Buddy::CreateAllocationRequest( + } + + bool VmaBlockMetadata_Buddy::MakeRequestedAllocationsLost( +- uint32_t currentFrameIndex, +- uint32_t frameInUseCount, ++ uint32_t /*currentFrameIndex*/, ++ uint32_t /*frameInUseCount*/, + VmaAllocationRequest* pAllocationRequest) + { + /* +@@ -10604,7 +10612,7 @@ bool VmaBlockMetadata_Buddy::MakeRequestedAllocationsLost( + return pAllocationRequest->itemsToMakeLostCount == 0; + } + +-uint32_t VmaBlockMetadata_Buddy::MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount) ++uint32_t VmaBlockMetadata_Buddy::MakeAllocationsLost(uint32_t /*currentFrameIndex*/, uint32_t /*frameInUseCount*/) + { + /* + Lost allocations are not supported in buddy allocator at the moment. +@@ -10615,9 +10623,9 @@ uint32_t VmaBlockMetadata_Buddy::MakeAllocationsLost(uint32_t currentFrameIndex, + + void VmaBlockMetadata_Buddy::Alloc( + const VmaAllocationRequest& request, +- VmaSuballocationType type, ++ VmaSuballocationType /*type*/, + VkDeviceSize allocSize, +- bool upperAddress, ++ bool /*upperAddress*/, + VmaAllocation hAllocation) + { + const uint32_t targetLevel = AllocSizeToLevel(allocSize); +@@ -10941,7 +10949,7 @@ void VmaBlockMetadata_Buddy::PrintDetailedMapNode(class VmaJsonWriter& json, con + //////////////////////////////////////////////////////////////////////////////// + // class VmaDeviceMemoryBlock + +-VmaDeviceMemoryBlock::VmaDeviceMemoryBlock(VmaAllocator hAllocator) : ++VmaDeviceMemoryBlock::VmaDeviceMemoryBlock(VmaAllocator /*hAllocator*/) : + m_pMetadata(VMA_NULL), + m_MemoryTypeIndex(UINT32_MAX), + m_Id(0), +@@ -11691,6 +11699,7 @@ VkResult VmaBlockVector::AllocatePage( + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBestRequestBlock->WriteMagicValueAroundAllocation(m_hAllocator, bestRequest.offset, size); ++ (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); + } + return VK_SUCCESS; +@@ -11729,6 +11738,7 @@ void VmaBlockVector::Free( + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBlock->ValidateMagicValueAroundAllocation(m_hAllocator, hAllocation->GetOffset(), hAllocation->GetSize()); ++ (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to validate magic value."); + } + +@@ -11894,6 +11904,7 @@ VkResult VmaBlockVector::AllocateFromBlock( + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBlock->WriteMagicValueAroundAllocation(m_hAllocator, currRequest.offset, size); ++ (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); + } + return VK_SUCCESS; +@@ -11903,7 +11914,8 @@ VkResult VmaBlockVector::AllocateFromBlock( + + VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIndex) + { +- VkMemoryAllocateInfo allocInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO }; ++ VkMemoryAllocateInfo allocInfo = {}; ++ allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + allocInfo.memoryTypeIndex = m_MemoryTypeIndex; + allocInfo.allocationSize = blockSize; + VkDeviceMemory mem = VK_NULL_HANDLE; +@@ -11991,7 +12003,8 @@ void VmaBlockVector::ApplyDefragmentationMovesCpu( + if(pDefragCtx->res == VK_SUCCESS) + { + const VkDeviceSize nonCoherentAtomSize = m_hAllocator->m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; +- VkMappedMemoryRange memRange = { VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE }; ++ VkMappedMemoryRange memRange = {}; ++ memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE; + + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { +@@ -12076,7 +12089,8 @@ void VmaBlockVector::ApplyDefragmentationMovesGpu( + + // Go over all blocks. Create and bind buffer for whole block if necessary. + { +- VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; ++ VkBufferCreateInfo bufCreateInfo = {}; ++ bufCreateInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; + bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | + VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +@@ -12101,8 +12115,9 @@ void VmaBlockVector::ApplyDefragmentationMovesGpu( + // Go over all moves. Post data transfer commands to command buffer. + if(pDefragCtx->res == VK_SUCCESS) + { +- const VkDeviceSize nonCoherentAtomSize = m_hAllocator->m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; +- VkMappedMemoryRange memRange = { VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE }; ++ /*const VkDeviceSize nonCoherentAtomSize = m_hAllocator->m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; ++ VkMappedMemoryRange memRange = {}; ++ memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;*/ + + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { +@@ -12435,10 +12450,10 @@ VmaDefragmentationAlgorithm_Generic::VmaDefragmentationAlgorithm_Generic( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex, +- bool overlappingMoveSupported) : ++ bool /*overlappingMoveSupported*/) : + VmaDefragmentationAlgorithm(hAllocator, pBlockVector, currentFrameIndex), +- m_AllAllocations(false), + m_AllocationCount(0), ++ m_AllAllocations(false), + m_BytesMoved(0), + m_AllocationsMoved(0), + m_Blocks(VmaStlAllocator<BlockInfo*>(hAllocator->GetAllocationCallbacks())) +@@ -12813,7 +12828,7 @@ VkResult VmaDefragmentationAlgorithm_Fast::Defragment( + size_t freeSpaceOrigBlockIndex = m_BlockInfos[freeSpaceInfoIndex].origBlockIndex; + VmaDeviceMemoryBlock* pFreeSpaceBlock = m_pBlockVector->GetBlock(freeSpaceOrigBlockIndex); + VmaBlockMetadata_Generic* pFreeSpaceMetadata = (VmaBlockMetadata_Generic*)pFreeSpaceBlock->m_pMetadata; +- VkDeviceSize freeSpaceBlockSize = pFreeSpaceMetadata->GetSize(); ++ /*VkDeviceSize freeSpaceBlockSize = pFreeSpaceMetadata->GetSize();*/ + + // Same block + if(freeSpaceInfoIndex == srcBlockInfoIndex) +@@ -13098,7 +13113,7 @@ VmaBlockVectorDefragmentationContext::VmaBlockVectorDefragmentationContext( + VmaPool hCustomPool, + VmaBlockVector* pBlockVector, + uint32_t currFrameIndex, +- uint32_t algorithmFlags) : ++ uint32_t /*algorithmFlags*/) : + res(VK_SUCCESS), + mutexLocked(false), + blockContexts(VmaStlAllocator<VmaBlockDefragmentationContext>(hAllocator->GetAllocationCallbacks())), +@@ -13106,7 +13121,7 @@ VmaBlockVectorDefragmentationContext::VmaBlockVectorDefragmentationContext( + m_hCustomPool(hCustomPool), + m_pBlockVector(pBlockVector), + m_CurrFrameIndex(currFrameIndex), +- m_AlgorithmFlags(algorithmFlags), ++ /*m_AlgorithmFlags(algorithmFlags),*/ + m_pAlgorithm(VMA_NULL), + m_Allocations(VmaStlAllocator<AllocInfo>(hAllocator->GetAllocationCallbacks())), + m_AllAllocations(false) +@@ -14311,19 +14326,21 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( + bool map, + bool isUserDataString, + void* pUserData, +- VkBuffer dedicatedBuffer, +- VkImage dedicatedImage, ++ VkBuffer /*dedicatedBuffer*/, ++ VkImage /*dedicatedImage*/, + size_t allocationCount, + VmaAllocation* pAllocations) + { + VMA_ASSERT(allocationCount > 0 && pAllocations); + +- VkMemoryAllocateInfo allocInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO }; ++ VkMemoryAllocateInfo allocInfo = {}; ++ allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + allocInfo.memoryTypeIndex = memTypeIndex; + allocInfo.allocationSize = size; + + #if VMA_DEDICATED_ALLOCATION +- VkMemoryDedicatedAllocateInfoKHR dedicatedAllocInfo = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO_KHR }; ++ VkMemoryDedicatedAllocateInfoKHR dedicatedAllocInfo = {}; ++ dedicatedAllocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO_KHR; + if(m_UseKhrDedicatedAllocation) + { + if(dedicatedBuffer != VK_NULL_HANDLE) +@@ -14341,7 +14358,7 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( + #endif // #if VMA_DEDICATED_ALLOCATION + + size_t allocIndex; +- VkResult res; ++ VkResult res = VK_SUCCESS; + for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + res = AllocateDedicatedMemoryPage( +@@ -14460,12 +14477,15 @@ void VmaAllocator_T::GetBufferMemoryRequirements( + #if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { +- VkBufferMemoryRequirementsInfo2KHR memReqInfo = { VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2_KHR }; ++ VkBufferMemoryRequirementsInfo2KHR memReqInfo = {}; ++ memReqInfo.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2_KHR; + memReqInfo.buffer = hBuffer; + +- VkMemoryDedicatedRequirementsKHR memDedicatedReq = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR }; ++ VkMemoryDedicatedRequirementsKHR memDedicatedReq = {}; ++ memDedicatedReq.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR; + +- VkMemoryRequirements2KHR memReq2 = { VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR }; ++ VkMemoryRequirements2KHR memReq2 = {}; ++ memReq2.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR; + memReq2.pNext = &memDedicatedReq; + + (*m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); +@@ -14492,12 +14512,15 @@ void VmaAllocator_T::GetImageMemoryRequirements( + #if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { +- VkImageMemoryRequirementsInfo2KHR memReqInfo = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2_KHR }; ++ VkImageMemoryRequirementsInfo2KHR memReqInfo = {}; ++ memReqInfo.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2_KHR; + memReqInfo.image = hImage; + +- VkMemoryDedicatedRequirementsKHR memDedicatedReq = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR }; ++ VkMemoryDedicatedRequirementsKHR memDedicatedReq = {}; ++ memDedicatedReq.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR; + +- VkMemoryRequirements2KHR memReq2 = { VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR }; ++ VkMemoryRequirements2KHR memReq2 = {}; ++ memReq2.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR; + memReq2.pNext = &memDedicatedReq; + + (*m_VulkanFunctions.vkGetImageMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); +@@ -14734,7 +14757,7 @@ VkResult VmaAllocator_T::ResizeAllocation( + } + else + { +- return VK_ERROR_OUT_OF_POOL_MEMORY; ++ return VkResult(-1000069000); // VK_ERROR_OUT_OF_POOL_MEMORY + } + default: + VMA_ASSERT(0); +@@ -15000,6 +15023,7 @@ void VmaAllocator_T::DestroyPool(VmaPool pool) + { + VmaMutexLockWrite lock(m_PoolsMutex, m_UseMutex); + bool success = VmaVectorRemoveSorted<VmaPointerLess>(m_Pools, pool); ++ (void) success; + VMA_ASSERT(success && "Pool not found in Allocator."); + } + +@@ -15248,7 +15272,8 @@ void VmaAllocator_T::FlushOrInvalidateAllocation( + + const VkDeviceSize nonCoherentAtomSize = m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; + +- VkMappedMemoryRange memRange = { VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE }; ++ VkMappedMemoryRange memRange = {}; ++ memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE; + memRange.memory = hAllocation->GetMemory(); + + switch(hAllocation->GetType()) +@@ -15321,6 +15346,7 @@ void VmaAllocator_T::FreeDedicatedMemory(VmaAllocation allocation) + AllocationVectorType* const pDedicatedAllocations = m_pDedicatedAllocations[memTypeIndex]; + VMA_ASSERT(pDedicatedAllocations); + bool success = VmaVectorRemoveSorted<VmaPointerLess>(*pDedicatedAllocations, allocation); ++ (void) success; + VMA_ASSERT(success); + } + diff --git a/src/3rdparty/VulkanMemoryAllocator/patches/0002-Fix-gcc8-warning.patch b/src/3rdparty/VulkanMemoryAllocator/patches/0002-Fix-gcc8-warning.patch new file mode 100644 index 0000000000..57a2f1a0f1 --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/patches/0002-Fix-gcc8-warning.patch @@ -0,0 +1,14 @@ +diff --git a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +index fbe6f9e3e8..f043bdc289 100644 +--- a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h ++++ b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +@@ -12074,7 +12074,8 @@ void VmaBlockVector::ApplyDefragmentationMovesGpu( + const size_t blockCount = m_Blocks.size(); + + pDefragCtx->blockContexts.resize(blockCount); +- memset(pDefragCtx->blockContexts.data(), 0, blockCount * sizeof(VmaBlockDefragmentationContext)); ++ for (size_t i = 0; i < blockCount; ++i) ++ pDefragCtx->blockContexts[i] = VmaBlockDefragmentationContext(); + + // Go over all moves. Mark blocks that are used with BLOCK_FLAG_USED. + const size_t moveCount = moves.size(); diff --git a/src/3rdparty/VulkanMemoryAllocator/patches/0003-Disable-srwlock-for-mingw.patch b/src/3rdparty/VulkanMemoryAllocator/patches/0003-Disable-srwlock-for-mingw.patch new file mode 100644 index 0000000000..ab7acfe40b --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/patches/0003-Disable-srwlock-for-mingw.patch @@ -0,0 +1,13 @@ +diff --git a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +index f043bdc289..2355de091f 100644 +--- a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h ++++ b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h +@@ -3298,7 +3298,7 @@ void *aligned_alloc(size_t alignment, size_t size) + std::shared_mutex m_Mutex; + }; + #define VMA_RW_MUTEX VmaRWMutex +- #elif defined(_WIN32) ++ #elif defined(_WIN32) && !defined(__MINGW32__) + // Use SRWLOCK from WinAPI. + class VmaRWMutex + { diff --git a/src/3rdparty/VulkanMemoryAllocator/qt_attribution.json b/src/3rdparty/VulkanMemoryAllocator/qt_attribution.json new file mode 100644 index 0000000000..2548856ca7 --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/qt_attribution.json @@ -0,0 +1,16 @@ +[ + { + "Id": "VulkanMemoryAllocator", + "Name": "Vulkan Memory Allocator", + "QDocModule": "qtrhi", + "Description": "Vulkan Memory Allocator", + "QtUsage": "Memory management for the Vulkan backend of QRhi.", + + "Homepage": "https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator", + "Version": "2.2.0", + "License": "MIT License", + "LicenseId": "MIT", + "LicenseFile": "LICENSE.txt", + "Copyright": "Copyright (c) 2017-2018 Advanced Micro Devices, Inc. All rights reserved." + } +] diff --git a/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h new file mode 100644 index 0000000000..2355de091f --- /dev/null +++ b/src/3rdparty/VulkanMemoryAllocator/vk_mem_alloc.h @@ -0,0 +1,16790 @@ +// +// Copyright (c) 2017-2018 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +// + +#ifndef AMD_VULKAN_MEMORY_ALLOCATOR_H +#define AMD_VULKAN_MEMORY_ALLOCATOR_H + +#ifdef __cplusplus +extern "C" { +#endif + +/** \mainpage Vulkan Memory Allocator + +<b>Version 2.2.0</b> (2018-12-13) + +Copyright (c) 2017-2018 Advanced Micro Devices, Inc. All rights reserved. \n +License: MIT + +Documentation of all members: vk_mem_alloc.h + +\section main_table_of_contents Table of contents + +- <b>User guide</b> + - \subpage quick_start + - [Project setup](@ref quick_start_project_setup) + - [Initialization](@ref quick_start_initialization) + - [Resource allocation](@ref quick_start_resource_allocation) + - \subpage choosing_memory_type + - [Usage](@ref choosing_memory_type_usage) + - [Required and preferred flags](@ref choosing_memory_type_required_preferred_flags) + - [Explicit memory types](@ref choosing_memory_type_explicit_memory_types) + - [Custom memory pools](@ref choosing_memory_type_custom_memory_pools) + - \subpage memory_mapping + - [Mapping functions](@ref memory_mapping_mapping_functions) + - [Persistently mapped memory](@ref memory_mapping_persistently_mapped_memory) + - [Cache control](@ref memory_mapping_cache_control) + - [Finding out if memory is mappable](@ref memory_mapping_finding_if_memory_mappable) + - \subpage custom_memory_pools + - [Choosing memory type index](@ref custom_memory_pools_MemTypeIndex) + - [Linear allocation algorithm](@ref linear_algorithm) + - [Free-at-once](@ref linear_algorithm_free_at_once) + - [Stack](@ref linear_algorithm_stack) + - [Double stack](@ref linear_algorithm_double_stack) + - [Ring buffer](@ref linear_algorithm_ring_buffer) + - [Buddy allocation algorithm](@ref buddy_algorithm) + - \subpage defragmentation + - [Defragmenting CPU memory](@ref defragmentation_cpu) + - [Defragmenting GPU memory](@ref defragmentation_gpu) + - [Additional notes](@ref defragmentation_additional_notes) + - [Writing custom allocation algorithm](@ref defragmentation_custom_algorithm) + - \subpage lost_allocations + - \subpage statistics + - [Numeric statistics](@ref statistics_numeric_statistics) + - [JSON dump](@ref statistics_json_dump) + - \subpage allocation_annotation + - [Allocation user data](@ref allocation_user_data) + - [Allocation names](@ref allocation_names) + - \subpage debugging_memory_usage + - [Memory initialization](@ref debugging_memory_usage_initialization) + - [Margins](@ref debugging_memory_usage_margins) + - [Corruption detection](@ref debugging_memory_usage_corruption_detection) + - \subpage record_and_replay +- \subpage usage_patterns + - [Simple patterns](@ref usage_patterns_simple) + - [Advanced patterns](@ref usage_patterns_advanced) +- \subpage configuration + - [Pointers to Vulkan functions](@ref config_Vulkan_functions) + - [Custom host memory allocator](@ref custom_memory_allocator) + - [Device memory allocation callbacks](@ref allocation_callbacks) + - [Device heap memory limit](@ref heap_memory_limit) + - \subpage vk_khr_dedicated_allocation +- \subpage general_considerations + - [Thread safety](@ref general_considerations_thread_safety) + - [Validation layer warnings](@ref general_considerations_validation_layer_warnings) + - [Allocation algorithm](@ref general_considerations_allocation_algorithm) + - [Features not supported](@ref general_considerations_features_not_supported) + +\section main_see_also See also + +- [Product page on GPUOpen](https://gpuopen.com/gaming-product/vulkan-memory-allocator/) +- [Source repository on GitHub](https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator) + + + + +\page quick_start Quick start + +\section quick_start_project_setup Project setup + +Vulkan Memory Allocator comes in form of a single header file. +You don't need to build it as a separate library project. +You can add this file directly to your project and submit it to code repository next to your other source files. + +"Single header" doesn't mean that everything is contained in C/C++ declarations, +like it tends to be in case of inline functions or C++ templates. +It means that implementation is bundled with interface in a single file and needs to be extracted using preprocessor macro. +If you don't do it properly, you will get linker errors. + +To do it properly: + +-# Include "vk_mem_alloc.h" file in each CPP file where you want to use the library. + This includes declarations of all members of the library. +-# In exacly one CPP file define following macro before this include. + It enables also internal definitions. + +\code +#define VMA_IMPLEMENTATION +#include "vk_mem_alloc.h" +\endcode + +It may be a good idea to create dedicated CPP file just for this purpose. + +Note on language: This library is written in C++, but has C-compatible interface. +Thus you can include and use vk_mem_alloc.h in C or C++ code, but full +implementation with `VMA_IMPLEMENTATION` macro must be compiled as C++, NOT as C. + +Please note that this library includes header `<vulkan/vulkan.h>`, which in turn +includes `<windows.h>` on Windows. If you need some specific macros defined +before including these headers (like `WIN32_LEAN_AND_MEAN` or +`WINVER` for Windows, `VK_USE_PLATFORM_WIN32_KHR` for Vulkan), you must define +them before every `#include` of this library. + + +\section quick_start_initialization Initialization + +At program startup: + +-# Initialize Vulkan to have `VkPhysicalDevice` and `VkDevice` object. +-# Fill VmaAllocatorCreateInfo structure and create #VmaAllocator object by + calling vmaCreateAllocator(). + +\code +VmaAllocatorCreateInfo allocatorInfo = {}; +allocatorInfo.physicalDevice = physicalDevice; +allocatorInfo.device = device; + +VmaAllocator allocator; +vmaCreateAllocator(&allocatorInfo, &allocator); +\endcode + +\section quick_start_resource_allocation Resource allocation + +When you want to create a buffer or image: + +-# Fill `VkBufferCreateInfo` / `VkImageCreateInfo` structure. +-# Fill VmaAllocationCreateInfo structure. +-# Call vmaCreateBuffer() / vmaCreateImage() to get `VkBuffer`/`VkImage` with memory + already allocated and bound to it. + +\code +VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufferInfo.size = 65536; +bufferInfo.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +Don't forget to destroy your objects when no longer needed: + +\code +vmaDestroyBuffer(allocator, buffer, allocation); +vmaDestroyAllocator(allocator); +\endcode + + +\page choosing_memory_type Choosing memory type + +Physical devices in Vulkan support various combinations of memory heaps and +types. Help with choosing correct and optimal memory type for your specific +resource is one of the key features of this library. You can use it by filling +appropriate members of VmaAllocationCreateInfo structure, as described below. +You can also combine multiple methods. + +-# If you just want to find memory type index that meets your requirements, you + can use function vmaFindMemoryTypeIndex(). +-# If you want to allocate a region of device memory without association with any + specific image or buffer, you can use function vmaAllocateMemory(). Usage of + this function is not recommended and usually not needed. +-# If you already have a buffer or an image created, you want to allocate memory + for it and then you will bind it yourself, you can use function + vmaAllocateMemoryForBuffer(), vmaAllocateMemoryForImage(). + For binding you should use functions: vmaBindBufferMemory(), vmaBindImageMemory(). +-# If you want to create a buffer or an image, allocate memory for it and bind + them together, all in one call, you can use function vmaCreateBuffer(), + vmaCreateImage(). This is the recommended way to use this library. + +When using 3. or 4., the library internally queries Vulkan for memory types +supported for that buffer or image (function `vkGetBufferMemoryRequirements()`) +and uses only one of these types. + +If no memory type can be found that meets all the requirements, these functions +return `VK_ERROR_FEATURE_NOT_PRESENT`. + +You can leave VmaAllocationCreateInfo structure completely filled with zeros. +It means no requirements are specified for memory type. +It is valid, although not very useful. + +\section choosing_memory_type_usage Usage + +The easiest way to specify memory requirements is to fill member +VmaAllocationCreateInfo::usage using one of the values of enum #VmaMemoryUsage. +It defines high level, common usage types. +For more details, see description of this enum. + +For example, if you want to create a uniform buffer that will be filled using +transfer only once or infrequently and used for rendering every frame, you can +do it using following code: + +\code +VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufferInfo.size = 65536; +bufferInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +\section choosing_memory_type_required_preferred_flags Required and preferred flags + +You can specify more detailed requirements by filling members +VmaAllocationCreateInfo::requiredFlags and VmaAllocationCreateInfo::preferredFlags +with a combination of bits from enum `VkMemoryPropertyFlags`. For example, +if you want to create a buffer that will be persistently mapped on host (so it +must be `HOST_VISIBLE`) and preferably will also be `HOST_COHERENT` and `HOST_CACHED`, +use following code: + +\code +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.requiredFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; +allocInfo.preferredFlags = VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; +allocInfo.flags = VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +A memory type is chosen that has all the required flags and as many preferred +flags set as possible. + +If you use VmaAllocationCreateInfo::usage, it is just internally converted to +a set of required and preferred flags. + +\section choosing_memory_type_explicit_memory_types Explicit memory types + +If you inspected memory types available on the physical device and you have +a preference for memory types that you want to use, you can fill member +VmaAllocationCreateInfo::memoryTypeBits. It is a bit mask, where each bit set +means that a memory type with that index is allowed to be used for the +allocation. Special value 0, just like `UINT32_MAX`, means there are no +restrictions to memory type index. + +Please note that this member is NOT just a memory type index. +Still you can use it to choose just one, specific memory type. +For example, if you already determined that your buffer should be created in +memory type 2, use following code: + +\code +uint32_t memoryTypeIndex = 2; + +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.memoryTypeBits = 1u << memoryTypeIndex; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +\section choosing_memory_type_custom_memory_pools Custom memory pools + +If you allocate from custom memory pool, all the ways of specifying memory +requirements described above are not applicable and the aforementioned members +of VmaAllocationCreateInfo structure are ignored. Memory type is selected +explicitly when creating the pool and then used to make all the allocations from +that pool. For further details, see \ref custom_memory_pools. + + +\page memory_mapping Memory mapping + +To "map memory" in Vulkan means to obtain a CPU pointer to `VkDeviceMemory`, +to be able to read from it or write to it in CPU code. +Mapping is possible only of memory allocated from a memory type that has +`VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` flag. +Functions `vkMapMemory()`, `vkUnmapMemory()` are designed for this purpose. +You can use them directly with memory allocated by this library, +but it is not recommended because of following issue: +Mapping the same `VkDeviceMemory` block multiple times is illegal - only one mapping at a time is allowed. +This includes mapping disjoint regions. Mapping is not reference-counted internally by Vulkan. +Because of this, Vulkan Memory Allocator provides following facilities: + +\section memory_mapping_mapping_functions Mapping functions + +The library provides following functions for mapping of a specific #VmaAllocation: vmaMapMemory(), vmaUnmapMemory(). +They are safer and more convenient to use than standard Vulkan functions. +You can map an allocation multiple times simultaneously - mapping is reference-counted internally. +You can also map different allocations simultaneously regardless of whether they use the same `VkDeviceMemory` block. +The way it's implemented is that the library always maps entire memory block, not just region of the allocation. +For further details, see description of vmaMapMemory() function. +Example: + +\code +// Having these objects initialized: + +struct ConstantBuffer +{ + ... +}; +ConstantBuffer constantBufferData; + +VmaAllocator allocator; +VkBuffer constantBuffer; +VmaAllocation constantBufferAllocation; + +// You can map and fill your buffer using following code: + +void* mappedData; +vmaMapMemory(allocator, constantBufferAllocation, &mappedData); +memcpy(mappedData, &constantBufferData, sizeof(constantBufferData)); +vmaUnmapMemory(allocator, constantBufferAllocation); +\endcode + +When mapping, you may see a warning from Vulkan validation layer similar to this one: + +<i>Mapping an image with layout VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL can result in undefined behavior if this memory is used by the device. Only GENERAL or PREINITIALIZED should be used.</i> + +It happens because the library maps entire `VkDeviceMemory` block, where different +types of images and buffers may end up together, especially on GPUs with unified memory like Intel. +You can safely ignore it if you are sure you access only memory of the intended +object that you wanted to map. + + +\section memory_mapping_persistently_mapped_memory Persistently mapped memory + +Kepping your memory persistently mapped is generally OK in Vulkan. +You don't need to unmap it before using its data on the GPU. +The library provides a special feature designed for that: +Allocations made with #VMA_ALLOCATION_CREATE_MAPPED_BIT flag set in +VmaAllocationCreateInfo::flags stay mapped all the time, +so you can just access CPU pointer to it any time +without a need to call any "map" or "unmap" function. +Example: + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = sizeof(ConstantBuffer); +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_CPU_ONLY; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +// Buffer is already mapped. You can access its memory. +memcpy(allocInfo.pMappedData, &constantBufferData, sizeof(constantBufferData)); +\endcode + +There are some exceptions though, when you should consider mapping memory only for a short period of time: + +- When operating system is Windows 7 or 8.x (Windows 10 is not affected because it uses WDDM2), + device is discrete AMD GPU, + and memory type is the special 256 MiB pool of `DEVICE_LOCAL + HOST_VISIBLE` memory + (selected when you use #VMA_MEMORY_USAGE_CPU_TO_GPU), + then whenever a memory block allocated from this memory type stays mapped + for the time of any call to `vkQueueSubmit()` or `vkQueuePresentKHR()`, this + block is migrated by WDDM to system RAM, which degrades performance. It doesn't + matter if that particular memory block is actually used by the command buffer + being submitted. +- On Mac/MoltenVK there is a known bug - [Issue #175](https://github.com/KhronosGroup/MoltenVK/issues/175) + which requires unmapping before GPU can see updated texture. +- Keeping many large memory blocks mapped may impact performance or stability of some debugging tools. + +\section memory_mapping_cache_control Cache control + +Memory in Vulkan doesn't need to be unmapped before using it on GPU, +but unless a memory types has `VK_MEMORY_PROPERTY_HOST_COHERENT_BIT` flag set, +you need to manually invalidate cache before reading of mapped pointer +and flush cache after writing to mapped pointer. +Vulkan provides following functions for this purpose `vkFlushMappedMemoryRanges()`, +`vkInvalidateMappedMemoryRanges()`, but this library provides more convenient +functions that refer to given allocation object: vmaFlushAllocation(), +vmaInvalidateAllocation(). + +Regions of memory specified for flush/invalidate must be aligned to +`VkPhysicalDeviceLimits::nonCoherentAtomSize`. This is automatically ensured by the library. +In any memory type that is `HOST_VISIBLE` but not `HOST_COHERENT`, all allocations +within blocks are aligned to this value, so their offsets are always multiply of +`nonCoherentAtomSize` and two different allocations never share same "line" of this size. + +Please note that memory allocated with #VMA_MEMORY_USAGE_CPU_ONLY is guaranteed to be `HOST_COHERENT`. + +Also, Windows drivers from all 3 PC GPU vendors (AMD, Intel, NVIDIA) +currently provide `HOST_COHERENT` flag on all memory types that are +`HOST_VISIBLE`, so on this platform you may not need to bother. + +\section memory_mapping_finding_if_memory_mappable Finding out if memory is mappable + +It may happen that your allocation ends up in memory that is `HOST_VISIBLE` (available for mapping) +despite it wasn't explicitly requested. +For example, application may work on integrated graphics with unified memory (like Intel) or +allocation from video memory might have failed, so the library chose system memory as fallback. + +You can detect this case and map such allocation to access its memory on CPU directly, +instead of launching a transfer operation. +In order to do that: inspect `allocInfo.memoryType`, call vmaGetMemoryTypeProperties(), +and look for `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` flag in properties of that memory type. + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = sizeof(ConstantBuffer); +bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.preferredFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +VkMemoryPropertyFlags memFlags; +vmaGetMemoryTypeProperties(allocator, allocInfo.memoryType, &memFlags); +if((memFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) +{ + // Allocation ended up in mappable memory. You can map it and access it directly. + void* mappedData; + vmaMapMemory(allocator, alloc, &mappedData); + memcpy(mappedData, &constantBufferData, sizeof(constantBufferData)); + vmaUnmapMemory(allocator, alloc); +} +else +{ + // Allocation ended up in non-mappable memory. + // You need to create CPU-side buffer in VMA_MEMORY_USAGE_CPU_ONLY and make a transfer. +} +\endcode + +You can even use #VMA_ALLOCATION_CREATE_MAPPED_BIT flag while creating allocations +that are not necessarily `HOST_VISIBLE` (e.g. using #VMA_MEMORY_USAGE_GPU_ONLY). +If the allocation ends up in memory type that is `HOST_VISIBLE`, it will be persistently mapped and you can use it directly. +If not, the flag is just ignored. +Example: + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = sizeof(ConstantBuffer); +bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +if(allocInfo.pUserData != nullptr) +{ + // Allocation ended up in mappable memory. + // It's persistently mapped. You can access it directly. + memcpy(allocInfo.pMappedData, &constantBufferData, sizeof(constantBufferData)); +} +else +{ + // Allocation ended up in non-mappable memory. + // You need to create CPU-side buffer in VMA_MEMORY_USAGE_CPU_ONLY and make a transfer. +} +\endcode + + +\page custom_memory_pools Custom memory pools + +A memory pool contains a number of `VkDeviceMemory` blocks. +The library automatically creates and manages default pool for each memory type available on the device. +Default memory pool automatically grows in size. +Size of allocated blocks is also variable and managed automatically. + +You can create custom pool and allocate memory out of it. +It can be useful if you want to: + +- Keep certain kind of allocations separate from others. +- Enforce particular, fixed size of Vulkan memory blocks. +- Limit maximum amount of Vulkan memory allocated for that pool. +- Reserve minimum or fixed amount of Vulkan memory always preallocated for that pool. + +To use custom memory pools: + +-# Fill VmaPoolCreateInfo structure. +-# Call vmaCreatePool() to obtain #VmaPool handle. +-# When making an allocation, set VmaAllocationCreateInfo::pool to this handle. + You don't need to specify any other parameters of this structure, like `usage`. + +Example: + +\code +// Create a pool that can have at most 2 blocks, 128 MiB each. +VmaPoolCreateInfo poolCreateInfo = {}; +poolCreateInfo.memoryTypeIndex = ... +poolCreateInfo.blockSize = 128ull * 1024 * 1024; +poolCreateInfo.maxBlockCount = 2; + +VmaPool pool; +vmaCreatePool(allocator, &poolCreateInfo, &pool); + +// Allocate a buffer out of it. +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 1024; +bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.pool = pool; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); +\endcode + +You have to free all allocations made from this pool before destroying it. + +\code +vmaDestroyBuffer(allocator, buf, alloc); +vmaDestroyPool(allocator, pool); +\endcode + +\section custom_memory_pools_MemTypeIndex Choosing memory type index + +When creating a pool, you must explicitly specify memory type index. +To find the one suitable for your buffers or images, you can use helper functions +vmaFindMemoryTypeIndexForBufferInfo(), vmaFindMemoryTypeIndexForImageInfo(). +You need to provide structures with example parameters of buffers or images +that you are going to create in that pool. + +\code +VkBufferCreateInfo exampleBufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +exampleBufCreateInfo.size = 1024; // Whatever. +exampleBufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; // Change if needed. + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; // Change if needed. + +uint32_t memTypeIndex; +vmaFindMemoryTypeIndexForBufferInfo(allocator, &exampleBufCreateInfo, &allocCreateInfo, &memTypeIndex); + +VmaPoolCreateInfo poolCreateInfo = {}; +poolCreateInfo.memoryTypeIndex = memTypeIndex; +// ... +\endcode + +When creating buffers/images allocated in that pool, provide following parameters: + +- `VkBufferCreateInfo`: Prefer to pass same parameters as above. + Otherwise you risk creating resources in a memory type that is not suitable for them, which may result in undefined behavior. + Using different `VK_BUFFER_USAGE_` flags may work, but you shouldn't create images in a pool intended for buffers + or the other way around. +- VmaAllocationCreateInfo: You don't need to pass same parameters. Fill only `pool` member. + Other members are ignored anyway. + +\section linear_algorithm Linear allocation algorithm + +Each Vulkan memory block managed by this library has accompanying metadata that +keeps track of used and unused regions. By default, the metadata structure and +algorithm tries to find best place for new allocations among free regions to +optimize memory usage. This way you can allocate and free objects in any order. + + + +Sometimes there is a need to use simpler, linear allocation algorithm. You can +create custom pool that uses such algorithm by adding flag +#VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT to VmaPoolCreateInfo::flags while creating +#VmaPool object. Then an alternative metadata management is used. It always +creates new allocations after last one and doesn't reuse free regions after +allocations freed in the middle. It results in better allocation performance and +less memory consumed by metadata. + + + +With this one flag, you can create a custom pool that can be used in many ways: +free-at-once, stack, double stack, and ring buffer. See below for details. + +\subsection linear_algorithm_free_at_once Free-at-once + +In a pool that uses linear algorithm, you still need to free all the allocations +individually, e.g. by using vmaFreeMemory() or vmaDestroyBuffer(). You can free +them in any order. New allocations are always made after last one - free space +in the middle is not reused. However, when you release all the allocation and +the pool becomes empty, allocation starts from the beginning again. This way you +can use linear algorithm to speed up creation of allocations that you are going +to release all at once. + + + +This mode is also available for pools created with VmaPoolCreateInfo::maxBlockCount +value that allows multiple memory blocks. + +\subsection linear_algorithm_stack Stack + +When you free an allocation that was created last, its space can be reused. +Thanks to this, if you always release allocations in the order opposite to their +creation (LIFO - Last In First Out), you can achieve behavior of a stack. + + + +This mode is also available for pools created with VmaPoolCreateInfo::maxBlockCount +value that allows multiple memory blocks. + +\subsection linear_algorithm_double_stack Double stack + +The space reserved by a custom pool with linear algorithm may be used by two +stacks: + +- First, default one, growing up from offset 0. +- Second, "upper" one, growing down from the end towards lower offsets. + +To make allocation from upper stack, add flag #VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT +to VmaAllocationCreateInfo::flags. + + + +Double stack is available only in pools with one memory block - +VmaPoolCreateInfo::maxBlockCount must be 1. Otherwise behavior is undefined. + +When the two stacks' ends meet so there is not enough space between them for a +new allocation, such allocation fails with usual +`VK_ERROR_OUT_OF_DEVICE_MEMORY` error. + +\subsection linear_algorithm_ring_buffer Ring buffer + +When you free some allocations from the beginning and there is not enough free space +for a new one at the end of a pool, allocator's "cursor" wraps around to the +beginning and starts allocation there. Thanks to this, if you always release +allocations in the same order as you created them (FIFO - First In First Out), +you can achieve behavior of a ring buffer / queue. + + + +Pools with linear algorithm support [lost allocations](@ref lost_allocations) when used as ring buffer. +If there is not enough free space for a new allocation, but existing allocations +from the front of the queue can become lost, they become lost and the allocation +succeeds. + + + +Ring buffer is available only in pools with one memory block - +VmaPoolCreateInfo::maxBlockCount must be 1. Otherwise behavior is undefined. + +\section buddy_algorithm Buddy allocation algorithm + +There is another allocation algorithm that can be used with custom pools, called +"buddy". Its internal data structure is based on a tree of blocks, each having +size that is a power of two and a half of its parent's size. When you want to +allocate memory of certain size, a free node in the tree is located. If it's too +large, it is recursively split into two halves (called "buddies"). However, if +requested allocation size is not a power of two, the size of a tree node is +aligned up to the nearest power of two and the remaining space is wasted. When +two buddy nodes become free, they are merged back into one larger node. + + + +The advantage of buddy allocation algorithm over default algorithm is faster +allocation and deallocation, as well as smaller external fragmentation. The +disadvantage is more wasted space (internal fragmentation). + +For more information, please read ["Buddy memory allocation" on Wikipedia](https://en.wikipedia.org/wiki/Buddy_memory_allocation) +or other sources that describe this concept in general. + +To use buddy allocation algorithm with a custom pool, add flag +#VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT to VmaPoolCreateInfo::flags while creating +#VmaPool object. + +Several limitations apply to pools that use buddy algorithm: + +- It is recommended to use VmaPoolCreateInfo::blockSize that is a power of two. + Otherwise, only largest power of two smaller than the size is used for + allocations. The remaining space always stays unused. +- [Margins](@ref debugging_memory_usage_margins) and + [corruption detection](@ref debugging_memory_usage_corruption_detection) + don't work in such pools. +- [Lost allocations](@ref lost_allocations) don't work in such pools. You can + use them, but they never become lost. Support may be added in the future. +- [Defragmentation](@ref defragmentation) doesn't work with allocations made from + such pool. + +\page defragmentation Defragmentation + +Interleaved allocations and deallocations of many objects of varying size can +cause fragmentation over time, which can lead to a situation where the library is unable +to find a continuous range of free memory for a new allocation despite there is +enough free space, just scattered across many small free ranges between existing +allocations. + +To mitigate this problem, you can use defragmentation feature: +structure #VmaDefragmentationInfo2, function vmaDefragmentationBegin(), vmaDefragmentationEnd(). +Given set of allocations, +this function can move them to compact used memory, ensure more continuous free +space and possibly also free some `VkDeviceMemory` blocks. + +What the defragmentation does is: + +- Updates #VmaAllocation objects to point to new `VkDeviceMemory` and offset. + After allocation has been moved, its VmaAllocationInfo::deviceMemory and/or + VmaAllocationInfo::offset changes. You must query them again using + vmaGetAllocationInfo() if you need them. +- Moves actual data in memory. + +What it doesn't do, so you need to do it yourself: + +- Recreate buffers and images that were bound to allocations that were defragmented and + bind them with their new places in memory. + You must use `vkDestroyBuffer()`, `vkDestroyImage()`, + `vkCreateBuffer()`, `vkCreateImage()` for that purpose and NOT vmaDestroyBuffer(), + vmaDestroyImage(), vmaCreateBuffer(), vmaCreateImage(), because you don't need to + destroy or create allocation objects! +- Recreate views and update descriptors that point to these buffers and images. + +\section defragmentation_cpu Defragmenting CPU memory + +Following example demonstrates how you can run defragmentation on CPU. +Only allocations created in memory types that are `HOST_VISIBLE` can be defragmented. +Others are ignored. + +The way it works is: + +- It temporarily maps entire memory blocks when necessary. +- It moves data using `memmove()` function. + +\code +// Given following variables already initialized: +VkDevice device; +VmaAllocator allocator; +std::vector<VkBuffer> buffers; +std::vector<VmaAllocation> allocations; + + +const uint32_t allocCount = (uint32_t)allocations.size(); +std::vector<VkBool32> allocationsChanged(allocCount); + +VmaDefragmentationInfo2 defragInfo = {}; +defragInfo.allocationCount = allocCount; +defragInfo.pAllocations = allocations.data(); +defragInfo.pAllocationsChanged = allocationsChanged.data(); +defragInfo.maxCpuBytesToMove = VK_WHOLE_SIZE; // No limit. +defragInfo.maxCpuAllocationsToMove = UINT32_MAX; // No limit. + +VmaDefragmentationContext defragCtx; +vmaDefragmentationBegin(allocator, &defragInfo, nullptr, &defragCtx); +vmaDefragmentationEnd(allocator, defragCtx); + +for(uint32_t i = 0; i < allocCount; ++i) +{ + if(allocationsChanged[i]) + { + // Destroy buffer that is immutably bound to memory region which is no longer valid. + vkDestroyBuffer(device, buffers[i], nullptr); + + // Create new buffer with same parameters. + VkBufferCreateInfo bufferInfo = ...; + vkCreateBuffer(device, &bufferInfo, nullptr, &buffers[i]); + + // You can make dummy call to vkGetBufferMemoryRequirements here to silence validation layer warning. + + // Bind new buffer to new memory region. Data contained in it is already moved. + VmaAllocationInfo allocInfo; + vmaGetAllocationInfo(allocator, allocations[i], &allocInfo); + vkBindBufferMemory(device, buffers[i], allocInfo.deviceMemory, allocInfo.offset); + } +} +\endcode + +Setting VmaDefragmentationInfo2::pAllocationsChanged is optional. +This output array tells whether particular allocation in VmaDefragmentationInfo2::pAllocations at the same index +has been modified during defragmentation. +You can pass null, but you then need to query every allocation passed to defragmentation +for new parameters using vmaGetAllocationInfo() if you might need to recreate and rebind a buffer or image associated with it. + +If you use [Custom memory pools](@ref choosing_memory_type_custom_memory_pools), +you can fill VmaDefragmentationInfo2::poolCount and VmaDefragmentationInfo2::pPools +instead of VmaDefragmentationInfo2::allocationCount and VmaDefragmentationInfo2::pAllocations +to defragment all allocations in given pools. +You cannot use VmaDefragmentationInfo2::pAllocationsChanged in that case. +You can also combine both methods. + +\section defragmentation_gpu Defragmenting GPU memory + +It is also possible to defragment allocations created in memory types that are not `HOST_VISIBLE`. +To do that, you need to pass a command buffer that meets requirements as described in +VmaDefragmentationInfo2::commandBuffer. The way it works is: + +- It creates temporary buffers and binds them to entire memory blocks when necessary. +- It issues `vkCmdCopyBuffer()` to passed command buffer. + +Example: + +\code +// Given following variables already initialized: +VkDevice device; +VmaAllocator allocator; +VkCommandBuffer commandBuffer; +std::vector<VkBuffer> buffers; +std::vector<VmaAllocation> allocations; + + +const uint32_t allocCount = (uint32_t)allocations.size(); +std::vector<VkBool32> allocationsChanged(allocCount); + +VkCommandBufferBeginInfo cmdBufBeginInfo = ...; +vkBeginCommandBuffer(commandBuffer, &cmdBufBeginInfo); + +VmaDefragmentationInfo2 defragInfo = {}; +defragInfo.allocationCount = allocCount; +defragInfo.pAllocations = allocations.data(); +defragInfo.pAllocationsChanged = allocationsChanged.data(); +defragInfo.maxGpuBytesToMove = VK_WHOLE_SIZE; // Notice it's "GPU" this time. +defragInfo.maxGpuAllocationsToMove = UINT32_MAX; // Notice it's "GPU" this time. +defragInfo.commandBuffer = commandBuffer; + +VmaDefragmentationContext defragCtx; +vmaDefragmentationBegin(allocator, &defragInfo, nullptr, &defragCtx); + +vkEndCommandBuffer(commandBuffer); + +// Submit commandBuffer. +// Wait for a fence that ensures commandBuffer execution finished. + +vmaDefragmentationEnd(allocator, defragCtx); + +for(uint32_t i = 0; i < allocCount; ++i) +{ + if(allocationsChanged[i]) + { + // Destroy buffer that is immutably bound to memory region which is no longer valid. + vkDestroyBuffer(device, buffers[i], nullptr); + + // Create new buffer with same parameters. + VkBufferCreateInfo bufferInfo = ...; + vkCreateBuffer(device, &bufferInfo, nullptr, &buffers[i]); + + // You can make dummy call to vkGetBufferMemoryRequirements here to silence validation layer warning. + + // Bind new buffer to new memory region. Data contained in it is already moved. + VmaAllocationInfo allocInfo; + vmaGetAllocationInfo(allocator, allocations[i], &allocInfo); + vkBindBufferMemory(device, buffers[i], allocInfo.deviceMemory, allocInfo.offset); + } +} +\endcode + +You can combine these two methods by specifying non-zero `maxGpu*` as well as `maxCpu*` parameters. +The library automatically chooses best method to defragment each memory pool. + +You may try not to block your entire program to wait until defragmentation finishes, +but do it in the background, as long as you carefully fullfill requirements described +in function vmaDefragmentationBegin(). + +\section defragmentation_additional_notes Additional notes + +While using defragmentation, you may experience validation layer warnings, which you just need to ignore. +See [Validation layer warnings](@ref general_considerations_validation_layer_warnings). + +If you defragment allocations bound to images, these images should be created with +`VK_IMAGE_CREATE_ALIAS_BIT` flag, to make sure that new image created with same +parameters and pointing to data copied to another memory region will interpret +its contents consistently. Otherwise you may experience corrupted data on some +implementations, e.g. due to different pixel swizzling used internally by the graphics driver. + +If you defragment allocations bound to images, new images to be bound to new +memory region after defragmentation should be created with `VK_IMAGE_LAYOUT_PREINITIALIZED` +and then transitioned to their original layout from before defragmentation using +an image memory barrier. + +Please don't expect memory to be fully compacted after defragmentation. +Algorithms inside are based on some heuristics that try to maximize number of Vulkan +memory blocks to make totally empty to release them, as well as to maximimze continuous +empty space inside remaining blocks, while minimizing the number and size of allocations that +need to be moved. Some fragmentation may still remain - this is normal. + +\section defragmentation_custom_algorithm Writing custom defragmentation algorithm + +If you want to implement your own, custom defragmentation algorithm, +there is infrastructure prepared for that, +but it is not exposed through the library API - you need to hack its source code. +Here are steps needed to do this: + +-# Main thing you need to do is to define your own class derived from base abstract + class `VmaDefragmentationAlgorithm` and implement your version of its pure virtual methods. + See definition and comments of this class for details. +-# Your code needs to interact with device memory block metadata. + If you need more access to its data than it's provided by its public interface, + declare your new class as a friend class e.g. in class `VmaBlockMetadata_Generic`. +-# If you want to create a flag that would enable your algorithm or pass some additional + flags to configure it, add them to `VmaDefragmentationFlagBits` and use them in + VmaDefragmentationInfo2::flags. +-# Modify function `VmaBlockVectorDefragmentationContext::Begin` to create object + of your new class whenever needed. + + +\page lost_allocations Lost allocations + +If your game oversubscribes video memory, if may work OK in previous-generation +graphics APIs (DirectX 9, 10, 11, OpenGL) because resources are automatically +paged to system RAM. In Vulkan you can't do it because when you run out of +memory, an allocation just fails. If you have more data (e.g. textures) that can +fit into VRAM and you don't need it all at once, you may want to upload them to +GPU on demand and "push out" ones that are not used for a long time to make room +for the new ones, effectively using VRAM (or a cartain memory pool) as a form of +cache. Vulkan Memory Allocator can help you with that by supporting a concept of +"lost allocations". + +To create an allocation that can become lost, include #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT +flag in VmaAllocationCreateInfo::flags. Before using a buffer or image bound to +such allocation in every new frame, you need to query it if it's not lost. +To check it, call vmaTouchAllocation(). +If the allocation is lost, you should not use it or buffer/image bound to it. +You mustn't forget to destroy this allocation and this buffer/image. +vmaGetAllocationInfo() can also be used for checking status of the allocation. +Allocation is lost when returned VmaAllocationInfo::deviceMemory == `VK_NULL_HANDLE`. + +To create an allocation that can make some other allocations lost to make room +for it, use #VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT flag. You will +usually use both flags #VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT and +#VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT at the same time. + +Warning! Current implementation uses quite naive, brute force algorithm, +which can make allocation calls that use #VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT +flag quite slow. A new, more optimal algorithm and data structure to speed this +up is planned for the future. + +<b>Q: When interleaving creation of new allocations with usage of existing ones, +how do you make sure that an allocation won't become lost while it's used in the +current frame?</b> + +It is ensured because vmaTouchAllocation() / vmaGetAllocationInfo() not only returns allocation +status/parameters and checks whether it's not lost, but when it's not, it also +atomically marks it as used in the current frame, which makes it impossible to +become lost in that frame. It uses lockless algorithm, so it works fast and +doesn't involve locking any internal mutex. + +<b>Q: What if my allocation may still be in use by the GPU when it's rendering a +previous frame while I already submit new frame on the CPU?</b> + +You can make sure that allocations "touched" by vmaTouchAllocation() / vmaGetAllocationInfo() will not +become lost for a number of additional frames back from the current one by +specifying this number as VmaAllocatorCreateInfo::frameInUseCount (for default +memory pool) and VmaPoolCreateInfo::frameInUseCount (for custom pool). + +<b>Q: How do you inform the library when new frame starts?</b> + +You need to call function vmaSetCurrentFrameIndex(). + +Example code: + +\code +struct MyBuffer +{ + VkBuffer m_Buf = nullptr; + VmaAllocation m_Alloc = nullptr; + + // Called when the buffer is really needed in the current frame. + void EnsureBuffer(); +}; + +void MyBuffer::EnsureBuffer() +{ + // Buffer has been created. + if(m_Buf != VK_NULL_HANDLE) + { + // Check if its allocation is not lost + mark it as used in current frame. + if(vmaTouchAllocation(allocator, m_Alloc)) + { + // It's all OK - safe to use m_Buf. + return; + } + } + + // Buffer not yet exists or lost - destroy and recreate it. + + vmaDestroyBuffer(allocator, m_Buf, m_Alloc); + + VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; + bufCreateInfo.size = 1024; + bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + + VmaAllocationCreateInfo allocCreateInfo = {}; + allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; + allocCreateInfo.flags = VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT | + VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT; + + vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &m_Buf, &m_Alloc, nullptr); +} +\endcode + +When using lost allocations, you may see some Vulkan validation layer warnings +about overlapping regions of memory bound to different kinds of buffers and +images. This is still valid as long as you implement proper handling of lost +allocations (like in the example above) and don't use them. + +You can create an allocation that is already in lost state from the beginning using function +vmaCreateLostAllocation(). It may be useful if you need a "dummy" allocation that is not null. + +You can call function vmaMakePoolAllocationsLost() to set all eligible allocations +in a specified custom pool to lost state. +Allocations that have been "touched" in current frame or VmaPoolCreateInfo::frameInUseCount frames back +cannot become lost. + +<b>Q: Can I touch allocation that cannot become lost?</b> + +Yes, although it has no visible effect. +Calls to vmaGetAllocationInfo() and vmaTouchAllocation() update last use frame index +also for allocations that cannot become lost, but the only way to observe it is to dump +internal allocator state using vmaBuildStatsString(). +You can use this feature for debugging purposes to explicitly mark allocations that you use +in current frame and then analyze JSON dump to see for how long each allocation stays unused. + + +\page statistics Statistics + +This library contains functions that return information about its internal state, +especially the amount of memory allocated from Vulkan. +Please keep in mind that these functions need to traverse all internal data structures +to gather these information, so they may be quite time-consuming. +Don't call them too often. + +\section statistics_numeric_statistics Numeric statistics + +You can query for overall statistics of the allocator using function vmaCalculateStats(). +Information are returned using structure #VmaStats. +It contains #VmaStatInfo - number of allocated blocks, number of allocations +(occupied ranges in these blocks), number of unused (free) ranges in these blocks, +number of bytes used and unused (but still allocated from Vulkan) and other information. +They are summed across memory heaps, memory types and total for whole allocator. + +You can query for statistics of a custom pool using function vmaGetPoolStats(). +Information are returned using structure #VmaPoolStats. + +You can query for information about specific allocation using function vmaGetAllocationInfo(). +It fill structure #VmaAllocationInfo. + +\section statistics_json_dump JSON dump + +You can dump internal state of the allocator to a string in JSON format using function vmaBuildStatsString(). +The result is guaranteed to be correct JSON. +It uses ANSI encoding. +Any strings provided by user (see [Allocation names](@ref allocation_names)) +are copied as-is and properly escaped for JSON, so if they use UTF-8, ISO-8859-2 or any other encoding, +this JSON string can be treated as using this encoding. +It must be freed using function vmaFreeStatsString(). + +The format of this JSON string is not part of official documentation of the library, +but it will not change in backward-incompatible way without increasing library major version number +and appropriate mention in changelog. + +The JSON string contains all the data that can be obtained using vmaCalculateStats(). +It can also contain detailed map of allocated memory blocks and their regions - +free and occupied by allocations. +This allows e.g. to visualize the memory or assess fragmentation. + + +\page allocation_annotation Allocation names and user data + +\section allocation_user_data Allocation user data + +You can annotate allocations with your own information, e.g. for debugging purposes. +To do that, fill VmaAllocationCreateInfo::pUserData field when creating +an allocation. It's an opaque `void*` pointer. You can use it e.g. as a pointer, +some handle, index, key, ordinal number or any other value that would associate +the allocation with your custom metadata. + +\code +VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +// Fill bufferInfo... + +MyBufferMetadata* pMetadata = CreateBufferMetadata(); + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.pUserData = pMetadata; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocCreateInfo, &buffer, &allocation, nullptr); +\endcode + +The pointer may be later retrieved as VmaAllocationInfo::pUserData: + +\code +VmaAllocationInfo allocInfo; +vmaGetAllocationInfo(allocator, allocation, &allocInfo); +MyBufferMetadata* pMetadata = (MyBufferMetadata*)allocInfo.pUserData; +\endcode + +It can also be changed using function vmaSetAllocationUserData(). + +Values of (non-zero) allocations' `pUserData` are printed in JSON report created by +vmaBuildStatsString(), in hexadecimal form. + +\section allocation_names Allocation names + +There is alternative mode available where `pUserData` pointer is used to point to +a null-terminated string, giving a name to the allocation. To use this mode, +set #VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT flag in VmaAllocationCreateInfo::flags. +Then `pUserData` passed as VmaAllocationCreateInfo::pUserData or argument to +vmaSetAllocationUserData() must be either null or pointer to a null-terminated string. +The library creates internal copy of the string, so the pointer you pass doesn't need +to be valid for whole lifetime of the allocation. You can free it after the call. + +\code +VkImageCreateInfo imageInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO }; +// Fill imageInfo... + +std::string imageName = "Texture: "; +imageName += fileName; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT; +allocCreateInfo.pUserData = imageName.c_str(); + +VkImage image; +VmaAllocation allocation; +vmaCreateImage(allocator, &imageInfo, &allocCreateInfo, &image, &allocation, nullptr); +\endcode + +The value of `pUserData` pointer of the allocation will be different than the one +you passed when setting allocation's name - pointing to a buffer managed +internally that holds copy of the string. + +\code +VmaAllocationInfo allocInfo; +vmaGetAllocationInfo(allocator, allocation, &allocInfo); +const char* imageName = (const char*)allocInfo.pUserData; +printf("Image name: %s\n", imageName); +\endcode + +That string is also printed in JSON report created by vmaBuildStatsString(). + + +\page debugging_memory_usage Debugging incorrect memory usage + +If you suspect a bug with memory usage, like usage of uninitialized memory or +memory being overwritten out of bounds of an allocation, +you can use debug features of this library to verify this. + +\section debugging_memory_usage_initialization Memory initialization + +If you experience a bug with incorrect and nondeterministic data in your program and you suspect uninitialized memory to be used, +you can enable automatic memory initialization to verify this. +To do it, define macro `VMA_DEBUG_INITIALIZE_ALLOCATIONS` to 1. + +\code +#define VMA_DEBUG_INITIALIZE_ALLOCATIONS 1 +#include "vk_mem_alloc.h" +\endcode + +It makes memory of all new allocations initialized to bit pattern `0xDCDCDCDC`. +Before an allocation is destroyed, its memory is filled with bit pattern `0xEFEFEFEF`. +Memory is automatically mapped and unmapped if necessary. + +If you find these values while debugging your program, good chances are that you incorrectly +read Vulkan memory that is allocated but not initialized, or already freed, respectively. + +Memory initialization works only with memory types that are `HOST_VISIBLE`. +It works also with dedicated allocations. +It doesn't work with allocations created with #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag, +as they cannot be mapped. + +\section debugging_memory_usage_margins Margins + +By default, allocations are laid out in memory blocks next to each other if possible +(considering required alignment, `bufferImageGranularity`, and `nonCoherentAtomSize`). + + + +Define macro `VMA_DEBUG_MARGIN` to some non-zero value (e.g. 16) to enforce specified +number of bytes as a margin before and after every allocation. + +\code +#define VMA_DEBUG_MARGIN 16 +#include "vk_mem_alloc.h" +\endcode + + + +If your bug goes away after enabling margins, it means it may be caused by memory +being overwritten outside of allocation boundaries. It is not 100% certain though. +Change in application behavior may also be caused by different order and distribution +of allocations across memory blocks after margins are applied. + +The margin is applied also before first and after last allocation in a block. +It may occur only once between two adjacent allocations. + +Margins work with all types of memory. + +Margin is applied only to allocations made out of memory blocks and not to dedicated +allocations, which have their own memory block of specific size. +It is thus not applied to allocations made using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT flag +or those automatically decided to put into dedicated allocations, e.g. due to its +large size or recommended by VK_KHR_dedicated_allocation extension. +Margins are also not active in custom pools created with #VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT flag. + +Margins appear in [JSON dump](@ref statistics_json_dump) as part of free space. + +Note that enabling margins increases memory usage and fragmentation. + +\section debugging_memory_usage_corruption_detection Corruption detection + +You can additionally define macro `VMA_DEBUG_DETECT_CORRUPTION` to 1 to enable validation +of contents of the margins. + +\code +#define VMA_DEBUG_MARGIN 16 +#define VMA_DEBUG_DETECT_CORRUPTION 1 +#include "vk_mem_alloc.h" +\endcode + +When this feature is enabled, number of bytes specified as `VMA_DEBUG_MARGIN` +(it must be multiply of 4) before and after every allocation is filled with a magic number. +This idea is also know as "canary". +Memory is automatically mapped and unmapped if necessary. + +This number is validated automatically when the allocation is destroyed. +If it's not equal to the expected value, `VMA_ASSERT()` is executed. +It clearly means that either CPU or GPU overwritten the memory outside of boundaries of the allocation, +which indicates a serious bug. + +You can also explicitly request checking margins of all allocations in all memory blocks +that belong to specified memory types by using function vmaCheckCorruption(), +or in memory blocks that belong to specified custom pool, by using function +vmaCheckPoolCorruption(). + +Margin validation (corruption detection) works only for memory types that are +`HOST_VISIBLE` and `HOST_COHERENT`. + + +\page record_and_replay Record and replay + +\section record_and_replay_introduction Introduction + +While using the library, sequence of calls to its functions together with their +parameters can be recorded to a file and later replayed using standalone player +application. It can be useful to: + +- Test correctness - check if same sequence of calls will not cause crash or + failures on a target platform. +- Gather statistics - see number of allocations, peak memory usage, number of + calls etc. +- Benchmark performance - see how much time it takes to replay the whole + sequence. + +\section record_and_replay_usage Usage + +<b>To record sequence of calls to a file:</b> Fill in +VmaAllocatorCreateInfo::pRecordSettings member while creating #VmaAllocator +object. File is opened and written during whole lifetime of the allocator. + +<b>To replay file:</b> Use VmaReplay - standalone command-line program. +Precompiled binary can be found in "bin" directory. +Its source can be found in "src/VmaReplay" directory. +Its project is generated by Premake. +Command line syntax is printed when the program is launched without parameters. +Basic usage: + + VmaReplay.exe MyRecording.csv + +<b>Documentation of file format</b> can be found in file: "docs/Recording file format.md". +It's a human-readable, text file in CSV format (Comma Separated Values). + +\section record_and_replay_additional_considerations Additional considerations + +- Replaying file that was recorded on a different GPU (with different parameters + like `bufferImageGranularity`, `nonCoherentAtomSize`, and especially different + set of memory heaps and types) may give different performance and memory usage + results, as well as issue some warnings and errors. +- Current implementation of recording in VMA, as well as VmaReplay application, is + coded and tested only on Windows. Inclusion of recording code is driven by + `VMA_RECORDING_ENABLED` macro. Support for other platforms should be easy to + add. Contributions are welcomed. +- Currently calls to vmaDefragment() function are not recorded. + + +\page usage_patterns Recommended usage patterns + +See also slides from talk: +[Sawicki, Adam. Advanced Graphics Techniques Tutorial: Memory management in Vulkan and DX12. Game Developers Conference, 2018](https://www.gdcvault.com/play/1025458/Advanced-Graphics-Techniques-Tutorial-New) + + +\section usage_patterns_simple Simple patterns + +\subsection usage_patterns_simple_render_targets Render targets + +<b>When:</b> +Any resources that you frequently write and read on GPU, +e.g. images used as color attachments (aka "render targets"), depth-stencil attachments, +images/buffers used as storage image/buffer (aka "Unordered Access View (UAV)"). + +<b>What to do:</b> +Create them in video memory that is fastest to access from GPU using +#VMA_MEMORY_USAGE_GPU_ONLY. + +Consider using [VK_KHR_dedicated_allocation](@ref vk_khr_dedicated_allocation) extension +and/or manually creating them as dedicated allocations using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT, +especially if they are large or if you plan to destroy and recreate them e.g. when +display resolution changes. +Prefer to create such resources first and all other GPU resources (like textures and vertex buffers) later. + +\subsection usage_patterns_simple_immutable_resources Immutable resources + +<b>When:</b> +Any resources that you fill on CPU only once (aka "immutable") or infrequently +and then read frequently on GPU, +e.g. textures, vertex and index buffers, constant buffers that don't change often. + +<b>What to do:</b> +Create them in video memory that is fastest to access from GPU using +#VMA_MEMORY_USAGE_GPU_ONLY. + +To initialize content of such resource, create a CPU-side (aka "staging") copy of it +in system memory - #VMA_MEMORY_USAGE_CPU_ONLY, map it, fill it, +and submit a transfer from it to the GPU resource. +You can keep the staging copy if you need it for another upload transfer in the future. +If you don't, you can destroy it or reuse this buffer for uploading different resource +after the transfer finishes. + +Prefer to create just buffers in system memory rather than images, even for uploading textures. +Use `vkCmdCopyBufferToImage()`. +Dont use images with `VK_IMAGE_TILING_LINEAR`. + +\subsection usage_patterns_dynamic_resources Dynamic resources + +<b>When:</b> +Any resources that change frequently (aka "dynamic"), e.g. every frame or every draw call, +written on CPU, read on GPU. + +<b>What to do:</b> +Create them using #VMA_MEMORY_USAGE_CPU_TO_GPU. +You can map it and write to it directly on CPU, as well as read from it on GPU. + +This is a more complex situation. Different solutions are possible, +and the best one depends on specific GPU type, but you can use this simple approach for the start. +Prefer to write to such resource sequentially (e.g. using `memcpy`). +Don't perform random access or any reads from it on CPU, as it may be very slow. + +\subsection usage_patterns_readback Readback + +<b>When:</b> +Resources that contain data written by GPU that you want to read back on CPU, +e.g. results of some computations. + +<b>What to do:</b> +Create them using #VMA_MEMORY_USAGE_GPU_TO_CPU. +You can write to them directly on GPU, as well as map and read them on CPU. + +\section usage_patterns_advanced Advanced patterns + +\subsection usage_patterns_integrated_graphics Detecting integrated graphics + +You can support integrated graphics (like Intel HD Graphics, AMD APU) better +by detecting it in Vulkan. +To do it, call `vkGetPhysicalDeviceProperties()`, inspect +`VkPhysicalDeviceProperties::deviceType` and look for `VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU`. +When you find it, you can assume that memory is unified and all memory types are comparably fast +to access from GPU, regardless of `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. + +You can then sum up sizes of all available memory heaps and treat them as useful for +your GPU resources, instead of only `DEVICE_LOCAL` ones. +You can also prefer to create your resources in memory types that are `HOST_VISIBLE` to map them +directly instead of submitting explicit transfer (see below). + +\subsection usage_patterns_direct_vs_transfer Direct access versus transfer + +For resources that you frequently write on CPU and read on GPU, many solutions are possible: + +-# Create one copy in video memory using #VMA_MEMORY_USAGE_GPU_ONLY, + second copy in system memory using #VMA_MEMORY_USAGE_CPU_ONLY and submit explicit tranfer each time. +-# Create just single copy using #VMA_MEMORY_USAGE_CPU_TO_GPU, map it and fill it on CPU, + read it directly on GPU. +-# Create just single copy using #VMA_MEMORY_USAGE_CPU_ONLY, map it and fill it on CPU, + read it directly on GPU. + +Which solution is the most efficient depends on your resource and especially on the GPU. +It is best to measure it and then make the decision. +Some general recommendations: + +- On integrated graphics use (2) or (3) to avoid unnecesary time and memory overhead + related to using a second copy and making transfer. +- For small resources (e.g. constant buffers) use (2). + Discrete AMD cards have special 256 MiB pool of video memory that is directly mappable. + Even if the resource ends up in system memory, its data may be cached on GPU after first + fetch over PCIe bus. +- For larger resources (e.g. textures), decide between (1) and (2). + You may want to differentiate NVIDIA and AMD, e.g. by looking for memory type that is + both `DEVICE_LOCAL` and `HOST_VISIBLE`. When you find it, use (2), otherwise use (1). + +Similarly, for resources that you frequently write on GPU and read on CPU, multiple +solutions are possible: + +-# Create one copy in video memory using #VMA_MEMORY_USAGE_GPU_ONLY, + second copy in system memory using #VMA_MEMORY_USAGE_GPU_TO_CPU and submit explicit tranfer each time. +-# Create just single copy using #VMA_MEMORY_USAGE_GPU_TO_CPU, write to it directly on GPU, + map it and read it on CPU. + +You should take some measurements to decide which option is faster in case of your specific +resource. + +If you don't want to specialize your code for specific types of GPUs, you can still make +an simple optimization for cases when your resource ends up in mappable memory to use it +directly in this case instead of creating CPU-side staging copy. +For details see [Finding out if memory is mappable](@ref memory_mapping_finding_if_memory_mappable). + + +\page configuration Configuration + +Please check "CONFIGURATION SECTION" in the code to find macros that you can define +before each include of this file or change directly in this file to provide +your own implementation of basic facilities like assert, `min()` and `max()` functions, +mutex, atomic etc. +The library uses its own implementation of containers by default, but you can switch to using +STL containers instead. + +\section config_Vulkan_functions Pointers to Vulkan functions + +The library uses Vulkan functions straight from the `vulkan.h` header by default. +If you want to provide your own pointers to these functions, e.g. fetched using +`vkGetInstanceProcAddr()` and `vkGetDeviceProcAddr()`: + +-# Define `VMA_STATIC_VULKAN_FUNCTIONS 0`. +-# Provide valid pointers through VmaAllocatorCreateInfo::pVulkanFunctions. + +\section custom_memory_allocator Custom host memory allocator + +If you use custom allocator for CPU memory rather than default operator `new` +and `delete` from C++, you can make this library using your allocator as well +by filling optional member VmaAllocatorCreateInfo::pAllocationCallbacks. These +functions will be passed to Vulkan, as well as used by the library itself to +make any CPU-side allocations. + +\section allocation_callbacks Device memory allocation callbacks + +The library makes calls to `vkAllocateMemory()` and `vkFreeMemory()` internally. +You can setup callbacks to be informed about these calls, e.g. for the purpose +of gathering some statistics. To do it, fill optional member +VmaAllocatorCreateInfo::pDeviceMemoryCallbacks. + +\section heap_memory_limit Device heap memory limit + +If you want to test how your program behaves with limited amount of Vulkan device +memory available without switching your graphics card to one that really has +smaller VRAM, you can use a feature of this library intended for this purpose. +To do it, fill optional member VmaAllocatorCreateInfo::pHeapSizeLimit. + + + +\page vk_khr_dedicated_allocation VK_KHR_dedicated_allocation + +VK_KHR_dedicated_allocation is a Vulkan extension which can be used to improve +performance on some GPUs. It augments Vulkan API with possibility to query +driver whether it prefers particular buffer or image to have its own, dedicated +allocation (separate `VkDeviceMemory` block) for better efficiency - to be able +to do some internal optimizations. + +The extension is supported by this library. It will be used automatically when +enabled. To enable it: + +1 . When creating Vulkan device, check if following 2 device extensions are +supported (call `vkEnumerateDeviceExtensionProperties()`). +If yes, enable them (fill `VkDeviceCreateInfo::ppEnabledExtensionNames`). + +- VK_KHR_get_memory_requirements2 +- VK_KHR_dedicated_allocation + +If you enabled these extensions: + +2 . Use #VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT flag when creating +your #VmaAllocator`to inform the library that you enabled required extensions +and you want the library to use them. + +\code +allocatorInfo.flags |= VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT; + +vmaCreateAllocator(&allocatorInfo, &allocator); +\endcode + +That's all. The extension will be automatically used whenever you create a +buffer using vmaCreateBuffer() or image using vmaCreateImage(). + +When using the extension together with Vulkan Validation Layer, you will receive +warnings like this: + + vkBindBufferMemory(): Binding memory to buffer 0x33 but vkGetBufferMemoryRequirements() has not been called on that buffer. + +It is OK, you should just ignore it. It happens because you use function +`vkGetBufferMemoryRequirements2KHR()` instead of standard +`vkGetBufferMemoryRequirements()`, while the validation layer seems to be +unaware of it. + +To learn more about this extension, see: + +- [VK_KHR_dedicated_allocation in Vulkan specification](https://www.khronos.org/registry/vulkan/specs/1.0-extensions/html/vkspec.html#VK_KHR_dedicated_allocation) +- [VK_KHR_dedicated_allocation unofficial manual](http://asawicki.info/articles/VK_KHR_dedicated_allocation.php5) + + + +\page general_considerations General considerations + +\section general_considerations_thread_safety Thread safety + +- The library has no global state, so separate #VmaAllocator objects can be used + independently. + There should be no need to create multiple such objects though - one per `VkDevice` is enough. +- By default, all calls to functions that take #VmaAllocator as first parameter + are safe to call from multiple threads simultaneously because they are + synchronized internally when needed. +- When the allocator is created with #VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT + flag, calls to functions that take such #VmaAllocator object must be + synchronized externally. +- Access to a #VmaAllocation object must be externally synchronized. For example, + you must not call vmaGetAllocationInfo() and vmaMapMemory() from different + threads at the same time if you pass the same #VmaAllocation object to these + functions. + +\section general_considerations_validation_layer_warnings Validation layer warnings + +When using this library, you can meet following types of warnings issued by +Vulkan validation layer. They don't necessarily indicate a bug, so you may need +to just ignore them. + +- *vkBindBufferMemory(): Binding memory to buffer 0xeb8e4 but vkGetBufferMemoryRequirements() has not been called on that buffer.* + - It happens when VK_KHR_dedicated_allocation extension is enabled. + `vkGetBufferMemoryRequirements2KHR` function is used instead, while validation layer seems to be unaware of it. +- *Mapping an image with layout VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL can result in undefined behavior if this memory is used by the device. Only GENERAL or PREINITIALIZED should be used.* + - It happens when you map a buffer or image, because the library maps entire + `VkDeviceMemory` block, where different types of images and buffers may end + up together, especially on GPUs with unified memory like Intel. +- *Non-linear image 0xebc91 is aliased with linear buffer 0xeb8e4 which may indicate a bug.* + - It happens when you use lost allocations, and a new image or buffer is + created in place of an existing object that bacame lost. + - It may happen also when you use [defragmentation](@ref defragmentation). + +\section general_considerations_allocation_algorithm Allocation algorithm + +The library uses following algorithm for allocation, in order: + +-# Try to find free range of memory in existing blocks. +-# If failed, try to create a new block of `VkDeviceMemory`, with preferred block size. +-# If failed, try to create such block with size/2, size/4, size/8. +-# If failed and #VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT flag was + specified, try to find space in existing blocks, possilby making some other + allocations lost. +-# If failed, try to allocate separate `VkDeviceMemory` for this allocation, + just like when you use #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +-# If failed, choose other memory type that meets the requirements specified in + VmaAllocationCreateInfo and go to point 1. +-# If failed, return `VK_ERROR_OUT_OF_DEVICE_MEMORY`. + +\section general_considerations_features_not_supported Features not supported + +Features deliberately excluded from the scope of this library: + +- Data transfer. Uploading (straming) and downloading data of buffers and images + between CPU and GPU memory and related synchronization is responsibility of the user. +- Allocations for imported/exported external memory. They tend to require + explicit memory type index and dedicated allocation anyway, so they don't + interact with main features of this library. Such special purpose allocations + should be made manually, using `vkCreateBuffer()` and `vkAllocateMemory()`. +- Recreation of buffers and images. Although the library has functions for + buffer and image creation (vmaCreateBuffer(), vmaCreateImage()), you need to + recreate these objects yourself after defragmentation. That's because the big + structures `VkBufferCreateInfo`, `VkImageCreateInfo` are not stored in + #VmaAllocation object. +- Handling CPU memory allocation failures. When dynamically creating small C++ + objects in CPU memory (not Vulkan memory), allocation failures are not checked + and handled gracefully, because that would complicate code significantly and + is usually not needed in desktop PC applications anyway. +- Code free of any compiler warnings. Maintaining the library to compile and + work correctly on so many different platforms is hard enough. Being free of + any warnings, on any version of any compiler, is simply not feasible. +- This is a C++ library with C interface. + Bindings or ports to any other programming languages are welcomed as external projects and + are not going to be included into this repository. + +*/ + +/* +Define this macro to 0/1 to disable/enable support for recording functionality, +available through VmaAllocatorCreateInfo::pRecordSettings. +*/ +#ifndef VMA_RECORDING_ENABLED + #ifdef _WIN32 + #define VMA_RECORDING_ENABLED 1 + #else + #define VMA_RECORDING_ENABLED 0 + #endif +#endif + +#ifndef NOMINMAX + #define NOMINMAX // For windows.h +#endif + +#ifndef VULKAN_H_ + #include <vulkan/vulkan.h> +#endif + +#if VMA_RECORDING_ENABLED + #include <windows.h> +#endif + +#if !defined(VMA_DEDICATED_ALLOCATION) + #if VK_KHR_get_memory_requirements2 && VK_KHR_dedicated_allocation + #define VMA_DEDICATED_ALLOCATION 1 + #else + #define VMA_DEDICATED_ALLOCATION 0 + #endif +#endif + +/** \struct VmaAllocator +\brief Represents main object of this library initialized. + +Fill structure #VmaAllocatorCreateInfo and call function vmaCreateAllocator() to create it. +Call function vmaDestroyAllocator() to destroy it. + +It is recommended to create just one object of this type per `VkDevice` object, +right after Vulkan is initialized and keep it alive until before Vulkan device is destroyed. +*/ +VK_DEFINE_HANDLE(VmaAllocator) + +/// Callback function called after successful vkAllocateMemory. +typedef void (VKAPI_PTR *PFN_vmaAllocateDeviceMemoryFunction)( + VmaAllocator allocator, + uint32_t memoryType, + VkDeviceMemory memory, + VkDeviceSize size); +/// Callback function called before vkFreeMemory. +typedef void (VKAPI_PTR *PFN_vmaFreeDeviceMemoryFunction)( + VmaAllocator allocator, + uint32_t memoryType, + VkDeviceMemory memory, + VkDeviceSize size); + +/** \brief Set of callbacks that the library will call for `vkAllocateMemory` and `vkFreeMemory`. + +Provided for informative purpose, e.g. to gather statistics about number of +allocations or total amount of memory allocated in Vulkan. + +Used in VmaAllocatorCreateInfo::pDeviceMemoryCallbacks. +*/ +typedef struct VmaDeviceMemoryCallbacks { + /// Optional, can be null. + PFN_vmaAllocateDeviceMemoryFunction pfnAllocate; + /// Optional, can be null. + PFN_vmaFreeDeviceMemoryFunction pfnFree; +} VmaDeviceMemoryCallbacks; + +/// Flags for created #VmaAllocator. +typedef enum VmaAllocatorCreateFlagBits { + /** \brief Allocator and all objects created from it will not be synchronized internally, so you must guarantee they are used from only one thread at a time or synchronized externally by you. + + Using this flag may increase performance because internal mutexes are not used. + */ + VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT = 0x00000001, + /** \brief Enables usage of VK_KHR_dedicated_allocation extension. + + Using this extenion will automatically allocate dedicated blocks of memory for + some buffers and images instead of suballocating place for them out of bigger + memory blocks (as if you explicitly used #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT + flag) when it is recommended by the driver. It may improve performance on some + GPUs. + + You may set this flag only if you found out that following device extensions are + supported, you enabled them while creating Vulkan device passed as + VmaAllocatorCreateInfo::device, and you want them to be used internally by this + library: + + - VK_KHR_get_memory_requirements2 + - VK_KHR_dedicated_allocation + +When this flag is set, you can experience following warnings reported by Vulkan +validation layer. You can ignore them. + +> vkBindBufferMemory(): Binding memory to buffer 0x2d but vkGetBufferMemoryRequirements() has not been called on that buffer. + */ + VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT = 0x00000002, + + VMA_ALLOCATOR_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaAllocatorCreateFlagBits; +typedef VkFlags VmaAllocatorCreateFlags; + +/** \brief Pointers to some Vulkan functions - a subset used by the library. + +Used in VmaAllocatorCreateInfo::pVulkanFunctions. +*/ +typedef struct VmaVulkanFunctions { + PFN_vkGetPhysicalDeviceProperties vkGetPhysicalDeviceProperties; + PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties; + PFN_vkAllocateMemory vkAllocateMemory; + PFN_vkFreeMemory vkFreeMemory; + PFN_vkMapMemory vkMapMemory; + PFN_vkUnmapMemory vkUnmapMemory; + PFN_vkFlushMappedMemoryRanges vkFlushMappedMemoryRanges; + PFN_vkInvalidateMappedMemoryRanges vkInvalidateMappedMemoryRanges; + PFN_vkBindBufferMemory vkBindBufferMemory; + PFN_vkBindImageMemory vkBindImageMemory; + PFN_vkGetBufferMemoryRequirements vkGetBufferMemoryRequirements; + PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements; + PFN_vkCreateBuffer vkCreateBuffer; + PFN_vkDestroyBuffer vkDestroyBuffer; + PFN_vkCreateImage vkCreateImage; + PFN_vkDestroyImage vkDestroyImage; + PFN_vkCmdCopyBuffer vkCmdCopyBuffer; +#if VMA_DEDICATED_ALLOCATION + PFN_vkGetBufferMemoryRequirements2KHR vkGetBufferMemoryRequirements2KHR; + PFN_vkGetImageMemoryRequirements2KHR vkGetImageMemoryRequirements2KHR; +#endif +} VmaVulkanFunctions; + +/// Flags to be used in VmaRecordSettings::flags. +typedef enum VmaRecordFlagBits { + /** \brief Enables flush after recording every function call. + + Enable it if you expect your application to crash, which may leave recording file truncated. + It may degrade performance though. + */ + VMA_RECORD_FLUSH_AFTER_CALL_BIT = 0x00000001, + + VMA_RECORD_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaRecordFlagBits; +typedef VkFlags VmaRecordFlags; + +/// Parameters for recording calls to VMA functions. To be used in VmaAllocatorCreateInfo::pRecordSettings. +typedef struct VmaRecordSettings +{ + /// Flags for recording. Use #VmaRecordFlagBits enum. + VmaRecordFlags flags; + /** \brief Path to the file that should be written by the recording. + + Suggested extension: "csv". + If the file already exists, it will be overwritten. + It will be opened for the whole time #VmaAllocator object is alive. + If opening this file fails, creation of the whole allocator object fails. + */ + const char* pFilePath; +} VmaRecordSettings; + +/// Description of a Allocator to be created. +typedef struct VmaAllocatorCreateInfo +{ + /// Flags for created allocator. Use #VmaAllocatorCreateFlagBits enum. + VmaAllocatorCreateFlags flags; + /// Vulkan physical device. + /** It must be valid throughout whole lifetime of created allocator. */ + VkPhysicalDevice physicalDevice; + /// Vulkan device. + /** It must be valid throughout whole lifetime of created allocator. */ + VkDevice device; + /// Preferred size of a single `VkDeviceMemory` block to be allocated from large heaps > 1 GiB. Optional. + /** Set to 0 to use default, which is currently 256 MiB. */ + VkDeviceSize preferredLargeHeapBlockSize; + /// Custom CPU memory allocation callbacks. Optional. + /** Optional, can be null. When specified, will also be used for all CPU-side memory allocations. */ + const VkAllocationCallbacks* pAllocationCallbacks; + /// Informative callbacks for `vkAllocateMemory`, `vkFreeMemory`. Optional. + /** Optional, can be null. */ + const VmaDeviceMemoryCallbacks* pDeviceMemoryCallbacks; + /** \brief Maximum number of additional frames that are in use at the same time as current frame. + + This value is used only when you make allocations with + VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag. Such allocation cannot become + lost if allocation.lastUseFrameIndex >= allocator.currentFrameIndex - frameInUseCount. + + For example, if you double-buffer your command buffers, so resources used for + rendering in previous frame may still be in use by the GPU at the moment you + allocate resources needed for the current frame, set this value to 1. + + If you want to allow any allocations other than used in the current frame to + become lost, set this value to 0. + */ + uint32_t frameInUseCount; + /** \brief Either null or a pointer to an array of limits on maximum number of bytes that can be allocated out of particular Vulkan memory heap. + + If not NULL, it must be a pointer to an array of + `VkPhysicalDeviceMemoryProperties::memoryHeapCount` elements, defining limit on + maximum number of bytes that can be allocated out of particular Vulkan memory + heap. + + Any of the elements may be equal to `VK_WHOLE_SIZE`, which means no limit on that + heap. This is also the default in case of `pHeapSizeLimit` = NULL. + + If there is a limit defined for a heap: + + - If user tries to allocate more memory from that heap using this allocator, + the allocation fails with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. + - If the limit is smaller than heap size reported in `VkMemoryHeap::size`, the + value of this limit will be reported instead when using vmaGetMemoryProperties(). + + Warning! Using this feature may not be equivalent to installing a GPU with + smaller amount of memory, because graphics driver doesn't necessary fail new + allocations with `VK_ERROR_OUT_OF_DEVICE_MEMORY` result when memory capacity is + exceeded. It may return success and just silently migrate some device memory + blocks to system RAM. This driver behavior can also be controlled using + VK_AMD_memory_overallocation_behavior extension. + */ + const VkDeviceSize* pHeapSizeLimit; + /** \brief Pointers to Vulkan functions. Can be null if you leave define `VMA_STATIC_VULKAN_FUNCTIONS 1`. + + If you leave define `VMA_STATIC_VULKAN_FUNCTIONS 1` in configuration section, + you can pass null as this member, because the library will fetch pointers to + Vulkan functions internally in a static way, like: + + vulkanFunctions.vkAllocateMemory = &vkAllocateMemory; + + Fill this member if you want to provide your own pointers to Vulkan functions, + e.g. fetched using `vkGetInstanceProcAddr()` and `vkGetDeviceProcAddr()`. + */ + const VmaVulkanFunctions* pVulkanFunctions; + /** \brief Parameters for recording of VMA calls. Can be null. + + If not null, it enables recording of calls to VMA functions to a file. + If support for recording is not enabled using `VMA_RECORDING_ENABLED` macro, + creation of the allocator object fails with `VK_ERROR_FEATURE_NOT_PRESENT`. + */ + const VmaRecordSettings* pRecordSettings; +} VmaAllocatorCreateInfo; + +/// Creates Allocator object. +VkResult vmaCreateAllocator( + const VmaAllocatorCreateInfo* pCreateInfo, + VmaAllocator* pAllocator); + +/// Destroys allocator object. +void vmaDestroyAllocator( + VmaAllocator allocator); + +/** +PhysicalDeviceProperties are fetched from physicalDevice by the allocator. +You can access it here, without fetching it again on your own. +*/ +void vmaGetPhysicalDeviceProperties( + VmaAllocator allocator, + const VkPhysicalDeviceProperties** ppPhysicalDeviceProperties); + +/** +PhysicalDeviceMemoryProperties are fetched from physicalDevice by the allocator. +You can access it here, without fetching it again on your own. +*/ +void vmaGetMemoryProperties( + VmaAllocator allocator, + const VkPhysicalDeviceMemoryProperties** ppPhysicalDeviceMemoryProperties); + +/** +\brief Given Memory Type Index, returns Property Flags of this memory type. + +This is just a convenience function. Same information can be obtained using +vmaGetMemoryProperties(). +*/ +void vmaGetMemoryTypeProperties( + VmaAllocator allocator, + uint32_t memoryTypeIndex, + VkMemoryPropertyFlags* pFlags); + +/** \brief Sets index of the current frame. + +This function must be used if you make allocations with +#VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT and +#VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT flags to inform the allocator +when a new frame begins. Allocations queried using vmaGetAllocationInfo() cannot +become lost in the current frame. +*/ +void vmaSetCurrentFrameIndex( + VmaAllocator allocator, + uint32_t frameIndex); + +/** \brief Calculated statistics of memory usage in entire allocator. +*/ +typedef struct VmaStatInfo +{ + /// Number of `VkDeviceMemory` Vulkan memory blocks allocated. + uint32_t blockCount; + /// Number of #VmaAllocation allocation objects allocated. + uint32_t allocationCount; + /// Number of free ranges of memory between allocations. + uint32_t unusedRangeCount; + /// Total number of bytes occupied by all allocations. + VkDeviceSize usedBytes; + /// Total number of bytes occupied by unused ranges. + VkDeviceSize unusedBytes; + VkDeviceSize allocationSizeMin, allocationSizeAvg, allocationSizeMax; + VkDeviceSize unusedRangeSizeMin, unusedRangeSizeAvg, unusedRangeSizeMax; +} VmaStatInfo; + +/// General statistics from current state of Allocator. +typedef struct VmaStats +{ + VmaStatInfo memoryType[VK_MAX_MEMORY_TYPES]; + VmaStatInfo memoryHeap[VK_MAX_MEMORY_HEAPS]; + VmaStatInfo total; +} VmaStats; + +/// Retrieves statistics from current state of the Allocator. +void vmaCalculateStats( + VmaAllocator allocator, + VmaStats* pStats); + +#define VMA_STATS_STRING_ENABLED 1 + +#if VMA_STATS_STRING_ENABLED + +/// Builds and returns statistics as string in JSON format. +/** @param[out] ppStatsString Must be freed using vmaFreeStatsString() function. +*/ +void vmaBuildStatsString( + VmaAllocator allocator, + char** ppStatsString, + VkBool32 detailedMap); + +void vmaFreeStatsString( + VmaAllocator allocator, + char* pStatsString); + +#endif // #if VMA_STATS_STRING_ENABLED + +/** \struct VmaPool +\brief Represents custom memory pool + +Fill structure VmaPoolCreateInfo and call function vmaCreatePool() to create it. +Call function vmaDestroyPool() to destroy it. + +For more information see [Custom memory pools](@ref choosing_memory_type_custom_memory_pools). +*/ +VK_DEFINE_HANDLE(VmaPool) + +typedef enum VmaMemoryUsage +{ + /** No intended memory usage specified. + Use other members of VmaAllocationCreateInfo to specify your requirements. + */ + VMA_MEMORY_USAGE_UNKNOWN = 0, + /** Memory will be used on device only, so fast access from the device is preferred. + It usually means device-local GPU (video) memory. + No need to be mappable on host. + It is roughly equivalent of `D3D12_HEAP_TYPE_DEFAULT`. + + Usage: + + - Resources written and read by device, e.g. images used as attachments. + - Resources transferred from host once (immutable) or infrequently and read by + device multiple times, e.g. textures to be sampled, vertex buffers, uniform + (constant) buffers, and majority of other types of resources used on GPU. + + Allocation may still end up in `HOST_VISIBLE` memory on some implementations. + In such case, you are free to map it. + You can use #VMA_ALLOCATION_CREATE_MAPPED_BIT with this usage type. + */ + VMA_MEMORY_USAGE_GPU_ONLY = 1, + /** Memory will be mappable on host. + It usually means CPU (system) memory. + Guarantees to be `HOST_VISIBLE` and `HOST_COHERENT`. + CPU access is typically uncached. Writes may be write-combined. + Resources created in this pool may still be accessible to the device, but access to them can be slow. + It is roughly equivalent of `D3D12_HEAP_TYPE_UPLOAD`. + + Usage: Staging copy of resources used as transfer source. + */ + VMA_MEMORY_USAGE_CPU_ONLY = 2, + /** + Memory that is both mappable on host (guarantees to be `HOST_VISIBLE`) and preferably fast to access by GPU. + CPU access is typically uncached. Writes may be write-combined. + + Usage: Resources written frequently by host (dynamic), read by device. E.g. textures, vertex buffers, uniform buffers updated every frame or every draw call. + */ + VMA_MEMORY_USAGE_CPU_TO_GPU = 3, + /** Memory mappable on host (guarantees to be `HOST_VISIBLE`) and cached. + It is roughly equivalent of `D3D12_HEAP_TYPE_READBACK`. + + Usage: + + - Resources written by device, read by host - results of some computations, e.g. screen capture, average scene luminance for HDR tone mapping. + - Any resources read or accessed randomly on host, e.g. CPU-side copy of vertex buffer used as source of transfer, but also used for collision detection. + */ + VMA_MEMORY_USAGE_GPU_TO_CPU = 4, + VMA_MEMORY_USAGE_MAX_ENUM = 0x7FFFFFFF +} VmaMemoryUsage; + +/// Flags to be passed as VmaAllocationCreateInfo::flags. +typedef enum VmaAllocationCreateFlagBits { + /** \brief Set this flag if the allocation should have its own memory block. + + Use it for special, big resources, like fullscreen images used as attachments. + + This flag must also be used for host visible resources that you want to map + simultaneously because otherwise they might end up as regions of the same + `VkDeviceMemory`, while mapping same `VkDeviceMemory` multiple times + simultaneously is illegal. + + You should not use this flag if VmaAllocationCreateInfo::pool is not null. + */ + VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT = 0x00000001, + + /** \brief Set this flag to only try to allocate from existing `VkDeviceMemory` blocks and never create new such block. + + If new allocation cannot be placed in any of the existing blocks, allocation + fails with `VK_ERROR_OUT_OF_DEVICE_MEMORY` error. + + You should not use #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT and + #VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT at the same time. It makes no sense. + + If VmaAllocationCreateInfo::pool is not null, this flag is implied and ignored. */ + VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT = 0x00000002, + /** \brief Set this flag to use a memory that will be persistently mapped and retrieve pointer to it. + + Pointer to mapped memory will be returned through VmaAllocationInfo::pMappedData. + + Is it valid to use this flag for allocation made from memory type that is not + `HOST_VISIBLE`. This flag is then ignored and memory is not mapped. This is + useful if you need an allocation that is efficient to use on GPU + (`DEVICE_LOCAL`) and still want to map it directly if possible on platforms that + support it (e.g. Intel GPU). + + You should not use this flag together with #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT. + */ + VMA_ALLOCATION_CREATE_MAPPED_BIT = 0x00000004, + /** Allocation created with this flag can become lost as a result of another + allocation with #VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT flag, so you + must check it before use. + + To check if allocation is not lost, call vmaGetAllocationInfo() and check if + VmaAllocationInfo::deviceMemory is not `VK_NULL_HANDLE`. + + For details about supporting lost allocations, see Lost Allocations + chapter of User Guide on Main Page. + + You should not use this flag together with #VMA_ALLOCATION_CREATE_MAPPED_BIT. + */ + VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT = 0x00000008, + /** While creating allocation using this flag, other allocations that were + created with flag #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT can become lost. + + For details about supporting lost allocations, see Lost Allocations + chapter of User Guide on Main Page. + */ + VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT = 0x00000010, + /** Set this flag to treat VmaAllocationCreateInfo::pUserData as pointer to a + null-terminated string. Instead of copying pointer value, a local copy of the + string is made and stored in allocation's `pUserData`. The string is automatically + freed together with the allocation. It is also used in vmaBuildStatsString(). + */ + VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT = 0x00000020, + /** Allocation will be created from upper stack in a double stack pool. + + This flag is only allowed for custom pools created with #VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT flag. + */ + VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT = 0x00000040, + + /** Allocation strategy that chooses smallest possible free range for the + allocation. + */ + VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT = 0x00010000, + /** Allocation strategy that chooses biggest possible free range for the + allocation. + */ + VMA_ALLOCATION_CREATE_STRATEGY_WORST_FIT_BIT = 0x00020000, + /** Allocation strategy that chooses first suitable free range for the + allocation. + + "First" doesn't necessarily means the one with smallest offset in memory, + but rather the one that is easiest and fastest to find. + */ + VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT = 0x00040000, + + /** Allocation strategy that tries to minimize memory usage. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT = VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT, + /** Allocation strategy that tries to minimize allocation time. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT = VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT, + /** Allocation strategy that tries to minimize memory fragmentation. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_FRAGMENTATION_BIT = VMA_ALLOCATION_CREATE_STRATEGY_WORST_FIT_BIT, + + /** A bit mask to extract only `STRATEGY` bits from entire set of flags. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MASK = + VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT | + VMA_ALLOCATION_CREATE_STRATEGY_WORST_FIT_BIT | + VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT, + + VMA_ALLOCATION_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaAllocationCreateFlagBits; +typedef VkFlags VmaAllocationCreateFlags; + +typedef struct VmaAllocationCreateInfo +{ + /// Use #VmaAllocationCreateFlagBits enum. + VmaAllocationCreateFlags flags; + /** \brief Intended usage of memory. + + You can leave #VMA_MEMORY_USAGE_UNKNOWN if you specify memory requirements in other way. \n + If `pool` is not null, this member is ignored. + */ + VmaMemoryUsage usage; + /** \brief Flags that must be set in a Memory Type chosen for an allocation. + + Leave 0 if you specify memory requirements in other way. \n + If `pool` is not null, this member is ignored.*/ + VkMemoryPropertyFlags requiredFlags; + /** \brief Flags that preferably should be set in a memory type chosen for an allocation. + + Set to 0 if no additional flags are prefered. \n + If `pool` is not null, this member is ignored. */ + VkMemoryPropertyFlags preferredFlags; + /** \brief Bitmask containing one bit set for every memory type acceptable for this allocation. + + Value 0 is equivalent to `UINT32_MAX` - it means any memory type is accepted if + it meets other requirements specified by this structure, with no further + restrictions on memory type index. \n + If `pool` is not null, this member is ignored. + */ + uint32_t memoryTypeBits; + /** \brief Pool that this allocation should be created in. + + Leave `VK_NULL_HANDLE` to allocate from default pool. If not null, members: + `usage`, `requiredFlags`, `preferredFlags`, `memoryTypeBits` are ignored. + */ + VmaPool pool; + /** \brief Custom general-purpose pointer that will be stored in #VmaAllocation, can be read as VmaAllocationInfo::pUserData and changed using vmaSetAllocationUserData(). + + If #VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT is used, it must be either + null or pointer to a null-terminated string. The string will be then copied to + internal buffer, so it doesn't need to be valid after allocation call. + */ + void* pUserData; +} VmaAllocationCreateInfo; + +/** +\brief Helps to find memoryTypeIndex, given memoryTypeBits and VmaAllocationCreateInfo. + +This algorithm tries to find a memory type that: + +- Is allowed by memoryTypeBits. +- Contains all the flags from pAllocationCreateInfo->requiredFlags. +- Matches intended usage. +- Has as many flags from pAllocationCreateInfo->preferredFlags as possible. + +\return Returns VK_ERROR_FEATURE_NOT_PRESENT if not found. Receiving such result +from this function or any other allocating function probably means that your +device doesn't support any memory type with requested features for the specific +type of resource you want to use it for. Please check parameters of your +resource, like image layout (OPTIMAL versus LINEAR) or mip level count. +*/ +VkResult vmaFindMemoryTypeIndex( + VmaAllocator allocator, + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex); + +/** +\brief Helps to find memoryTypeIndex, given VkBufferCreateInfo and VmaAllocationCreateInfo. + +It can be useful e.g. to determine value to be used as VmaPoolCreateInfo::memoryTypeIndex. +It internally creates a temporary, dummy buffer that never has memory bound. +It is just a convenience function, equivalent to calling: + +- `vkCreateBuffer` +- `vkGetBufferMemoryRequirements` +- `vmaFindMemoryTypeIndex` +- `vkDestroyBuffer` +*/ +VkResult vmaFindMemoryTypeIndexForBufferInfo( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex); + +/** +\brief Helps to find memoryTypeIndex, given VkImageCreateInfo and VmaAllocationCreateInfo. + +It can be useful e.g. to determine value to be used as VmaPoolCreateInfo::memoryTypeIndex. +It internally creates a temporary, dummy image that never has memory bound. +It is just a convenience function, equivalent to calling: + +- `vkCreateImage` +- `vkGetImageMemoryRequirements` +- `vmaFindMemoryTypeIndex` +- `vkDestroyImage` +*/ +VkResult vmaFindMemoryTypeIndexForImageInfo( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex); + +/// Flags to be passed as VmaPoolCreateInfo::flags. +typedef enum VmaPoolCreateFlagBits { + /** \brief Use this flag if you always allocate only buffers and linear images or only optimal images out of this pool and so Buffer-Image Granularity can be ignored. + + This is an optional optimization flag. + + If you always allocate using vmaCreateBuffer(), vmaCreateImage(), + vmaAllocateMemoryForBuffer(), then you don't need to use it because allocator + knows exact type of your allocations so it can handle Buffer-Image Granularity + in the optimal way. + + If you also allocate using vmaAllocateMemoryForImage() or vmaAllocateMemory(), + exact type of such allocations is not known, so allocator must be conservative + in handling Buffer-Image Granularity, which can lead to suboptimal allocation + (wasted memory). In that case, if you can make sure you always allocate only + buffers and linear images or only optimal images out of this pool, use this flag + to make allocator disregard Buffer-Image Granularity and so make allocations + faster and more optimal. + */ + VMA_POOL_CREATE_IGNORE_BUFFER_IMAGE_GRANULARITY_BIT = 0x00000002, + + /** \brief Enables alternative, linear allocation algorithm in this pool. + + Specify this flag to enable linear allocation algorithm, which always creates + new allocations after last one and doesn't reuse space from allocations freed in + between. It trades memory consumption for simplified algorithm and data + structure, which has better performance and uses less memory for metadata. + + By using this flag, you can achieve behavior of free-at-once, stack, + ring buffer, and double stack. For details, see documentation chapter + \ref linear_algorithm. + + When using this flag, you must specify VmaPoolCreateInfo::maxBlockCount == 1 (or 0 for default). + + For more details, see [Linear allocation algorithm](@ref linear_algorithm). + */ + VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT = 0x00000004, + + /** \brief Enables alternative, buddy allocation algorithm in this pool. + + It operates on a tree of blocks, each having size that is a power of two and + a half of its parent's size. Comparing to default algorithm, this one provides + faster allocation and deallocation and decreased external fragmentation, + at the expense of more memory wasted (internal fragmentation). + + For more details, see [Buddy allocation algorithm](@ref buddy_algorithm). + */ + VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT = 0x00000008, + + /** Bit mask to extract only `ALGORITHM` bits from entire set of flags. + */ + VMA_POOL_CREATE_ALGORITHM_MASK = + VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT | + VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT, + + VMA_POOL_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaPoolCreateFlagBits; +typedef VkFlags VmaPoolCreateFlags; + +/** \brief Describes parameter of created #VmaPool. +*/ +typedef struct VmaPoolCreateInfo { + /** \brief Vulkan memory type index to allocate this pool from. + */ + uint32_t memoryTypeIndex; + /** \brief Use combination of #VmaPoolCreateFlagBits. + */ + VmaPoolCreateFlags flags; + /** \brief Size of a single `VkDeviceMemory` block to be allocated as part of this pool, in bytes. Optional. + + Specify nonzero to set explicit, constant size of memory blocks used by this + pool. + + Leave 0 to use default and let the library manage block sizes automatically. + Sizes of particular blocks may vary. + */ + VkDeviceSize blockSize; + /** \brief Minimum number of blocks to be always allocated in this pool, even if they stay empty. + + Set to 0 to have no preallocated blocks and allow the pool be completely empty. + */ + size_t minBlockCount; + /** \brief Maximum number of blocks that can be allocated in this pool. Optional. + + Set to 0 to use default, which is `SIZE_MAX`, which means no limit. + + Set to same value as VmaPoolCreateInfo::minBlockCount to have fixed amount of memory allocated + throughout whole lifetime of this pool. + */ + size_t maxBlockCount; + /** \brief Maximum number of additional frames that are in use at the same time as current frame. + + This value is used only when you make allocations with + #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag. Such allocation cannot become + lost if allocation.lastUseFrameIndex >= allocator.currentFrameIndex - frameInUseCount. + + For example, if you double-buffer your command buffers, so resources used for + rendering in previous frame may still be in use by the GPU at the moment you + allocate resources needed for the current frame, set this value to 1. + + If you want to allow any allocations other than used in the current frame to + become lost, set this value to 0. + */ + uint32_t frameInUseCount; +} VmaPoolCreateInfo; + +/** \brief Describes parameter of existing #VmaPool. +*/ +typedef struct VmaPoolStats { + /** \brief Total amount of `VkDeviceMemory` allocated from Vulkan for this pool, in bytes. + */ + VkDeviceSize size; + /** \brief Total number of bytes in the pool not used by any #VmaAllocation. + */ + VkDeviceSize unusedSize; + /** \brief Number of #VmaAllocation objects created from this pool that were not destroyed or lost. + */ + size_t allocationCount; + /** \brief Number of continuous memory ranges in the pool not used by any #VmaAllocation. + */ + size_t unusedRangeCount; + /** \brief Size of the largest continuous free memory region available for new allocation. + + Making a new allocation of that size is not guaranteed to succeed because of + possible additional margin required to respect alignment and buffer/image + granularity. + */ + VkDeviceSize unusedRangeSizeMax; + /** \brief Number of `VkDeviceMemory` blocks allocated for this pool. + */ + size_t blockCount; +} VmaPoolStats; + +/** \brief Allocates Vulkan device memory and creates #VmaPool object. + +@param allocator Allocator object. +@param pCreateInfo Parameters of pool to create. +@param[out] pPool Handle to created pool. +*/ +VkResult vmaCreatePool( + VmaAllocator allocator, + const VmaPoolCreateInfo* pCreateInfo, + VmaPool* pPool); + +/** \brief Destroys #VmaPool object and frees Vulkan device memory. +*/ +void vmaDestroyPool( + VmaAllocator allocator, + VmaPool pool); + +/** \brief Retrieves statistics of existing #VmaPool object. + +@param allocator Allocator object. +@param pool Pool object. +@param[out] pPoolStats Statistics of specified pool. +*/ +void vmaGetPoolStats( + VmaAllocator allocator, + VmaPool pool, + VmaPoolStats* pPoolStats); + +/** \brief Marks all allocations in given pool as lost if they are not used in current frame or VmaPoolCreateInfo::frameInUseCount back from now. + +@param allocator Allocator object. +@param pool Pool. +@param[out] pLostAllocationCount Number of allocations marked as lost. Optional - pass null if you don't need this information. +*/ +void vmaMakePoolAllocationsLost( + VmaAllocator allocator, + VmaPool pool, + size_t* pLostAllocationCount); + +/** \brief Checks magic number in margins around all allocations in given memory pool in search for corruptions. + +Corruption detection is enabled only when `VMA_DEBUG_DETECT_CORRUPTION` macro is defined to nonzero, +`VMA_DEBUG_MARGIN` is defined to nonzero and the pool is created in memory type that is +`HOST_VISIBLE` and `HOST_COHERENT`. For more information, see [Corruption detection](@ref debugging_memory_usage_corruption_detection). + +Possible return values: + +- `VK_ERROR_FEATURE_NOT_PRESENT` - corruption detection is not enabled for specified pool. +- `VK_SUCCESS` - corruption detection has been performed and succeeded. +- `VK_ERROR_VALIDATION_FAILED_EXT` - corruption detection has been performed and found memory corruptions around one of the allocations. + `VMA_ASSERT` is also fired in that case. +- Other value: Error returned by Vulkan, e.g. memory mapping failure. +*/ +VkResult vmaCheckPoolCorruption(VmaAllocator allocator, VmaPool pool); + +/** \struct VmaAllocation +\brief Represents single memory allocation. + +It may be either dedicated block of `VkDeviceMemory` or a specific region of a bigger block of this type +plus unique offset. + +There are multiple ways to create such object. +You need to fill structure VmaAllocationCreateInfo. +For more information see [Choosing memory type](@ref choosing_memory_type). + +Although the library provides convenience functions that create Vulkan buffer or image, +allocate memory for it and bind them together, +binding of the allocation to a buffer or an image is out of scope of the allocation itself. +Allocation object can exist without buffer/image bound, +binding can be done manually by the user, and destruction of it can be done +independently of destruction of the allocation. + +The object also remembers its size and some other information. +To retrieve this information, use function vmaGetAllocationInfo() and inspect +returned structure VmaAllocationInfo. + +Some kinds allocations can be in lost state. +For more information, see [Lost allocations](@ref lost_allocations). +*/ +VK_DEFINE_HANDLE(VmaAllocation) + +/** \brief Parameters of #VmaAllocation objects, that can be retrieved using function vmaGetAllocationInfo(). +*/ +typedef struct VmaAllocationInfo { + /** \brief Memory type index that this allocation was allocated from. + + It never changes. + */ + uint32_t memoryType; + /** \brief Handle to Vulkan memory object. + + Same memory object can be shared by multiple allocations. + + It can change after call to vmaDefragment() if this allocation is passed to the function, or if allocation is lost. + + If the allocation is lost, it is equal to `VK_NULL_HANDLE`. + */ + VkDeviceMemory deviceMemory; + /** \brief Offset into deviceMemory object to the beginning of this allocation, in bytes. (deviceMemory, offset) pair is unique to this allocation. + + It can change after call to vmaDefragment() if this allocation is passed to the function, or if allocation is lost. + */ + VkDeviceSize offset; + /** \brief Size of this allocation, in bytes. + + It never changes, unless allocation is lost. + */ + VkDeviceSize size; + /** \brief Pointer to the beginning of this allocation as mapped data. + + If the allocation hasn't been mapped using vmaMapMemory() and hasn't been + created with #VMA_ALLOCATION_CREATE_MAPPED_BIT flag, this value null. + + It can change after call to vmaMapMemory(), vmaUnmapMemory(). + It can also change after call to vmaDefragment() if this allocation is passed to the function. + */ + void* pMappedData; + /** \brief Custom general-purpose pointer that was passed as VmaAllocationCreateInfo::pUserData or set using vmaSetAllocationUserData(). + + It can change after call to vmaSetAllocationUserData() for this allocation. + */ + void* pUserData; +} VmaAllocationInfo; + +/** \brief General purpose memory allocation. + +@param[out] pAllocation Handle to allocated memory. +@param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). + +You should free the memory using vmaFreeMemory() or vmaFreeMemoryPages(). + +It is recommended to use vmaAllocateMemoryForBuffer(), vmaAllocateMemoryForImage(), +vmaCreateBuffer(), vmaCreateImage() instead whenever possible. +*/ +VkResult vmaAllocateMemory( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo); + +/** \brief General purpose memory allocation for multiple allocation objects at once. + +@param allocator Allocator object. +@param pVkMemoryRequirements Memory requirements for each allocation. +@param pCreateInfo Creation parameters for each alloction. +@param allocationCount Number of allocations to make. +@param[out] pAllocations Pointer to array that will be filled with handles to created allocations. +@param[out] pAllocationInfo Optional. Pointer to array that will be filled with parameters of created allocations. + +You should free the memory using vmaFreeMemory() or vmaFreeMemoryPages(). + +Word "pages" is just a suggestion to use this function to allocate pieces of memory needed for sparse binding. +It is just a general purpose allocation function able to make multiple allocations at once. +It may be internally optimized to be more efficient than calling vmaAllocateMemory() `allocationCount` times. + +All allocations are made using same parameters. All of them are created out of the same memory pool and type. +If any allocation fails, all allocations already made within this function call are also freed, so that when +returned result is not `VK_SUCCESS`, `pAllocation` array is always entirely filled with `VK_NULL_HANDLE`. +*/ +VkResult vmaAllocateMemoryPages( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + size_t allocationCount, + VmaAllocation* pAllocations, + VmaAllocationInfo* pAllocationInfo); + +/** +@param[out] pAllocation Handle to allocated memory. +@param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). + +You should free the memory using vmaFreeMemory(). +*/ +VkResult vmaAllocateMemoryForBuffer( + VmaAllocator allocator, + VkBuffer buffer, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo); + +/// Function similar to vmaAllocateMemoryForBuffer(). +VkResult vmaAllocateMemoryForImage( + VmaAllocator allocator, + VkImage image, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo); + +/** \brief Frees memory previously allocated using vmaAllocateMemory(), vmaAllocateMemoryForBuffer(), or vmaAllocateMemoryForImage(). + +Passing `VK_NULL_HANDLE` as `allocation` is valid. Such function call is just skipped. +*/ +void vmaFreeMemory( + VmaAllocator allocator, + VmaAllocation allocation); + +/** \brief Frees memory and destroys multiple allocations. + +Word "pages" is just a suggestion to use this function to free pieces of memory used for sparse binding. +It is just a general purpose function to free memory and destroy allocations made using e.g. vmaAllocateMemory(), +vmaAllocateMemoryPages() and other functions. +It may be internally optimized to be more efficient than calling vmaFreeMemory() `allocationCount` times. + +Allocations in `pAllocations` array can come from any memory pools and types. +Passing `VK_NULL_HANDLE` as elements of `pAllocations` array is valid. Such entries are just skipped. +*/ +void vmaFreeMemoryPages( + VmaAllocator allocator, + size_t allocationCount, + VmaAllocation* pAllocations); + +/** \brief Tries to resize an allocation in place, if there is enough free memory after it. + +Tries to change allocation's size without moving or reallocating it. +You can both shrink and grow allocation size. +When growing, it succeeds only when the allocation belongs to a memory block with enough +free space after it. + +Returns `VK_SUCCESS` if allocation's size has been successfully changed. +Returns `VK_ERROR_OUT_OF_POOL_MEMORY` if allocation's size could not be changed. + +After successful call to this function, VmaAllocationInfo::size of this allocation changes. +All other parameters stay the same: memory pool and type, alignment, offset, mapped pointer. + +- Calling this function on allocation that is in lost state fails with result `VK_ERROR_VALIDATION_FAILED_EXT`. +- Calling this function with `newSize` same as current allocation size does nothing and returns `VK_SUCCESS`. +- Resizing dedicated allocations, as well as allocations created in pools that use linear + or buddy algorithm, is not supported. + The function returns `VK_ERROR_FEATURE_NOT_PRESENT` in such cases. + Support may be added in the future. +*/ +VkResult vmaResizeAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize newSize); + +/** \brief Returns current information about specified allocation and atomically marks it as used in current frame. + +Current paramters of given allocation are returned in `pAllocationInfo`. + +This function also atomically "touches" allocation - marks it as used in current frame, +just like vmaTouchAllocation(). +If the allocation is in lost state, `pAllocationInfo->deviceMemory == VK_NULL_HANDLE`. + +Although this function uses atomics and doesn't lock any mutex, so it should be quite efficient, +you can avoid calling it too often. + +- You can retrieve same VmaAllocationInfo structure while creating your resource, from function + vmaCreateBuffer(), vmaCreateImage(). You can remember it if you are sure parameters don't change + (e.g. due to defragmentation or allocation becoming lost). +- If you just want to check if allocation is not lost, vmaTouchAllocation() will work faster. +*/ +void vmaGetAllocationInfo( + VmaAllocator allocator, + VmaAllocation allocation, + VmaAllocationInfo* pAllocationInfo); + +/** \brief Returns `VK_TRUE` if allocation is not lost and atomically marks it as used in current frame. + +If the allocation has been created with #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag, +this function returns `VK_TRUE` if it's not in lost state, so it can still be used. +It then also atomically "touches" the allocation - marks it as used in current frame, +so that you can be sure it won't become lost in current frame or next `frameInUseCount` frames. + +If the allocation is in lost state, the function returns `VK_FALSE`. +Memory of such allocation, as well as buffer or image bound to it, should not be used. +Lost allocation and the buffer/image still need to be destroyed. + +If the allocation has been created without #VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag, +this function always returns `VK_TRUE`. +*/ +VkBool32 vmaTouchAllocation( + VmaAllocator allocator, + VmaAllocation allocation); + +/** \brief Sets pUserData in given allocation to new value. + +If the allocation was created with VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT, +pUserData must be either null, or pointer to a null-terminated string. The function +makes local copy of the string and sets it as allocation's `pUserData`. String +passed as pUserData doesn't need to be valid for whole lifetime of the allocation - +you can free it after this call. String previously pointed by allocation's +pUserData is freed from memory. + +If the flag was not used, the value of pointer `pUserData` is just copied to +allocation's `pUserData`. It is opaque, so you can use it however you want - e.g. +as a pointer, ordinal number or some handle to you own data. +*/ +void vmaSetAllocationUserData( + VmaAllocator allocator, + VmaAllocation allocation, + void* pUserData); + +/** \brief Creates new allocation that is in lost state from the beginning. + +It can be useful if you need a dummy, non-null allocation. + +You still need to destroy created object using vmaFreeMemory(). + +Returned allocation is not tied to any specific memory pool or memory type and +not bound to any image or buffer. It has size = 0. It cannot be turned into +a real, non-empty allocation. +*/ +void vmaCreateLostAllocation( + VmaAllocator allocator, + VmaAllocation* pAllocation); + +/** \brief Maps memory represented by given allocation and returns pointer to it. + +Maps memory represented by given allocation to make it accessible to CPU code. +When succeeded, `*ppData` contains pointer to first byte of this memory. +If the allocation is part of bigger `VkDeviceMemory` block, the pointer is +correctly offseted to the beginning of region assigned to this particular +allocation. + +Mapping is internally reference-counted and synchronized, so despite raw Vulkan +function `vkMapMemory()` cannot be used to map same block of `VkDeviceMemory` +multiple times simultaneously, it is safe to call this function on allocations +assigned to the same memory block. Actual Vulkan memory will be mapped on first +mapping and unmapped on last unmapping. + +If the function succeeded, you must call vmaUnmapMemory() to unmap the +allocation when mapping is no longer needed or before freeing the allocation, at +the latest. + +It also safe to call this function multiple times on the same allocation. You +must call vmaUnmapMemory() same number of times as you called vmaMapMemory(). + +It is also safe to call this function on allocation created with +#VMA_ALLOCATION_CREATE_MAPPED_BIT flag. Its memory stays mapped all the time. +You must still call vmaUnmapMemory() same number of times as you called +vmaMapMemory(). You must not call vmaUnmapMemory() additional time to free the +"0-th" mapping made automatically due to #VMA_ALLOCATION_CREATE_MAPPED_BIT flag. + +This function fails when used on allocation made in memory type that is not +`HOST_VISIBLE`. + +This function always fails when called for allocation that was created with +#VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT flag. Such allocations cannot be +mapped. +*/ +VkResult vmaMapMemory( + VmaAllocator allocator, + VmaAllocation allocation, + void** ppData); + +/** \brief Unmaps memory represented by given allocation, mapped previously using vmaMapMemory(). + +For details, see description of vmaMapMemory(). +*/ +void vmaUnmapMemory( + VmaAllocator allocator, + VmaAllocation allocation); + +/** \brief Flushes memory of given allocation. + +Calls `vkFlushMappedMemoryRanges()` for memory associated with given range of given allocation. + +- `offset` must be relative to the beginning of allocation. +- `size` can be `VK_WHOLE_SIZE`. It means all memory from `offset` the the end of given allocation. +- `offset` and `size` don't have to be aligned. + They are internally rounded down/up to multiply of `nonCoherentAtomSize`. +- If `size` is 0, this call is ignored. +- If memory type that the `allocation` belongs to is not `HOST_VISIBLE` or it is `HOST_COHERENT`, + this call is ignored. +*/ +void vmaFlushAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size); + +/** \brief Invalidates memory of given allocation. + +Calls `vkInvalidateMappedMemoryRanges()` for memory associated with given range of given allocation. + +- `offset` must be relative to the beginning of allocation. +- `size` can be `VK_WHOLE_SIZE`. It means all memory from `offset` the the end of given allocation. +- `offset` and `size` don't have to be aligned. + They are internally rounded down/up to multiply of `nonCoherentAtomSize`. +- If `size` is 0, this call is ignored. +- If memory type that the `allocation` belongs to is not `HOST_VISIBLE` or it is `HOST_COHERENT`, + this call is ignored. +*/ +void vmaInvalidateAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size); + +/** \brief Checks magic number in margins around all allocations in given memory types (in both default and custom pools) in search for corruptions. + +@param memoryTypeBits Bit mask, where each bit set means that a memory type with that index should be checked. + +Corruption detection is enabled only when `VMA_DEBUG_DETECT_CORRUPTION` macro is defined to nonzero, +`VMA_DEBUG_MARGIN` is defined to nonzero and only for memory types that are +`HOST_VISIBLE` and `HOST_COHERENT`. For more information, see [Corruption detection](@ref debugging_memory_usage_corruption_detection). + +Possible return values: + +- `VK_ERROR_FEATURE_NOT_PRESENT` - corruption detection is not enabled for any of specified memory types. +- `VK_SUCCESS` - corruption detection has been performed and succeeded. +- `VK_ERROR_VALIDATION_FAILED_EXT` - corruption detection has been performed and found memory corruptions around one of the allocations. + `VMA_ASSERT` is also fired in that case. +- Other value: Error returned by Vulkan, e.g. memory mapping failure. +*/ +VkResult vmaCheckCorruption(VmaAllocator allocator, uint32_t memoryTypeBits); + +/** \struct VmaDefragmentationContext +\brief Represents Opaque object that represents started defragmentation process. + +Fill structure #VmaDefragmentationInfo2 and call function vmaDefragmentationBegin() to create it. +Call function vmaDefragmentationEnd() to destroy it. +*/ +VK_DEFINE_HANDLE(VmaDefragmentationContext) + +/// Flags to be used in vmaDefragmentationBegin(). None at the moment. Reserved for future use. +typedef enum VmaDefragmentationFlagBits { + VMA_DEFRAGMENTATION_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaDefragmentationFlagBits; +typedef VkFlags VmaDefragmentationFlags; + +/** \brief Parameters for defragmentation. + +To be used with function vmaDefragmentationBegin(). +*/ +typedef struct VmaDefragmentationInfo2 { + /** \brief Reserved for future use. Should be 0. + */ + VmaDefragmentationFlags flags; + /** \brief Number of allocations in `pAllocations` array. + */ + uint32_t allocationCount; + /** \brief Pointer to array of allocations that can be defragmented. + + The array should have `allocationCount` elements. + The array should not contain nulls. + Elements in the array should be unique - same allocation cannot occur twice. + It is safe to pass allocations that are in the lost state - they are ignored. + All allocations not present in this array are considered non-moveable during this defragmentation. + */ + VmaAllocation* pAllocations; + /** \brief Optional, output. Pointer to array that will be filled with information whether the allocation at certain index has been changed during defragmentation. + + The array should have `allocationCount` elements. + You can pass null if you are not interested in this information. + */ + VkBool32* pAllocationsChanged; + /** \brief Numer of pools in `pPools` array. + */ + uint32_t poolCount; + /** \brief Either null or pointer to array of pools to be defragmented. + + All the allocations in the specified pools can be moved during defragmentation + and there is no way to check if they were really moved as in `pAllocationsChanged`, + so you must query all the allocations in all these pools for new `VkDeviceMemory` + and offset using vmaGetAllocationInfo() if you might need to recreate buffers + and images bound to them. + + The array should have `poolCount` elements. + The array should not contain nulls. + Elements in the array should be unique - same pool cannot occur twice. + + Using this array is equivalent to specifying all allocations from the pools in `pAllocations`. + It might be more efficient. + */ + VmaPool* pPools; + /** \brief Maximum total numbers of bytes that can be copied while moving allocations to different places using transfers on CPU side, like `memcpy()`, `memmove()`. + + `VK_WHOLE_SIZE` means no limit. + */ + VkDeviceSize maxCpuBytesToMove; + /** \brief Maximum number of allocations that can be moved to a different place using transfers on CPU side, like `memcpy()`, `memmove()`. + + `UINT32_MAX` means no limit. + */ + uint32_t maxCpuAllocationsToMove; + /** \brief Maximum total numbers of bytes that can be copied while moving allocations to different places using transfers on GPU side, posted to `commandBuffer`. + + `VK_WHOLE_SIZE` means no limit. + */ + VkDeviceSize maxGpuBytesToMove; + /** \brief Maximum number of allocations that can be moved to a different place using transfers on GPU side, posted to `commandBuffer`. + + `UINT32_MAX` means no limit. + */ + uint32_t maxGpuAllocationsToMove; + /** \brief Optional. Command buffer where GPU copy commands will be posted. + + If not null, it must be a valid command buffer handle that supports Transfer queue type. + It must be in the recording state and outside of a render pass instance. + You need to submit it and make sure it finished execution before calling vmaDefragmentationEnd(). + + Passing null means that only CPU defragmentation will be performed. + */ + VkCommandBuffer commandBuffer; +} VmaDefragmentationInfo2; + +/** \brief Deprecated. Optional configuration parameters to be passed to function vmaDefragment(). + +\deprecated This is a part of the old interface. It is recommended to use structure #VmaDefragmentationInfo2 and function vmaDefragmentationBegin() instead. +*/ +typedef struct VmaDefragmentationInfo { + /** \brief Maximum total numbers of bytes that can be copied while moving allocations to different places. + + Default is `VK_WHOLE_SIZE`, which means no limit. + */ + VkDeviceSize maxBytesToMove; + /** \brief Maximum number of allocations that can be moved to different place. + + Default is `UINT32_MAX`, which means no limit. + */ + uint32_t maxAllocationsToMove; +} VmaDefragmentationInfo; + +/** \brief Statistics returned by function vmaDefragment(). */ +typedef struct VmaDefragmentationStats { + /// Total number of bytes that have been copied while moving allocations to different places. + VkDeviceSize bytesMoved; + /// Total number of bytes that have been released to the system by freeing empty `VkDeviceMemory` objects. + VkDeviceSize bytesFreed; + /// Number of allocations that have been moved to different places. + uint32_t allocationsMoved; + /// Number of empty `VkDeviceMemory` objects that have been released to the system. + uint32_t deviceMemoryBlocksFreed; +} VmaDefragmentationStats; + +/** \brief Begins defragmentation process. + +@param allocator Allocator object. +@param pInfo Structure filled with parameters of defragmentation. +@param[out] pStats Optional. Statistics of defragmentation. You can pass null if you are not interested in this information. +@param[out] pContext Context object that must be passed to vmaDefragmentationEnd() to finish defragmentation. +@return `VK_SUCCESS` and `*pContext == null` if defragmentation finished within this function call. `VK_NOT_READY` and `*pContext != null` if defragmentation has been started and you need to call vmaDefragmentationEnd() to finish it. Negative value in case of error. + +Use this function instead of old, deprecated vmaDefragment(). + +Warning! Between the call to vmaDefragmentationBegin() and vmaDefragmentationEnd(): + +- You should not use any of allocations passed as `pInfo->pAllocations` or + any allocations that belong to pools passed as `pInfo->pPools`, + including calling vmaGetAllocationInfo(), vmaTouchAllocation(), or access + their data. +- Some mutexes protecting internal data structures may be locked, so trying to + make or free any allocations, bind buffers or images, map memory, or launch + another simultaneous defragmentation in between may cause stall (when done on + another thread) or deadlock (when done on the same thread), unless you are + 100% sure that defragmented allocations are in different pools. +- Information returned via `pStats` and `pInfo->pAllocationsChanged` are undefined. + They become valid after call to vmaDefragmentationEnd(). +- If `pInfo->commandBuffer` is not null, you must submit that command buffer + and make sure it finished execution before calling vmaDefragmentationEnd(). +*/ +VkResult vmaDefragmentationBegin( + VmaAllocator allocator, + const VmaDefragmentationInfo2* pInfo, + VmaDefragmentationStats* pStats, + VmaDefragmentationContext *pContext); + +/** \brief Ends defragmentation process. + +Use this function to finish defragmentation started by vmaDefragmentationBegin(). +It is safe to pass `context == null`. The function then does nothing. +*/ +VkResult vmaDefragmentationEnd( + VmaAllocator allocator, + VmaDefragmentationContext context); + +/** \brief Deprecated. Compacts memory by moving allocations. + +@param pAllocations Array of allocations that can be moved during this compation. +@param allocationCount Number of elements in pAllocations and pAllocationsChanged arrays. +@param[out] pAllocationsChanged Array of boolean values that will indicate whether matching allocation in pAllocations array has been moved. This parameter is optional. Pass null if you don't need this information. +@param pDefragmentationInfo Configuration parameters. Optional - pass null to use default values. +@param[out] pDefragmentationStats Statistics returned by the function. Optional - pass null if you don't need this information. +@return `VK_SUCCESS` if completed, negative error code in case of error. + +\deprecated This is a part of the old interface. It is recommended to use structure #VmaDefragmentationInfo2 and function vmaDefragmentationBegin() instead. + +This function works by moving allocations to different places (different +`VkDeviceMemory` objects and/or different offsets) in order to optimize memory +usage. Only allocations that are in `pAllocations` array can be moved. All other +allocations are considered nonmovable in this call. Basic rules: + +- Only allocations made in memory types that have + `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` and `VK_MEMORY_PROPERTY_HOST_COHERENT_BIT` + flags can be compacted. You may pass other allocations but it makes no sense - + these will never be moved. +- Custom pools created with #VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT or + #VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT flag are not defragmented. Allocations + passed to this function that come from such pools are ignored. +- Allocations created with #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT or + created as dedicated allocations for any other reason are also ignored. +- Both allocations made with or without #VMA_ALLOCATION_CREATE_MAPPED_BIT + flag can be compacted. If not persistently mapped, memory will be mapped + temporarily inside this function if needed. +- You must not pass same #VmaAllocation object multiple times in `pAllocations` array. + +The function also frees empty `VkDeviceMemory` blocks. + +Warning: This function may be time-consuming, so you shouldn't call it too often +(like after every resource creation/destruction). +You can call it on special occasions (like when reloading a game level or +when you just destroyed a lot of objects). Calling it every frame may be OK, but +you should measure that on your platform. + +For more information, see [Defragmentation](@ref defragmentation) chapter. +*/ +VkResult vmaDefragment( + VmaAllocator allocator, + VmaAllocation* pAllocations, + size_t allocationCount, + VkBool32* pAllocationsChanged, + const VmaDefragmentationInfo *pDefragmentationInfo, + VmaDefragmentationStats* pDefragmentationStats); + +/** \brief Binds buffer to allocation. + +Binds specified buffer to region of memory represented by specified allocation. +Gets `VkDeviceMemory` handle and offset from the allocation. +If you want to create a buffer, allocate memory for it and bind them together separately, +you should use this function for binding instead of standard `vkBindBufferMemory()`, +because it ensures proper synchronization so that when a `VkDeviceMemory` object is used by multiple +allocations, calls to `vkBind*Memory()` or `vkMapMemory()` won't happen from multiple threads simultaneously +(which is illegal in Vulkan). + +It is recommended to use function vmaCreateBuffer() instead of this one. +*/ +VkResult vmaBindBufferMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkBuffer buffer); + +/** \brief Binds image to allocation. + +Binds specified image to region of memory represented by specified allocation. +Gets `VkDeviceMemory` handle and offset from the allocation. +If you want to create an image, allocate memory for it and bind them together separately, +you should use this function for binding instead of standard `vkBindImageMemory()`, +because it ensures proper synchronization so that when a `VkDeviceMemory` object is used by multiple +allocations, calls to `vkBind*Memory()` or `vkMapMemory()` won't happen from multiple threads simultaneously +(which is illegal in Vulkan). + +It is recommended to use function vmaCreateImage() instead of this one. +*/ +VkResult vmaBindImageMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkImage image); + +/** +@param[out] pBuffer Buffer that was created. +@param[out] pAllocation Allocation that was created. +@param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). + +This function automatically: + +-# Creates buffer. +-# Allocates appropriate memory for it. +-# Binds the buffer with the memory. + +If any of these operations fail, buffer and allocation are not created, +returned value is negative error code, *pBuffer and *pAllocation are null. + +If the function succeeded, you must destroy both buffer and allocation when you +no longer need them using either convenience function vmaDestroyBuffer() or +separately, using `vkDestroyBuffer()` and vmaFreeMemory(). + +If VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT flag was used, +VK_KHR_dedicated_allocation extension is used internally to query driver whether +it requires or prefers the new buffer to have dedicated allocation. If yes, +and if dedicated allocation is possible (VmaAllocationCreateInfo::pool is null +and VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT is not used), it creates dedicated +allocation for this buffer, just like when using +VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +*/ +VkResult vmaCreateBuffer( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkBuffer* pBuffer, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo); + +/** \brief Destroys Vulkan buffer and frees allocated memory. + +This is just a convenience function equivalent to: + +\code +vkDestroyBuffer(device, buffer, allocationCallbacks); +vmaFreeMemory(allocator, allocation); +\endcode + +It it safe to pass null as buffer and/or allocation. +*/ +void vmaDestroyBuffer( + VmaAllocator allocator, + VkBuffer buffer, + VmaAllocation allocation); + +/// Function similar to vmaCreateBuffer(). +VkResult vmaCreateImage( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkImage* pImage, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo); + +/** \brief Destroys Vulkan image and frees allocated memory. + +This is just a convenience function equivalent to: + +\code +vkDestroyImage(device, image, allocationCallbacks); +vmaFreeMemory(allocator, allocation); +\endcode + +It it safe to pass null as image and/or allocation. +*/ +void vmaDestroyImage( + VmaAllocator allocator, + VkImage image, + VmaAllocation allocation); + +#ifdef __cplusplus +} +#endif + +#endif // AMD_VULKAN_MEMORY_ALLOCATOR_H + +// For Visual Studio IntelliSense. +#if defined(__cplusplus) && defined(__INTELLISENSE__) +#define VMA_IMPLEMENTATION +#endif + +#ifdef VMA_IMPLEMENTATION +#undef VMA_IMPLEMENTATION + +#include <cstdint> +#include <cstdlib> +#include <cstring> + +/******************************************************************************* +CONFIGURATION SECTION + +Define some of these macros before each #include of this header or change them +here if you need other then default behavior depending on your environment. +*/ + +/* +Define this macro to 1 to make the library fetch pointers to Vulkan functions +internally, like: + + vulkanFunctions.vkAllocateMemory = &vkAllocateMemory; + +Define to 0 if you are going to provide you own pointers to Vulkan functions via +VmaAllocatorCreateInfo::pVulkanFunctions. +*/ +#if !defined(VMA_STATIC_VULKAN_FUNCTIONS) && !defined(VK_NO_PROTOTYPES) +#define VMA_STATIC_VULKAN_FUNCTIONS 1 +#endif + +// Define this macro to 1 to make the library use STL containers instead of its own implementation. +//#define VMA_USE_STL_CONTAINERS 1 + +/* Set this macro to 1 to make the library including and using STL containers: +std::pair, std::vector, std::list, std::unordered_map. + +Set it to 0 or undefined to make the library using its own implementation of +the containers. +*/ +#if VMA_USE_STL_CONTAINERS + #define VMA_USE_STL_VECTOR 1 + #define VMA_USE_STL_UNORDERED_MAP 1 + #define VMA_USE_STL_LIST 1 +#endif + +#ifndef VMA_USE_STL_SHARED_MUTEX + // Minimum Visual Studio 2015 Update 2 + #if defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 190023918 + #define VMA_USE_STL_SHARED_MUTEX 1 + #endif +#endif + +#if VMA_USE_STL_VECTOR + #include <vector> +#endif + +#if VMA_USE_STL_UNORDERED_MAP + #include <unordered_map> +#endif + +#if VMA_USE_STL_LIST + #include <list> +#endif + +/* +Following headers are used in this CONFIGURATION section only, so feel free to +remove them if not needed. +*/ +#include <cassert> // for assert +#include <algorithm> // for min, max +#include <mutex> +#include <atomic> // for std::atomic + +#ifndef VMA_NULL + // Value used as null pointer. Define it to e.g.: nullptr, NULL, 0, (void*)0. + #define VMA_NULL nullptr +#endif + +#if defined(__ANDROID_API__) && (__ANDROID_API__ < 16) +#include <cstdlib> +void *aligned_alloc(size_t alignment, size_t size) +{ + // alignment must be >= sizeof(void*) + if(alignment < sizeof(void*)) + { + alignment = sizeof(void*); + } + + return memalign(alignment, size); +} +#elif defined(__APPLE__) || defined(__ANDROID__) +#include <cstdlib> +void *aligned_alloc(size_t alignment, size_t size) +{ + // alignment must be >= sizeof(void*) + if(alignment < sizeof(void*)) + { + alignment = sizeof(void*); + } + + void *pointer; + if(posix_memalign(&pointer, alignment, size) == 0) + return pointer; + return VMA_NULL; +} +#endif + +// If your compiler is not compatible with C++11 and definition of +// aligned_alloc() function is missing, uncommeting following line may help: + +//#include <malloc.h> + +// Normal assert to check for programmer's errors, especially in Debug configuration. +#ifndef VMA_ASSERT + #ifdef _DEBUG + #define VMA_ASSERT(expr) assert(expr) + #else + #define VMA_ASSERT(expr) + #endif +#endif + +// Assert that will be called very often, like inside data structures e.g. operator[]. +// Making it non-empty can make program slow. +#ifndef VMA_HEAVY_ASSERT + #ifdef _DEBUG + #define VMA_HEAVY_ASSERT(expr) //VMA_ASSERT(expr) + #else + #define VMA_HEAVY_ASSERT(expr) + #endif +#endif + +#ifndef VMA_ALIGN_OF + #define VMA_ALIGN_OF(type) (__alignof(type)) +#endif + +#ifndef VMA_SYSTEM_ALIGNED_MALLOC + #if defined(_WIN32) + #define VMA_SYSTEM_ALIGNED_MALLOC(size, alignment) (_aligned_malloc((size), (alignment))) + #else + #define VMA_SYSTEM_ALIGNED_MALLOC(size, alignment) (aligned_alloc((alignment), (size) )) + #endif +#endif + +#ifndef VMA_SYSTEM_FREE + #if defined(_WIN32) + #define VMA_SYSTEM_FREE(ptr) _aligned_free(ptr) + #else + #define VMA_SYSTEM_FREE(ptr) free(ptr) + #endif +#endif + +#ifndef VMA_MIN + #define VMA_MIN(v1, v2) (std::min((v1), (v2))) +#endif + +#ifndef VMA_MAX + #define VMA_MAX(v1, v2) (std::max((v1), (v2))) +#endif + +#ifndef VMA_SWAP + #define VMA_SWAP(v1, v2) std::swap((v1), (v2)) +#endif + +#ifndef VMA_SORT + #define VMA_SORT(beg, end, cmp) std::sort(beg, end, cmp) +#endif + +#ifndef VMA_DEBUG_LOG + #define VMA_DEBUG_LOG(format, ...) + /* + #define VMA_DEBUG_LOG(format, ...) do { \ + printf(format, __VA_ARGS__); \ + printf("\n"); \ + } while(false) + */ +#endif + +// Define this macro to 1 to enable functions: vmaBuildStatsString, vmaFreeStatsString. +#if VMA_STATS_STRING_ENABLED + static inline void VmaUint32ToStr(char* outStr, size_t strLen, uint32_t num) + { + snprintf(outStr, strLen, "%u", static_cast<unsigned int>(num)); + } + static inline void VmaUint64ToStr(char* outStr, size_t strLen, uint64_t num) + { + snprintf(outStr, strLen, "%llu", static_cast<unsigned long long>(num)); + } + static inline void VmaPtrToStr(char* outStr, size_t strLen, const void* ptr) + { + snprintf(outStr, strLen, "%p", ptr); + } +#endif + +#ifndef VMA_MUTEX + class VmaMutex + { + public: + void Lock() { m_Mutex.lock(); } + void Unlock() { m_Mutex.unlock(); } + private: + std::mutex m_Mutex; + }; + #define VMA_MUTEX VmaMutex +#endif + +// Read-write mutex, where "read" is shared access, "write" is exclusive access. +#ifndef VMA_RW_MUTEX + #if VMA_USE_STL_SHARED_MUTEX + // Use std::shared_mutex from C++17. + #include <shared_mutex> + class VmaRWMutex + { + public: + void LockRead() { m_Mutex.lock_shared(); } + void UnlockRead() { m_Mutex.unlock_shared(); } + void LockWrite() { m_Mutex.lock(); } + void UnlockWrite() { m_Mutex.unlock(); } + private: + std::shared_mutex m_Mutex; + }; + #define VMA_RW_MUTEX VmaRWMutex + #elif defined(_WIN32) && !defined(__MINGW32__) + // Use SRWLOCK from WinAPI. + class VmaRWMutex + { + public: + VmaRWMutex() { InitializeSRWLock(&m_Lock); } + void LockRead() { AcquireSRWLockShared(&m_Lock); } + void UnlockRead() { ReleaseSRWLockShared(&m_Lock); } + void LockWrite() { AcquireSRWLockExclusive(&m_Lock); } + void UnlockWrite() { ReleaseSRWLockExclusive(&m_Lock); } + private: + SRWLOCK m_Lock; + }; + #define VMA_RW_MUTEX VmaRWMutex + #else + // Less efficient fallback: Use normal mutex. + class VmaRWMutex + { + public: + void LockRead() { m_Mutex.Lock(); } + void UnlockRead() { m_Mutex.Unlock(); } + void LockWrite() { m_Mutex.Lock(); } + void UnlockWrite() { m_Mutex.Unlock(); } + private: + VMA_MUTEX m_Mutex; + }; + #define VMA_RW_MUTEX VmaRWMutex + #endif // #if VMA_USE_STL_SHARED_MUTEX +#endif // #ifndef VMA_RW_MUTEX + +/* +If providing your own implementation, you need to implement a subset of std::atomic: + +- Constructor(uint32_t desired) +- uint32_t load() const +- void store(uint32_t desired) +- bool compare_exchange_weak(uint32_t& expected, uint32_t desired) +*/ +#ifndef VMA_ATOMIC_UINT32 + #define VMA_ATOMIC_UINT32 std::atomic<uint32_t> +#endif + +#ifndef VMA_DEBUG_ALWAYS_DEDICATED_MEMORY + /** + Every allocation will have its own memory block. + Define to 1 for debugging purposes only. + */ + #define VMA_DEBUG_ALWAYS_DEDICATED_MEMORY (0) +#endif + +#ifndef VMA_DEBUG_ALIGNMENT + /** + Minimum alignment of all allocations, in bytes. + Set to more than 1 for debugging purposes only. Must be power of two. + */ + #define VMA_DEBUG_ALIGNMENT (1) +#endif + +#ifndef VMA_DEBUG_MARGIN + /** + Minimum margin before and after every allocation, in bytes. + Set nonzero for debugging purposes only. + */ + #define VMA_DEBUG_MARGIN (0) +#endif + +#ifndef VMA_DEBUG_INITIALIZE_ALLOCATIONS + /** + Define this macro to 1 to automatically fill new allocations and destroyed + allocations with some bit pattern. + */ + #define VMA_DEBUG_INITIALIZE_ALLOCATIONS (0) +#endif + +#ifndef VMA_DEBUG_DETECT_CORRUPTION + /** + Define this macro to 1 together with non-zero value of VMA_DEBUG_MARGIN to + enable writing magic value to the margin before and after every allocation and + validating it, so that memory corruptions (out-of-bounds writes) are detected. + */ + #define VMA_DEBUG_DETECT_CORRUPTION (0) +#endif + +#ifndef VMA_DEBUG_GLOBAL_MUTEX + /** + Set this to 1 for debugging purposes only, to enable single mutex protecting all + entry calls to the library. Can be useful for debugging multithreading issues. + */ + #define VMA_DEBUG_GLOBAL_MUTEX (0) +#endif + +#ifndef VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY + /** + Minimum value for VkPhysicalDeviceLimits::bufferImageGranularity. + Set to more than 1 for debugging purposes only. Must be power of two. + */ + #define VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY (1) +#endif + +#ifndef VMA_SMALL_HEAP_MAX_SIZE + /// Maximum size of a memory heap in Vulkan to consider it "small". + #define VMA_SMALL_HEAP_MAX_SIZE (1024ull * 1024 * 1024) +#endif + +#ifndef VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE + /// Default size of a block allocated as single VkDeviceMemory from a "large" heap. + #define VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE (256ull * 1024 * 1024) +#endif + +#ifndef VMA_CLASS_NO_COPY + #define VMA_CLASS_NO_COPY(className) \ + private: \ + className(const className&) = delete; \ + className& operator=(const className&) = delete; +#endif + +static const uint32_t VMA_FRAME_INDEX_LOST = UINT32_MAX; + +// Decimal 2139416166, float NaN, little-endian binary 66 E6 84 7F. +static const uint32_t VMA_CORRUPTION_DETECTION_MAGIC_VALUE = 0x7F84E666; + +static const uint8_t VMA_ALLOCATION_FILL_PATTERN_CREATED = 0xDC; +static const uint8_t VMA_ALLOCATION_FILL_PATTERN_DESTROYED = 0xEF; + +/******************************************************************************* +END OF CONFIGURATION +*/ + +static const uint32_t VMA_ALLOCATION_INTERNAL_STRATEGY_MIN_OFFSET = 0x10000000u; + +static VkAllocationCallbacks VmaEmptyAllocationCallbacks = { + VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL }; + +// Returns number of bits set to 1 in (v). +static inline uint32_t VmaCountBitsSet(uint32_t v) +{ + uint32_t c = v - ((v >> 1) & 0x55555555); + c = ((c >> 2) & 0x33333333) + (c & 0x33333333); + c = ((c >> 4) + c) & 0x0F0F0F0F; + c = ((c >> 8) + c) & 0x00FF00FF; + c = ((c >> 16) + c) & 0x0000FFFF; + return c; +} + +// Aligns given value up to nearest multiply of align value. For example: VmaAlignUp(11, 8) = 16. +// Use types like uint32_t, uint64_t as T. +template <typename T> +static inline T VmaAlignUp(T val, T align) +{ + return (val + align - 1) / align * align; +} +// Aligns given value down to nearest multiply of align value. For example: VmaAlignUp(11, 8) = 8. +// Use types like uint32_t, uint64_t as T. +template <typename T> +static inline T VmaAlignDown(T val, T align) +{ + return val / align * align; +} + +// Division with mathematical rounding to nearest number. +template <typename T> +static inline T VmaRoundDiv(T x, T y) +{ + return (x + (y / (T)2)) / y; +} + +/* +Returns true if given number is a power of two. +T must be unsigned integer number or signed integer but always nonnegative. +For 0 returns true. +*/ +template <typename T> +inline bool VmaIsPow2(T x) +{ + return (x & (x-1)) == 0; +} + +// Returns smallest power of 2 greater or equal to v. +static inline uint32_t VmaNextPow2(uint32_t v) +{ + v--; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v++; + return v; +} +static inline uint64_t VmaNextPow2(uint64_t v) +{ + v--; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + v++; + return v; +} + +// Returns largest power of 2 less or equal to v. +static inline uint32_t VmaPrevPow2(uint32_t v) +{ + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v = v ^ (v >> 1); + return v; +} +static inline uint64_t VmaPrevPow2(uint64_t v) +{ + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + v = v ^ (v >> 1); + return v; +} + +static inline bool VmaStrIsEmpty(const char* pStr) +{ + return pStr == VMA_NULL || *pStr == '\0'; +} + +static const char* VmaAlgorithmToStr(uint32_t algorithm) +{ + switch(algorithm) + { + case VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT: + return "Linear"; + case VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT: + return "Buddy"; + case 0: + return "Default"; + default: + VMA_ASSERT(0); + return ""; + } +} + +#ifndef VMA_SORT + +template<typename Iterator, typename Compare> +Iterator VmaQuickSortPartition(Iterator beg, Iterator end, Compare cmp) +{ + Iterator centerValue = end; --centerValue; + Iterator insertIndex = beg; + for(Iterator memTypeIndex = beg; memTypeIndex < centerValue; ++memTypeIndex) + { + if(cmp(*memTypeIndex, *centerValue)) + { + if(insertIndex != memTypeIndex) + { + VMA_SWAP(*memTypeIndex, *insertIndex); + } + ++insertIndex; + } + } + if(insertIndex != centerValue) + { + VMA_SWAP(*insertIndex, *centerValue); + } + return insertIndex; +} + +template<typename Iterator, typename Compare> +void VmaQuickSort(Iterator beg, Iterator end, Compare cmp) +{ + if(beg < end) + { + Iterator it = VmaQuickSortPartition<Iterator, Compare>(beg, end, cmp); + VmaQuickSort<Iterator, Compare>(beg, it, cmp); + VmaQuickSort<Iterator, Compare>(it + 1, end, cmp); + } +} + +#define VMA_SORT(beg, end, cmp) VmaQuickSort(beg, end, cmp) + +#endif // #ifndef VMA_SORT + +/* +Returns true if two memory blocks occupy overlapping pages. +ResourceA must be in less memory offset than ResourceB. + +Algorithm is based on "Vulkan 1.0.39 - A Specification (with all registered Vulkan extensions)" +chapter 11.6 "Resource Memory Association", paragraph "Buffer-Image Granularity". +*/ +static inline bool VmaBlocksOnSamePage( + VkDeviceSize resourceAOffset, + VkDeviceSize resourceASize, + VkDeviceSize resourceBOffset, + VkDeviceSize pageSize) +{ + VMA_ASSERT(resourceAOffset + resourceASize <= resourceBOffset && resourceASize > 0 && pageSize > 0); + VkDeviceSize resourceAEnd = resourceAOffset + resourceASize - 1; + VkDeviceSize resourceAEndPage = resourceAEnd & ~(pageSize - 1); + VkDeviceSize resourceBStart = resourceBOffset; + VkDeviceSize resourceBStartPage = resourceBStart & ~(pageSize - 1); + return resourceAEndPage == resourceBStartPage; +} + +enum VmaSuballocationType +{ + VMA_SUBALLOCATION_TYPE_FREE = 0, + VMA_SUBALLOCATION_TYPE_UNKNOWN = 1, + VMA_SUBALLOCATION_TYPE_BUFFER = 2, + VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN = 3, + VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR = 4, + VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL = 5, + VMA_SUBALLOCATION_TYPE_MAX_ENUM = 0x7FFFFFFF +}; + +/* +Returns true if given suballocation types could conflict and must respect +VkPhysicalDeviceLimits::bufferImageGranularity. They conflict if one is buffer +or linear image and another one is optimal image. If type is unknown, behave +conservatively. +*/ +static inline bool VmaIsBufferImageGranularityConflict( + VmaSuballocationType suballocType1, + VmaSuballocationType suballocType2) +{ + if(suballocType1 > suballocType2) + { + VMA_SWAP(suballocType1, suballocType2); + } + + switch(suballocType1) + { + case VMA_SUBALLOCATION_TYPE_FREE: + return false; + case VMA_SUBALLOCATION_TYPE_UNKNOWN: + return true; + case VMA_SUBALLOCATION_TYPE_BUFFER: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL: + return false; + default: + VMA_ASSERT(0); + return true; + } +} + +static void VmaWriteMagicValue(void* pData, VkDeviceSize offset) +{ + uint32_t* pDst = (uint32_t*)((char*)pData + offset); + const size_t numberCount = VMA_DEBUG_MARGIN / sizeof(uint32_t); + for(size_t i = 0; i != numberCount; ++i, ++pDst) + { + *pDst = VMA_CORRUPTION_DETECTION_MAGIC_VALUE; + } +} + +static bool VmaValidateMagicValue(const void* pData, VkDeviceSize offset) +{ + const uint32_t* pSrc = (const uint32_t*)((const char*)pData + offset); + const size_t numberCount = VMA_DEBUG_MARGIN / sizeof(uint32_t); + for(size_t i = 0; i != numberCount; ++i, ++pSrc) + { + if(*pSrc != VMA_CORRUPTION_DETECTION_MAGIC_VALUE) + { + return false; + } + } + return true; +} + +// Helper RAII class to lock a mutex in constructor and unlock it in destructor (at the end of scope). +struct VmaMutexLock +{ + VMA_CLASS_NO_COPY(VmaMutexLock) +public: + VmaMutexLock(VMA_MUTEX& mutex, bool useMutex) : + m_pMutex(useMutex ? &mutex : VMA_NULL) + { if(m_pMutex) { m_pMutex->Lock(); } } + ~VmaMutexLock() + { if(m_pMutex) { m_pMutex->Unlock(); } } +private: + VMA_MUTEX* m_pMutex; +}; + +// Helper RAII class to lock a RW mutex in constructor and unlock it in destructor (at the end of scope), for reading. +struct VmaMutexLockRead +{ + VMA_CLASS_NO_COPY(VmaMutexLockRead) +public: + VmaMutexLockRead(VMA_RW_MUTEX& mutex, bool useMutex) : + m_pMutex(useMutex ? &mutex : VMA_NULL) + { if(m_pMutex) { m_pMutex->LockRead(); } } + ~VmaMutexLockRead() { if(m_pMutex) { m_pMutex->UnlockRead(); } } +private: + VMA_RW_MUTEX* m_pMutex; +}; + +// Helper RAII class to lock a RW mutex in constructor and unlock it in destructor (at the end of scope), for writing. +struct VmaMutexLockWrite +{ + VMA_CLASS_NO_COPY(VmaMutexLockWrite) +public: + VmaMutexLockWrite(VMA_RW_MUTEX& mutex, bool useMutex) : + m_pMutex(useMutex ? &mutex : VMA_NULL) + { if(m_pMutex) { m_pMutex->LockWrite(); } } + ~VmaMutexLockWrite() { if(m_pMutex) { m_pMutex->UnlockWrite(); } } +private: + VMA_RW_MUTEX* m_pMutex; +}; + +#if VMA_DEBUG_GLOBAL_MUTEX + static VMA_MUTEX gDebugGlobalMutex; + #define VMA_DEBUG_GLOBAL_MUTEX_LOCK VmaMutexLock debugGlobalMutexLock(gDebugGlobalMutex, true); +#else + #define VMA_DEBUG_GLOBAL_MUTEX_LOCK +#endif + +// Minimum size of a free suballocation to register it in the free suballocation collection. +static const VkDeviceSize VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER = 16; + +/* +Performs binary search and returns iterator to first element that is greater or +equal to (key), according to comparison (cmp). + +Cmp should return true if first argument is less than second argument. + +Returned value is the found element, if present in the collection or place where +new element with value (key) should be inserted. +*/ +template <typename CmpLess, typename IterT, typename KeyT> +static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT &key, CmpLess cmp) +{ + size_t down = 0, up = (end - beg); + while(down < up) + { + const size_t mid = (down + up) / 2; + if(cmp(*(beg+mid), key)) + { + down = mid + 1; + } + else + { + up = mid; + } + } + return beg + down; +} + +/* +Returns true if all pointers in the array are not-null and unique. +Warning! O(n^2) complexity. Use only inside VMA_HEAVY_ASSERT. +T must be pointer type, e.g. VmaAllocation, VmaPool. +*/ +template<typename T> +static bool VmaValidatePointerArray(uint32_t count, const T* arr) +{ + for(uint32_t i = 0; i < count; ++i) + { + const T iPtr = arr[i]; + if(iPtr == VMA_NULL) + { + return false; + } + for(uint32_t j = i + 1; j < count; ++j) + { + if(iPtr == arr[j]) + { + return false; + } + } + } + return true; +} + +//////////////////////////////////////////////////////////////////////////////// +// Memory allocation + +static void* VmaMalloc(const VkAllocationCallbacks* pAllocationCallbacks, size_t size, size_t alignment) +{ + if((pAllocationCallbacks != VMA_NULL) && + (pAllocationCallbacks->pfnAllocation != VMA_NULL)) + { + return (*pAllocationCallbacks->pfnAllocation)( + pAllocationCallbacks->pUserData, + size, + alignment, + VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); + } + else + { + return VMA_SYSTEM_ALIGNED_MALLOC(size, alignment); + } +} + +static void VmaFree(const VkAllocationCallbacks* pAllocationCallbacks, void* ptr) +{ + if((pAllocationCallbacks != VMA_NULL) && + (pAllocationCallbacks->pfnFree != VMA_NULL)) + { + (*pAllocationCallbacks->pfnFree)(pAllocationCallbacks->pUserData, ptr); + } + else + { + VMA_SYSTEM_FREE(ptr); + } +} + +template<typename T> +static T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks) +{ + return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T), VMA_ALIGN_OF(T)); +} + +template<typename T> +static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count) +{ + return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T) * count, VMA_ALIGN_OF(T)); +} + +#define vma_new(allocator, type) new(VmaAllocate<type>(allocator))(type) + +#define vma_new_array(allocator, type, count) new(VmaAllocateArray<type>((allocator), (count)))(type) + +template<typename T> +static void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr) +{ + ptr->~T(); + VmaFree(pAllocationCallbacks, ptr); +} + +template<typename T> +static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count) +{ + if(ptr != VMA_NULL) + { + for(size_t i = count; i--; ) + { + ptr[i].~T(); + } + VmaFree(pAllocationCallbacks, ptr); + } +} + +// STL-compatible allocator. +template<typename T> +class VmaStlAllocator +{ +public: + const VkAllocationCallbacks* const m_pCallbacks; + typedef T value_type; + + VmaStlAllocator(const VkAllocationCallbacks* pCallbacks) : m_pCallbacks(pCallbacks) { } + template<typename U> VmaStlAllocator(const VmaStlAllocator<U>& src) : m_pCallbacks(src.m_pCallbacks) { } + + T* allocate(size_t n) { return VmaAllocateArray<T>(m_pCallbacks, n); } + void deallocate(T* p, size_t /*n*/) { VmaFree(m_pCallbacks, p); } + + template<typename U> + bool operator==(const VmaStlAllocator<U>& rhs) const + { + return m_pCallbacks == rhs.m_pCallbacks; + } + template<typename U> + bool operator!=(const VmaStlAllocator<U>& rhs) const + { + return m_pCallbacks != rhs.m_pCallbacks; + } + + VmaStlAllocator& operator=(const VmaStlAllocator& x) = delete; +}; + +#if VMA_USE_STL_VECTOR + +#define VmaVector std::vector + +template<typename T, typename allocatorT> +static void VmaVectorInsert(std::vector<T, allocatorT>& vec, size_t index, const T& item) +{ + vec.insert(vec.begin() + index, item); +} + +template<typename T, typename allocatorT> +static void VmaVectorRemove(std::vector<T, allocatorT>& vec, size_t index) +{ + vec.erase(vec.begin() + index); +} + +#else // #if VMA_USE_STL_VECTOR + +/* Class with interface compatible with subset of std::vector. +T must be POD because constructors and destructors are not called and memcpy is +used for these objects. */ +template<typename T, typename AllocatorT> +class VmaVector +{ +public: + typedef T value_type; + + VmaVector(const AllocatorT& allocator) : + m_Allocator(allocator), + m_pArray(VMA_NULL), + m_Count(0), + m_Capacity(0) + { + } + + VmaVector(size_t count, const AllocatorT& allocator) : + m_Allocator(allocator), + m_pArray(count ? (T*)VmaAllocateArray<T>(allocator.m_pCallbacks, count) : VMA_NULL), + m_Count(count), + m_Capacity(count) + { + } + + VmaVector(const VmaVector<T, AllocatorT>& src) : + m_Allocator(src.m_Allocator), + m_pArray(src.m_Count ? (T*)VmaAllocateArray<T>(src.m_Allocator.m_pCallbacks, src.m_Count) : VMA_NULL), + m_Count(src.m_Count), + m_Capacity(src.m_Count) + { + if(m_Count != 0) + { + memcpy(m_pArray, src.m_pArray, m_Count * sizeof(T)); + } + } + + ~VmaVector() + { + VmaFree(m_Allocator.m_pCallbacks, m_pArray); + } + + VmaVector& operator=(const VmaVector<T, AllocatorT>& rhs) + { + if(&rhs != this) + { + resize(rhs.m_Count); + if(m_Count != 0) + { + memcpy(m_pArray, rhs.m_pArray, m_Count * sizeof(T)); + } + } + return *this; + } + + bool empty() const { return m_Count == 0; } + size_t size() const { return m_Count; } + T* data() { return m_pArray; } + const T* data() const { return m_pArray; } + + T& operator[](size_t index) + { + VMA_HEAVY_ASSERT(index < m_Count); + return m_pArray[index]; + } + const T& operator[](size_t index) const + { + VMA_HEAVY_ASSERT(index < m_Count); + return m_pArray[index]; + } + + T& front() + { + VMA_HEAVY_ASSERT(m_Count > 0); + return m_pArray[0]; + } + const T& front() const + { + VMA_HEAVY_ASSERT(m_Count > 0); + return m_pArray[0]; + } + T& back() + { + VMA_HEAVY_ASSERT(m_Count > 0); + return m_pArray[m_Count - 1]; + } + const T& back() const + { + VMA_HEAVY_ASSERT(m_Count > 0); + return m_pArray[m_Count - 1]; + } + + void reserve(size_t newCapacity, bool freeMemory = false) + { + newCapacity = VMA_MAX(newCapacity, m_Count); + + if((newCapacity < m_Capacity) && !freeMemory) + { + newCapacity = m_Capacity; + } + + if(newCapacity != m_Capacity) + { + T* const newArray = newCapacity ? VmaAllocateArray<T>(m_Allocator, newCapacity) : VMA_NULL; + if(m_Count != 0) + { + memcpy(newArray, m_pArray, m_Count * sizeof(T)); + } + VmaFree(m_Allocator.m_pCallbacks, m_pArray); + m_Capacity = newCapacity; + m_pArray = newArray; + } + } + + void resize(size_t newCount, bool freeMemory = false) + { + size_t newCapacity = m_Capacity; + if(newCount > m_Capacity) + { + newCapacity = VMA_MAX(newCount, VMA_MAX(m_Capacity * 3 / 2, (size_t)8)); + } + else if(freeMemory) + { + newCapacity = newCount; + } + + if(newCapacity != m_Capacity) + { + T* const newArray = newCapacity ? VmaAllocateArray<T>(m_Allocator.m_pCallbacks, newCapacity) : VMA_NULL; + const size_t elementsToCopy = VMA_MIN(m_Count, newCount); + if(elementsToCopy != 0) + { + memcpy(newArray, m_pArray, elementsToCopy * sizeof(T)); + } + VmaFree(m_Allocator.m_pCallbacks, m_pArray); + m_Capacity = newCapacity; + m_pArray = newArray; + } + + m_Count = newCount; + } + + void clear(bool freeMemory = false) + { + resize(0, freeMemory); + } + + void insert(size_t index, const T& src) + { + VMA_HEAVY_ASSERT(index <= m_Count); + const size_t oldCount = size(); + resize(oldCount + 1); + if(index < oldCount) + { + memmove(m_pArray + (index + 1), m_pArray + index, (oldCount - index) * sizeof(T)); + } + m_pArray[index] = src; + } + + void remove(size_t index) + { + VMA_HEAVY_ASSERT(index < m_Count); + const size_t oldCount = size(); + if(index < oldCount - 1) + { + memmove(m_pArray + index, m_pArray + (index + 1), (oldCount - index - 1) * sizeof(T)); + } + resize(oldCount - 1); + } + + void push_back(const T& src) + { + const size_t newIndex = size(); + resize(newIndex + 1); + m_pArray[newIndex] = src; + } + + void pop_back() + { + VMA_HEAVY_ASSERT(m_Count > 0); + resize(size() - 1); + } + + void push_front(const T& src) + { + insert(0, src); + } + + void pop_front() + { + VMA_HEAVY_ASSERT(m_Count > 0); + remove(0); + } + + typedef T* iterator; + + iterator begin() { return m_pArray; } + iterator end() { return m_pArray + m_Count; } + +private: + AllocatorT m_Allocator; + T* m_pArray; + size_t m_Count; + size_t m_Capacity; +}; + +template<typename T, typename allocatorT> +static void VmaVectorInsert(VmaVector<T, allocatorT>& vec, size_t index, const T& item) +{ + vec.insert(index, item); +} + +template<typename T, typename allocatorT> +static void VmaVectorRemove(VmaVector<T, allocatorT>& vec, size_t index) +{ + vec.remove(index); +} + +#endif // #if VMA_USE_STL_VECTOR + +template<typename CmpLess, typename VectorT> +size_t VmaVectorInsertSorted(VectorT& vector, const typename VectorT::value_type& value) +{ + const size_t indexToInsert = VmaBinaryFindFirstNotLess( + vector.data(), + vector.data() + vector.size(), + value, + CmpLess()) - vector.data(); + VmaVectorInsert(vector, indexToInsert, value); + return indexToInsert; +} + +template<typename CmpLess, typename VectorT> +bool VmaVectorRemoveSorted(VectorT& vector, const typename VectorT::value_type& value) +{ + CmpLess comparator; + typename VectorT::iterator it = VmaBinaryFindFirstNotLess( + vector.begin(), + vector.end(), + value, + comparator); + if((it != vector.end()) && !comparator(*it, value) && !comparator(value, *it)) + { + size_t indexToRemove = it - vector.begin(); + VmaVectorRemove(vector, indexToRemove); + return true; + } + return false; +} + +template<typename CmpLess, typename IterT, typename KeyT> +IterT VmaVectorFindSorted(const IterT& beg, const IterT& end, const KeyT& value) +{ + CmpLess comparator; + IterT it = VmaBinaryFindFirstNotLess<CmpLess, IterT, KeyT>( + beg, end, value, comparator); + if(it == end || + (!comparator(*it, value) && !comparator(value, *it))) + { + return it; + } + return end; +} + +//////////////////////////////////////////////////////////////////////////////// +// class VmaPoolAllocator + +/* +Allocator for objects of type T using a list of arrays (pools) to speed up +allocation. Number of elements that can be allocated is not bounded because +allocator can create multiple blocks. +*/ +template<typename T> +class VmaPoolAllocator +{ + VMA_CLASS_NO_COPY(VmaPoolAllocator) +public: + VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, size_t itemsPerBlock); + ~VmaPoolAllocator(); + void Clear(); + T* Alloc(); + void Free(T* ptr); + +private: + union Item + { + uint32_t NextFreeIndex; + T Value; + }; + + struct ItemBlock + { + Item* pItems; + uint32_t FirstFreeIndex; + }; + + const VkAllocationCallbacks* m_pAllocationCallbacks; + size_t m_ItemsPerBlock; + VmaVector< ItemBlock, VmaStlAllocator<ItemBlock> > m_ItemBlocks; + + ItemBlock& CreateNewBlock(); +}; + +template<typename T> +VmaPoolAllocator<T>::VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, size_t itemsPerBlock) : + m_pAllocationCallbacks(pAllocationCallbacks), + m_ItemsPerBlock(itemsPerBlock), + m_ItemBlocks(VmaStlAllocator<ItemBlock>(pAllocationCallbacks)) +{ + VMA_ASSERT(itemsPerBlock > 0); +} + +template<typename T> +VmaPoolAllocator<T>::~VmaPoolAllocator() +{ + Clear(); +} + +template<typename T> +void VmaPoolAllocator<T>::Clear() +{ + for(size_t i = m_ItemBlocks.size(); i--; ) + vma_delete_array(m_pAllocationCallbacks, m_ItemBlocks[i].pItems, m_ItemsPerBlock); + m_ItemBlocks.clear(); +} + +template<typename T> +T* VmaPoolAllocator<T>::Alloc() +{ + for(size_t i = m_ItemBlocks.size(); i--; ) + { + ItemBlock& block = m_ItemBlocks[i]; + // This block has some free items: Use first one. + if(block.FirstFreeIndex != UINT32_MAX) + { + Item* const pItem = &block.pItems[block.FirstFreeIndex]; + block.FirstFreeIndex = pItem->NextFreeIndex; + return &pItem->Value; + } + } + + // No block has free item: Create new one and use it. + ItemBlock& newBlock = CreateNewBlock(); + Item* const pItem = &newBlock.pItems[0]; + newBlock.FirstFreeIndex = pItem->NextFreeIndex; + return &pItem->Value; +} + +template<typename T> +void VmaPoolAllocator<T>::Free(T* ptr) +{ + // Search all memory blocks to find ptr. + for(size_t i = 0; i < m_ItemBlocks.size(); ++i) + { + ItemBlock& block = m_ItemBlocks[i]; + + // Casting to union. + Item* pItemPtr; + memcpy(&pItemPtr, &ptr, sizeof(pItemPtr)); + + // Check if pItemPtr is in address range of this block. + if((pItemPtr >= block.pItems) && (pItemPtr < block.pItems + m_ItemsPerBlock)) + { + const uint32_t index = static_cast<uint32_t>(pItemPtr - block.pItems); + pItemPtr->NextFreeIndex = block.FirstFreeIndex; + block.FirstFreeIndex = index; + return; + } + } + VMA_ASSERT(0 && "Pointer doesn't belong to this memory pool."); +} + +template<typename T> +typename VmaPoolAllocator<T>::ItemBlock& VmaPoolAllocator<T>::CreateNewBlock() +{ + ItemBlock newBlock = { + vma_new_array(m_pAllocationCallbacks, Item, m_ItemsPerBlock), 0 }; + + m_ItemBlocks.push_back(newBlock); + + // Setup singly-linked list of all free items in this block. + for(uint32_t i = 0; i < m_ItemsPerBlock - 1; ++i) + newBlock.pItems[i].NextFreeIndex = i + 1; + newBlock.pItems[m_ItemsPerBlock - 1].NextFreeIndex = UINT32_MAX; + return m_ItemBlocks.back(); +} + +//////////////////////////////////////////////////////////////////////////////// +// class VmaRawList, VmaList + +#if VMA_USE_STL_LIST + +#define VmaList std::list + +#else // #if VMA_USE_STL_LIST + +template<typename T> +struct VmaListItem +{ + VmaListItem* pPrev; + VmaListItem* pNext; + T Value; +}; + +// Doubly linked list. +template<typename T> +class VmaRawList +{ + VMA_CLASS_NO_COPY(VmaRawList) +public: + typedef VmaListItem<T> ItemType; + + VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks); + ~VmaRawList(); + void Clear(); + + size_t GetCount() const { return m_Count; } + bool IsEmpty() const { return m_Count == 0; } + + ItemType* Front() { return m_pFront; } + const ItemType* Front() const { return m_pFront; } + ItemType* Back() { return m_pBack; } + const ItemType* Back() const { return m_pBack; } + + ItemType* PushBack(); + ItemType* PushFront(); + ItemType* PushBack(const T& value); + ItemType* PushFront(const T& value); + void PopBack(); + void PopFront(); + + // Item can be null - it means PushBack. + ItemType* InsertBefore(ItemType* pItem); + // Item can be null - it means PushFront. + ItemType* InsertAfter(ItemType* pItem); + + ItemType* InsertBefore(ItemType* pItem, const T& value); + ItemType* InsertAfter(ItemType* pItem, const T& value); + + void Remove(ItemType* pItem); + +private: + const VkAllocationCallbacks* const m_pAllocationCallbacks; + VmaPoolAllocator<ItemType> m_ItemAllocator; + ItemType* m_pFront; + ItemType* m_pBack; + size_t m_Count; +}; + +template<typename T> +VmaRawList<T>::VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks) : + m_pAllocationCallbacks(pAllocationCallbacks), + m_ItemAllocator(pAllocationCallbacks, 128), + m_pFront(VMA_NULL), + m_pBack(VMA_NULL), + m_Count(0) +{ +} + +template<typename T> +VmaRawList<T>::~VmaRawList() +{ + // Intentionally not calling Clear, because that would be unnecessary + // computations to return all items to m_ItemAllocator as free. +} + +template<typename T> +void VmaRawList<T>::Clear() +{ + if(IsEmpty() == false) + { + ItemType* pItem = m_pBack; + while(pItem != VMA_NULL) + { + ItemType* const pPrevItem = pItem->pPrev; + m_ItemAllocator.Free(pItem); + pItem = pPrevItem; + } + m_pFront = VMA_NULL; + m_pBack = VMA_NULL; + m_Count = 0; + } +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::PushBack() +{ + ItemType* const pNewItem = m_ItemAllocator.Alloc(); + pNewItem->pNext = VMA_NULL; + if(IsEmpty()) + { + pNewItem->pPrev = VMA_NULL; + m_pFront = pNewItem; + m_pBack = pNewItem; + m_Count = 1; + } + else + { + pNewItem->pPrev = m_pBack; + m_pBack->pNext = pNewItem; + m_pBack = pNewItem; + ++m_Count; + } + return pNewItem; +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::PushFront() +{ + ItemType* const pNewItem = m_ItemAllocator.Alloc(); + pNewItem->pPrev = VMA_NULL; + if(IsEmpty()) + { + pNewItem->pNext = VMA_NULL; + m_pFront = pNewItem; + m_pBack = pNewItem; + m_Count = 1; + } + else + { + pNewItem->pNext = m_pFront; + m_pFront->pPrev = pNewItem; + m_pFront = pNewItem; + ++m_Count; + } + return pNewItem; +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::PushBack(const T& value) +{ + ItemType* const pNewItem = PushBack(); + pNewItem->Value = value; + return pNewItem; +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::PushFront(const T& value) +{ + ItemType* const pNewItem = PushFront(); + pNewItem->Value = value; + return pNewItem; +} + +template<typename T> +void VmaRawList<T>::PopBack() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const pBackItem = m_pBack; + ItemType* const pPrevItem = pBackItem->pPrev; + if(pPrevItem != VMA_NULL) + { + pPrevItem->pNext = VMA_NULL; + } + m_pBack = pPrevItem; + m_ItemAllocator.Free(pBackItem); + --m_Count; +} + +template<typename T> +void VmaRawList<T>::PopFront() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const pFrontItem = m_pFront; + ItemType* const pNextItem = pFrontItem->pNext; + if(pNextItem != VMA_NULL) + { + pNextItem->pPrev = VMA_NULL; + } + m_pFront = pNextItem; + m_ItemAllocator.Free(pFrontItem); + --m_Count; +} + +template<typename T> +void VmaRawList<T>::Remove(ItemType* pItem) +{ + VMA_HEAVY_ASSERT(pItem != VMA_NULL); + VMA_HEAVY_ASSERT(m_Count > 0); + + if(pItem->pPrev != VMA_NULL) + { + pItem->pPrev->pNext = pItem->pNext; + } + else + { + VMA_HEAVY_ASSERT(m_pFront == pItem); + m_pFront = pItem->pNext; + } + + if(pItem->pNext != VMA_NULL) + { + pItem->pNext->pPrev = pItem->pPrev; + } + else + { + VMA_HEAVY_ASSERT(m_pBack == pItem); + m_pBack = pItem->pPrev; + } + + m_ItemAllocator.Free(pItem); + --m_Count; +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::InsertBefore(ItemType* pItem) +{ + if(pItem != VMA_NULL) + { + ItemType* const prevItem = pItem->pPrev; + ItemType* const newItem = m_ItemAllocator.Alloc(); + newItem->pPrev = prevItem; + newItem->pNext = pItem; + pItem->pPrev = newItem; + if(prevItem != VMA_NULL) + { + prevItem->pNext = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_pFront == pItem); + m_pFront = newItem; + } + ++m_Count; + return newItem; + } + else + return PushBack(); +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::InsertAfter(ItemType* pItem) +{ + if(pItem != VMA_NULL) + { + ItemType* const nextItem = pItem->pNext; + ItemType* const newItem = m_ItemAllocator.Alloc(); + newItem->pNext = nextItem; + newItem->pPrev = pItem; + pItem->pNext = newItem; + if(nextItem != VMA_NULL) + { + nextItem->pPrev = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_pBack == pItem); + m_pBack = newItem; + } + ++m_Count; + return newItem; + } + else + return PushFront(); +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::InsertBefore(ItemType* pItem, const T& value) +{ + ItemType* const newItem = InsertBefore(pItem); + newItem->Value = value; + return newItem; +} + +template<typename T> +VmaListItem<T>* VmaRawList<T>::InsertAfter(ItemType* pItem, const T& value) +{ + ItemType* const newItem = InsertAfter(pItem); + newItem->Value = value; + return newItem; +} + +template<typename T, typename AllocatorT> +class VmaList +{ + VMA_CLASS_NO_COPY(VmaList) +public: + class iterator + { + public: + iterator() : + m_pList(VMA_NULL), + m_pItem(VMA_NULL) + { + } + + T& operator*() const + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + return m_pItem->Value; + } + T* operator->() const + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + return &m_pItem->Value; + } + + iterator& operator++() + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + m_pItem = m_pItem->pNext; + return *this; + } + iterator& operator--() + { + if(m_pItem != VMA_NULL) + { + m_pItem = m_pItem->pPrev; + } + else + { + VMA_HEAVY_ASSERT(!m_pList->IsEmpty()); + m_pItem = m_pList->Back(); + } + return *this; + } + + iterator operator++(int) + { + iterator result = *this; + ++*this; + return result; + } + iterator operator--(int) + { + iterator result = *this; + --*this; + return result; + } + + bool operator==(const iterator& rhs) const + { + VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); + return m_pItem == rhs.m_pItem; + } + bool operator!=(const iterator& rhs) const + { + VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); + return m_pItem != rhs.m_pItem; + } + + private: + VmaRawList<T>* m_pList; + VmaListItem<T>* m_pItem; + + iterator(VmaRawList<T>* pList, VmaListItem<T>* pItem) : + m_pList(pList), + m_pItem(pItem) + { + } + + friend class VmaList<T, AllocatorT>; + }; + + class const_iterator + { + public: + const_iterator() : + m_pList(VMA_NULL), + m_pItem(VMA_NULL) + { + } + + const_iterator(const iterator& src) : + m_pList(src.m_pList), + m_pItem(src.m_pItem) + { + } + + const T& operator*() const + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + return m_pItem->Value; + } + const T* operator->() const + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + return &m_pItem->Value; + } + + const_iterator& operator++() + { + VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); + m_pItem = m_pItem->pNext; + return *this; + } + const_iterator& operator--() + { + if(m_pItem != VMA_NULL) + { + m_pItem = m_pItem->pPrev; + } + else + { + VMA_HEAVY_ASSERT(!m_pList->IsEmpty()); + m_pItem = m_pList->Back(); + } + return *this; + } + + const_iterator operator++(int) + { + const_iterator result = *this; + ++*this; + return result; + } + const_iterator operator--(int) + { + const_iterator result = *this; + --*this; + return result; + } + + bool operator==(const const_iterator& rhs) const + { + VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); + return m_pItem == rhs.m_pItem; + } + bool operator!=(const const_iterator& rhs) const + { + VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); + return m_pItem != rhs.m_pItem; + } + + private: + const_iterator(const VmaRawList<T>* pList, const VmaListItem<T>* pItem) : + m_pList(pList), + m_pItem(pItem) + { + } + + const VmaRawList<T>* m_pList; + const VmaListItem<T>* m_pItem; + + friend class VmaList<T, AllocatorT>; + }; + + VmaList(const AllocatorT& allocator) : m_RawList(allocator.m_pCallbacks) { } + + bool empty() const { return m_RawList.IsEmpty(); } + size_t size() const { return m_RawList.GetCount(); } + + iterator begin() { return iterator(&m_RawList, m_RawList.Front()); } + iterator end() { return iterator(&m_RawList, VMA_NULL); } + + const_iterator cbegin() const { return const_iterator(&m_RawList, m_RawList.Front()); } + const_iterator cend() const { return const_iterator(&m_RawList, VMA_NULL); } + + void clear() { m_RawList.Clear(); } + void push_back(const T& value) { m_RawList.PushBack(value); } + void erase(iterator it) { m_RawList.Remove(it.m_pItem); } + iterator insert(iterator it, const T& value) { return iterator(&m_RawList, m_RawList.InsertBefore(it.m_pItem, value)); } + +private: + VmaRawList<T> m_RawList; +}; + +#endif // #if VMA_USE_STL_LIST + +//////////////////////////////////////////////////////////////////////////////// +// class VmaMap + +// Unused in this version. +#if 0 + +#if VMA_USE_STL_UNORDERED_MAP + +#define VmaPair std::pair + +#define VMA_MAP_TYPE(KeyT, ValueT) \ + std::unordered_map< KeyT, ValueT, std::hash<KeyT>, std::equal_to<KeyT>, VmaStlAllocator< std::pair<KeyT, ValueT> > > + +#else // #if VMA_USE_STL_UNORDERED_MAP + +template<typename T1, typename T2> +struct VmaPair +{ + T1 first; + T2 second; + + VmaPair() : first(), second() { } + VmaPair(const T1& firstSrc, const T2& secondSrc) : first(firstSrc), second(secondSrc) { } +}; + +/* Class compatible with subset of interface of std::unordered_map. +KeyT, ValueT must be POD because they will be stored in VmaVector. +*/ +template<typename KeyT, typename ValueT> +class VmaMap +{ +public: + typedef VmaPair<KeyT, ValueT> PairType; + typedef PairType* iterator; + + VmaMap(const VmaStlAllocator<PairType>& allocator) : m_Vector(allocator) { } + + iterator begin() { return m_Vector.begin(); } + iterator end() { return m_Vector.end(); } + + void insert(const PairType& pair); + iterator find(const KeyT& key); + void erase(iterator it); + +private: + VmaVector< PairType, VmaStlAllocator<PairType> > m_Vector; +}; + +#define VMA_MAP_TYPE(KeyT, ValueT) VmaMap<KeyT, ValueT> + +template<typename FirstT, typename SecondT> +struct VmaPairFirstLess +{ + bool operator()(const VmaPair<FirstT, SecondT>& lhs, const VmaPair<FirstT, SecondT>& rhs) const + { + return lhs.first < rhs.first; + } + bool operator()(const VmaPair<FirstT, SecondT>& lhs, const FirstT& rhsFirst) const + { + return lhs.first < rhsFirst; + } +}; + +template<typename KeyT, typename ValueT> +void VmaMap<KeyT, ValueT>::insert(const PairType& pair) +{ + const size_t indexToInsert = VmaBinaryFindFirstNotLess( + m_Vector.data(), + m_Vector.data() + m_Vector.size(), + pair, + VmaPairFirstLess<KeyT, ValueT>()) - m_Vector.data(); + VmaVectorInsert(m_Vector, indexToInsert, pair); +} + +template<typename KeyT, typename ValueT> +VmaPair<KeyT, ValueT>* VmaMap<KeyT, ValueT>::find(const KeyT& key) +{ + PairType* it = VmaBinaryFindFirstNotLess( + m_Vector.data(), + m_Vector.data() + m_Vector.size(), + key, + VmaPairFirstLess<KeyT, ValueT>()); + if((it != m_Vector.end()) && (it->first == key)) + { + return it; + } + else + { + return m_Vector.end(); + } +} + +template<typename KeyT, typename ValueT> +void VmaMap<KeyT, ValueT>::erase(iterator it) +{ + VmaVectorRemove(m_Vector, it - m_Vector.begin()); +} + +#endif // #if VMA_USE_STL_UNORDERED_MAP + +#endif // #if 0 + +//////////////////////////////////////////////////////////////////////////////// + +class VmaDeviceMemoryBlock; + +enum VMA_CACHE_OPERATION { VMA_CACHE_FLUSH, VMA_CACHE_INVALIDATE }; + +struct VmaAllocation_T +{ + VMA_CLASS_NO_COPY(VmaAllocation_T) +private: + static const uint8_t MAP_COUNT_FLAG_PERSISTENT_MAP = 0x80; + + enum FLAGS + { + FLAG_USER_DATA_STRING = 0x01, + }; + +public: + enum ALLOCATION_TYPE + { + ALLOCATION_TYPE_NONE, + ALLOCATION_TYPE_BLOCK, + ALLOCATION_TYPE_DEDICATED, + }; + + VmaAllocation_T(uint32_t currentFrameIndex, bool userDataString) : + m_Alignment(1), + m_Size(0), + m_pUserData(VMA_NULL), + m_LastUseFrameIndex(currentFrameIndex), + m_Type((uint8_t)ALLOCATION_TYPE_NONE), + m_SuballocationType((uint8_t)VMA_SUBALLOCATION_TYPE_UNKNOWN), + m_MapCount(0), + m_Flags(userDataString ? (uint8_t)FLAG_USER_DATA_STRING : 0) + { +#if VMA_STATS_STRING_ENABLED + m_CreationFrameIndex = currentFrameIndex; + m_BufferImageUsage = 0; +#endif + } + + ~VmaAllocation_T() + { + VMA_ASSERT((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) == 0 && "Allocation was not unmapped before destruction."); + + // Check if owned string was freed. + VMA_ASSERT(m_pUserData == VMA_NULL); + } + + void InitBlockAllocation( + VmaPool hPool, + VmaDeviceMemoryBlock* block, + VkDeviceSize offset, + VkDeviceSize alignment, + VkDeviceSize size, + VmaSuballocationType suballocationType, + bool mapped, + bool canBecomeLost) + { + VMA_ASSERT(m_Type == ALLOCATION_TYPE_NONE); + VMA_ASSERT(block != VMA_NULL); + m_Type = (uint8_t)ALLOCATION_TYPE_BLOCK; + m_Alignment = alignment; + m_Size = size; + m_MapCount = mapped ? MAP_COUNT_FLAG_PERSISTENT_MAP : 0; + m_SuballocationType = (uint8_t)suballocationType; + m_BlockAllocation.m_hPool = hPool; + m_BlockAllocation.m_Block = block; + m_BlockAllocation.m_Offset = offset; + m_BlockAllocation.m_CanBecomeLost = canBecomeLost; + } + + void InitLost() + { + VMA_ASSERT(m_Type == ALLOCATION_TYPE_NONE); + VMA_ASSERT(m_LastUseFrameIndex.load() == VMA_FRAME_INDEX_LOST); + m_Type = (uint8_t)ALLOCATION_TYPE_BLOCK; + m_BlockAllocation.m_hPool = VK_NULL_HANDLE; + m_BlockAllocation.m_Block = VMA_NULL; + m_BlockAllocation.m_Offset = 0; + m_BlockAllocation.m_CanBecomeLost = true; + } + + void ChangeBlockAllocation( + VmaAllocator hAllocator, + VmaDeviceMemoryBlock* block, + VkDeviceSize offset); + + void ChangeSize(VkDeviceSize newSize); + void ChangeOffset(VkDeviceSize newOffset); + + // pMappedData not null means allocation is created with MAPPED flag. + void InitDedicatedAllocation( + uint32_t memoryTypeIndex, + VkDeviceMemory hMemory, + VmaSuballocationType suballocationType, + void* pMappedData, + VkDeviceSize size) + { + VMA_ASSERT(m_Type == ALLOCATION_TYPE_NONE); + VMA_ASSERT(hMemory != VK_NULL_HANDLE); + m_Type = (uint8_t)ALLOCATION_TYPE_DEDICATED; + m_Alignment = 0; + m_Size = size; + m_SuballocationType = (uint8_t)suballocationType; + m_MapCount = (pMappedData != VMA_NULL) ? MAP_COUNT_FLAG_PERSISTENT_MAP : 0; + m_DedicatedAllocation.m_MemoryTypeIndex = memoryTypeIndex; + m_DedicatedAllocation.m_hMemory = hMemory; + m_DedicatedAllocation.m_pMappedData = pMappedData; + } + + ALLOCATION_TYPE GetType() const { return (ALLOCATION_TYPE)m_Type; } + VkDeviceSize GetAlignment() const { return m_Alignment; } + VkDeviceSize GetSize() const { return m_Size; } + bool IsUserDataString() const { return (m_Flags & FLAG_USER_DATA_STRING) != 0; } + void* GetUserData() const { return m_pUserData; } + void SetUserData(VmaAllocator hAllocator, void* pUserData); + VmaSuballocationType GetSuballocationType() const { return (VmaSuballocationType)m_SuballocationType; } + + VmaDeviceMemoryBlock* GetBlock() const + { + VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); + return m_BlockAllocation.m_Block; + } + VkDeviceSize GetOffset() const; + VkDeviceMemory GetMemory() const; + uint32_t GetMemoryTypeIndex() const; + bool IsPersistentMap() const { return (m_MapCount & MAP_COUNT_FLAG_PERSISTENT_MAP) != 0; } + void* GetMappedData() const; + bool CanBecomeLost() const; + VmaPool GetPool() const; + + uint32_t GetLastUseFrameIndex() const + { + return m_LastUseFrameIndex.load(); + } + bool CompareExchangeLastUseFrameIndex(uint32_t& expected, uint32_t desired) + { + return m_LastUseFrameIndex.compare_exchange_weak(expected, desired); + } + /* + - If hAllocation.LastUseFrameIndex + frameInUseCount < allocator.CurrentFrameIndex, + makes it lost by setting LastUseFrameIndex = VMA_FRAME_INDEX_LOST and returns true. + - Else, returns false. + + If hAllocation is already lost, assert - you should not call it then. + If hAllocation was not created with CAN_BECOME_LOST_BIT, assert. + */ + bool MakeLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); + + void DedicatedAllocCalcStatsInfo(VmaStatInfo& outInfo) + { + VMA_ASSERT(m_Type == ALLOCATION_TYPE_DEDICATED); + outInfo.blockCount = 1; + outInfo.allocationCount = 1; + outInfo.unusedRangeCount = 0; + outInfo.usedBytes = m_Size; + outInfo.unusedBytes = 0; + outInfo.allocationSizeMin = outInfo.allocationSizeMax = m_Size; + outInfo.unusedRangeSizeMin = UINT64_MAX; + outInfo.unusedRangeSizeMax = 0; + } + + void BlockAllocMap(); + void BlockAllocUnmap(); + VkResult DedicatedAllocMap(VmaAllocator hAllocator, void** ppData); + void DedicatedAllocUnmap(VmaAllocator hAllocator); + +#if VMA_STATS_STRING_ENABLED + uint32_t GetCreationFrameIndex() const { return m_CreationFrameIndex; } + uint32_t GetBufferImageUsage() const { return m_BufferImageUsage; } + + void InitBufferImageUsage(uint32_t bufferImageUsage) + { + VMA_ASSERT(m_BufferImageUsage == 0); + m_BufferImageUsage = bufferImageUsage; + } + + void PrintParameters(class VmaJsonWriter& json) const; +#endif + +private: + VkDeviceSize m_Alignment; + VkDeviceSize m_Size; + void* m_pUserData; + VMA_ATOMIC_UINT32 m_LastUseFrameIndex; + uint8_t m_Type; // ALLOCATION_TYPE + uint8_t m_SuballocationType; // VmaSuballocationType + // Bit 0x80 is set when allocation was created with VMA_ALLOCATION_CREATE_MAPPED_BIT. + // Bits with mask 0x7F are reference counter for vmaMapMemory()/vmaUnmapMemory(). + uint8_t m_MapCount; + uint8_t m_Flags; // enum FLAGS + + // Allocation out of VmaDeviceMemoryBlock. + struct BlockAllocation + { + VmaPool m_hPool; // Null if belongs to general memory. + VmaDeviceMemoryBlock* m_Block; + VkDeviceSize m_Offset; + bool m_CanBecomeLost; + }; + + // Allocation for an object that has its own private VkDeviceMemory. + struct DedicatedAllocation + { + uint32_t m_MemoryTypeIndex; + VkDeviceMemory m_hMemory; + void* m_pMappedData; // Not null means memory is mapped. + }; + + union + { + // Allocation out of VmaDeviceMemoryBlock. + BlockAllocation m_BlockAllocation; + // Allocation for an object that has its own private VkDeviceMemory. + DedicatedAllocation m_DedicatedAllocation; + }; + +#if VMA_STATS_STRING_ENABLED + uint32_t m_CreationFrameIndex; + uint32_t m_BufferImageUsage; // 0 if unknown. +#endif + + void FreeUserDataString(VmaAllocator hAllocator); +}; + +/* +Represents a region of VmaDeviceMemoryBlock that is either assigned and returned as +allocated memory block or free. +*/ +struct VmaSuballocation +{ + VkDeviceSize offset; + VkDeviceSize size; + VmaAllocation hAllocation; + VmaSuballocationType type; +}; + +// Comparator for offsets. +struct VmaSuballocationOffsetLess +{ + bool operator()(const VmaSuballocation& lhs, const VmaSuballocation& rhs) const + { + return lhs.offset < rhs.offset; + } +}; +struct VmaSuballocationOffsetGreater +{ + bool operator()(const VmaSuballocation& lhs, const VmaSuballocation& rhs) const + { + return lhs.offset > rhs.offset; + } +}; + +typedef VmaList< VmaSuballocation, VmaStlAllocator<VmaSuballocation> > VmaSuballocationList; + +// Cost of one additional allocation lost, as equivalent in bytes. +static const VkDeviceSize VMA_LOST_ALLOCATION_COST = 1048576; + +/* +Parameters of planned allocation inside a VmaDeviceMemoryBlock. + +If canMakeOtherLost was false: +- item points to a FREE suballocation. +- itemsToMakeLostCount is 0. + +If canMakeOtherLost was true: +- item points to first of sequence of suballocations, which are either FREE, + or point to VmaAllocations that can become lost. +- itemsToMakeLostCount is the number of VmaAllocations that need to be made lost for + the requested allocation to succeed. +*/ +struct VmaAllocationRequest +{ + VkDeviceSize offset; + VkDeviceSize sumFreeSize; // Sum size of free items that overlap with proposed allocation. + VkDeviceSize sumItemSize; // Sum size of items to make lost that overlap with proposed allocation. + VmaSuballocationList::iterator item; + size_t itemsToMakeLostCount; + void* customData; + + VkDeviceSize CalcCost() const + { + return sumItemSize + itemsToMakeLostCount * VMA_LOST_ALLOCATION_COST; + } +}; + +/* +Data structure used for bookkeeping of allocations and unused ranges of memory +in a single VkDeviceMemory block. +*/ +class VmaBlockMetadata +{ +public: + VmaBlockMetadata(VmaAllocator hAllocator); + virtual ~VmaBlockMetadata() { } + virtual void Init(VkDeviceSize size) { m_Size = size; } + + // Validates all data structures inside this object. If not valid, returns false. + virtual bool Validate() const = 0; + VkDeviceSize GetSize() const { return m_Size; } + virtual size_t GetAllocationCount() const = 0; + virtual VkDeviceSize GetSumFreeSize() const = 0; + virtual VkDeviceSize GetUnusedRangeSizeMax() const = 0; + // Returns true if this block is empty - contains only single free suballocation. + virtual bool IsEmpty() const = 0; + + virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const = 0; + // Shouldn't modify blockCount. + virtual void AddPoolStats(VmaPoolStats& inoutStats) const = 0; + +#if VMA_STATS_STRING_ENABLED + virtual void PrintDetailedMap(class VmaJsonWriter& json) const = 0; +#endif + + // Tries to find a place for suballocation with given parameters inside this block. + // If succeeded, fills pAllocationRequest and returns true. + // If failed, returns false. + virtual bool CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + // Always one of VMA_ALLOCATION_CREATE_STRATEGY_* or VMA_ALLOCATION_INTERNAL_STRATEGY_* flags. + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) = 0; + + virtual bool MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest) = 0; + + virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount) = 0; + + virtual VkResult CheckCorruption(const void* pBlockData) = 0; + + // Makes actual allocation based on request. Request must already be checked and valid. + virtual void Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation) = 0; + + // Frees suballocation assigned to given memory region. + virtual void Free(const VmaAllocation allocation) = 0; + virtual void FreeAtOffset(VkDeviceSize offset) = 0; + + // Tries to resize (grow or shrink) space for given allocation, in place. + virtual bool ResizeAllocation(const VmaAllocation /*alloc*/, VkDeviceSize /*newSize*/) { return false; } + +protected: + const VkAllocationCallbacks* GetAllocationCallbacks() const { return m_pAllocationCallbacks; } + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap_Begin(class VmaJsonWriter& json, + VkDeviceSize unusedBytes, + size_t allocationCount, + size_t unusedRangeCount) const; + void PrintDetailedMap_Allocation(class VmaJsonWriter& json, + VkDeviceSize offset, + VmaAllocation hAllocation) const; + void PrintDetailedMap_UnusedRange(class VmaJsonWriter& json, + VkDeviceSize offset, + VkDeviceSize size) const; + void PrintDetailedMap_End(class VmaJsonWriter& json) const; +#endif + +private: + VkDeviceSize m_Size; + const VkAllocationCallbacks* m_pAllocationCallbacks; +}; + +#define VMA_VALIDATE(cond) do { if(!(cond)) { \ + VMA_ASSERT(0 && "Validation failed: " #cond); \ + return false; \ + } } while(false) + +class VmaBlockMetadata_Generic : public VmaBlockMetadata +{ + VMA_CLASS_NO_COPY(VmaBlockMetadata_Generic) +public: + VmaBlockMetadata_Generic(VmaAllocator hAllocator); + virtual ~VmaBlockMetadata_Generic(); + virtual void Init(VkDeviceSize size); + + virtual bool Validate() const; + virtual size_t GetAllocationCount() const { return m_Suballocations.size() - m_FreeCount; } + virtual VkDeviceSize GetSumFreeSize() const { return m_SumFreeSize; } + virtual VkDeviceSize GetUnusedRangeSizeMax() const; + virtual bool IsEmpty() const; + + virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const; + virtual void AddPoolStats(VmaPoolStats& inoutStats) const; + +#if VMA_STATS_STRING_ENABLED + virtual void PrintDetailedMap(class VmaJsonWriter& json) const; +#endif + + virtual bool CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest); + + virtual bool MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest); + + virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); + + virtual VkResult CheckCorruption(const void* pBlockData); + + virtual void Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation); + + virtual void Free(const VmaAllocation allocation); + virtual void FreeAtOffset(VkDeviceSize offset); + + virtual bool ResizeAllocation(const VmaAllocation alloc, VkDeviceSize newSize); + + //////////////////////////////////////////////////////////////////////////////// + // For defragmentation + + bool IsBufferImageGranularityConflictPossible( + VkDeviceSize bufferImageGranularity, + VmaSuballocationType& inOutPrevSuballocType) const; + +private: + friend class VmaDefragmentationAlgorithm_Generic; + friend class VmaDefragmentationAlgorithm_Fast; + + uint32_t m_FreeCount; + VkDeviceSize m_SumFreeSize; + VmaSuballocationList m_Suballocations; + // Suballocations that are free and have size greater than certain threshold. + // Sorted by size, ascending. + VmaVector< VmaSuballocationList::iterator, VmaStlAllocator< VmaSuballocationList::iterator > > m_FreeSuballocationsBySize; + + bool ValidateFreeSuballocationList() const; + + // Checks if requested suballocation with given parameters can be placed in given pFreeSuballocItem. + // If yes, fills pOffset and returns true. If no, returns false. + bool CheckAllocation( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + VmaSuballocationList::const_iterator suballocItem, + bool canMakeOtherLost, + VkDeviceSize* pOffset, + size_t* itemsToMakeLostCount, + VkDeviceSize* pSumFreeSize, + VkDeviceSize* pSumItemSize) const; + // Given free suballocation, it merges it with following one, which must also be free. + void MergeFreeWithNext(VmaSuballocationList::iterator item); + // Releases given suballocation, making it free. + // Merges it with adjacent free suballocations if applicable. + // Returns iterator to new free suballocation at this place. + VmaSuballocationList::iterator FreeSuballocation(VmaSuballocationList::iterator suballocItem); + // Given free suballocation, it inserts it into sorted list of + // m_FreeSuballocationsBySize if it's suitable. + void RegisterFreeSuballocation(VmaSuballocationList::iterator item); + // Given free suballocation, it removes it from sorted list of + // m_FreeSuballocationsBySize if it's suitable. + void UnregisterFreeSuballocation(VmaSuballocationList::iterator item); +}; + +/* +Allocations and their references in internal data structure look like this: + +if(m_2ndVectorMode == SECOND_VECTOR_EMPTY): + + 0 +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | + | | + | | +GetSize() +-------+ + +if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER): + + 0 +-------+ + | Alloc | 2nd[0] + +-------+ + | Alloc | 2nd[1] + +-------+ + | ... | + +-------+ + | Alloc | 2nd[2nd.size() - 1] + +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | +GetSize() +-------+ + +if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK): + + 0 +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | + | | + | | + +-------+ + | Alloc | 2nd[2nd.size() - 1] + +-------+ + | ... | + +-------+ + | Alloc | 2nd[1] + +-------+ + | Alloc | 2nd[0] +GetSize() +-------+ + +*/ +class VmaBlockMetadata_Linear : public VmaBlockMetadata +{ + VMA_CLASS_NO_COPY(VmaBlockMetadata_Linear) +public: + VmaBlockMetadata_Linear(VmaAllocator hAllocator); + virtual ~VmaBlockMetadata_Linear(); + virtual void Init(VkDeviceSize size); + + virtual bool Validate() const; + virtual size_t GetAllocationCount() const; + virtual VkDeviceSize GetSumFreeSize() const { return m_SumFreeSize; } + virtual VkDeviceSize GetUnusedRangeSizeMax() const; + virtual bool IsEmpty() const { return GetAllocationCount() == 0; } + + virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const; + virtual void AddPoolStats(VmaPoolStats& inoutStats) const; + +#if VMA_STATS_STRING_ENABLED + virtual void PrintDetailedMap(class VmaJsonWriter& json) const; +#endif + + virtual bool CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest); + + virtual bool MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest); + + virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); + + virtual VkResult CheckCorruption(const void* pBlockData); + + virtual void Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation); + + virtual void Free(const VmaAllocation allocation); + virtual void FreeAtOffset(VkDeviceSize offset); + +private: + /* + There are two suballocation vectors, used in ping-pong way. + The one with index m_1stVectorIndex is called 1st. + The one with index (m_1stVectorIndex ^ 1) is called 2nd. + 2nd can be non-empty only when 1st is not empty. + When 2nd is not empty, m_2ndVectorMode indicates its mode of operation. + */ + typedef VmaVector< VmaSuballocation, VmaStlAllocator<VmaSuballocation> > SuballocationVectorType; + + enum SECOND_VECTOR_MODE + { + SECOND_VECTOR_EMPTY, + /* + Suballocations in 2nd vector are created later than the ones in 1st, but they + all have smaller offset. + */ + SECOND_VECTOR_RING_BUFFER, + /* + Suballocations in 2nd vector are upper side of double stack. + They all have offsets higher than those in 1st vector. + Top of this stack means smaller offsets, but higher indices in this vector. + */ + SECOND_VECTOR_DOUBLE_STACK, + }; + + VkDeviceSize m_SumFreeSize; + SuballocationVectorType m_Suballocations0, m_Suballocations1; + uint32_t m_1stVectorIndex; + SECOND_VECTOR_MODE m_2ndVectorMode; + + SuballocationVectorType& AccessSuballocations1st() { return m_1stVectorIndex ? m_Suballocations1 : m_Suballocations0; } + SuballocationVectorType& AccessSuballocations2nd() { return m_1stVectorIndex ? m_Suballocations0 : m_Suballocations1; } + const SuballocationVectorType& AccessSuballocations1st() const { return m_1stVectorIndex ? m_Suballocations1 : m_Suballocations0; } + const SuballocationVectorType& AccessSuballocations2nd() const { return m_1stVectorIndex ? m_Suballocations0 : m_Suballocations1; } + + // Number of items in 1st vector with hAllocation = null at the beginning. + size_t m_1stNullItemsBeginCount; + // Number of other items in 1st vector with hAllocation = null somewhere in the middle. + size_t m_1stNullItemsMiddleCount; + // Number of items in 2nd vector with hAllocation = null. + size_t m_2ndNullItemsCount; + + bool ShouldCompact1st() const; + void CleanupAfterFree(); +}; + +/* +- GetSize() is the original size of allocated memory block. +- m_UsableSize is this size aligned down to a power of two. + All allocations and calculations happen relative to m_UsableSize. +- GetUnusableSize() is the difference between them. + It is repoted as separate, unused range, not available for allocations. + +Node at level 0 has size = m_UsableSize. +Each next level contains nodes with size 2 times smaller than current level. +m_LevelCount is the maximum number of levels to use in the current object. +*/ +class VmaBlockMetadata_Buddy : public VmaBlockMetadata +{ + VMA_CLASS_NO_COPY(VmaBlockMetadata_Buddy) +public: + VmaBlockMetadata_Buddy(VmaAllocator hAllocator); + virtual ~VmaBlockMetadata_Buddy(); + virtual void Init(VkDeviceSize size); + + virtual bool Validate() const; + virtual size_t GetAllocationCount() const { return m_AllocationCount; } + virtual VkDeviceSize GetSumFreeSize() const { return m_SumFreeSize + GetUnusableSize(); } + virtual VkDeviceSize GetUnusedRangeSizeMax() const; + virtual bool IsEmpty() const { return m_Root->type == Node::TYPE_FREE; } + + virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const; + virtual void AddPoolStats(VmaPoolStats& inoutStats) const; + +#if VMA_STATS_STRING_ENABLED + virtual void PrintDetailedMap(class VmaJsonWriter& json) const; +#endif + + virtual bool CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest); + + virtual bool MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest); + + virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); + + virtual VkResult CheckCorruption(const void* /*pBlockData*/) { return VK_ERROR_FEATURE_NOT_PRESENT; } + + virtual void Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation); + + virtual void Free(const VmaAllocation allocation) { FreeAtOffset(allocation, allocation->GetOffset()); } + virtual void FreeAtOffset(VkDeviceSize offset) { FreeAtOffset(VMA_NULL, offset); } + +private: + static const VkDeviceSize MIN_NODE_SIZE = 32; + static const size_t MAX_LEVELS = 30; + + struct ValidationContext + { + size_t calculatedAllocationCount; + size_t calculatedFreeCount; + VkDeviceSize calculatedSumFreeSize; + + ValidationContext() : + calculatedAllocationCount(0), + calculatedFreeCount(0), + calculatedSumFreeSize(0) { } + }; + + struct Node + { + VkDeviceSize offset; + enum TYPE + { + TYPE_FREE, + TYPE_ALLOCATION, + TYPE_SPLIT, + TYPE_COUNT + } type; + Node* parent; + Node* buddy; + + union + { + struct + { + Node* prev; + Node* next; + } free; + struct + { + VmaAllocation alloc; + } allocation; + struct + { + Node* leftChild; + } split; + }; + }; + + // Size of the memory block aligned down to a power of two. + VkDeviceSize m_UsableSize; + uint32_t m_LevelCount; + + Node* m_Root; + struct { + Node* front; + Node* back; + } m_FreeList[MAX_LEVELS]; + // Number of nodes in the tree with type == TYPE_ALLOCATION. + size_t m_AllocationCount; + // Number of nodes in the tree with type == TYPE_FREE. + size_t m_FreeCount; + // This includes space wasted due to internal fragmentation. Doesn't include unusable size. + VkDeviceSize m_SumFreeSize; + + VkDeviceSize GetUnusableSize() const { return GetSize() - m_UsableSize; } + void DeleteNode(Node* node); + bool ValidateNode(ValidationContext& ctx, const Node* parent, const Node* curr, uint32_t level, VkDeviceSize levelNodeSize) const; + uint32_t AllocSizeToLevel(VkDeviceSize allocSize) const; + inline VkDeviceSize LevelToNodeSize(uint32_t level) const { return m_UsableSize >> level; } + // Alloc passed just for validation. Can be null. + void FreeAtOffset(VmaAllocation alloc, VkDeviceSize offset); + void CalcAllocationStatInfoNode(VmaStatInfo& outInfo, const Node* node, VkDeviceSize levelNodeSize) const; + // Adds node to the front of FreeList at given level. + // node->type must be FREE. + // node->free.prev, next can be undefined. + void AddToFreeListFront(uint32_t level, Node* node); + // Removes node from FreeList at given level. + // node->type must be FREE. + // node->free.prev, next stay untouched. + void RemoveFromFreeList(uint32_t level, Node* node); + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMapNode(class VmaJsonWriter& json, const Node* node, VkDeviceSize levelNodeSize) const; +#endif +}; + +/* +Represents a single block of device memory (`VkDeviceMemory`) with all the +data about its regions (aka suballocations, #VmaAllocation), assigned and free. + +Thread-safety: This class must be externally synchronized. +*/ +class VmaDeviceMemoryBlock +{ + VMA_CLASS_NO_COPY(VmaDeviceMemoryBlock) +public: + VmaBlockMetadata* m_pMetadata; + + VmaDeviceMemoryBlock(VmaAllocator hAllocator); + + ~VmaDeviceMemoryBlock() + { + VMA_ASSERT(m_MapCount == 0 && "VkDeviceMemory block is being destroyed while it is still mapped."); + VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); + } + + // Always call after construction. + void Init( + VmaAllocator hAllocator, + uint32_t newMemoryTypeIndex, + VkDeviceMemory newMemory, + VkDeviceSize newSize, + uint32_t id, + uint32_t algorithm); + // Always call before destruction. + void Destroy(VmaAllocator allocator); + + VkDeviceMemory GetDeviceMemory() const { return m_hMemory; } + uint32_t GetMemoryTypeIndex() const { return m_MemoryTypeIndex; } + uint32_t GetId() const { return m_Id; } + void* GetMappedData() const { return m_pMappedData; } + + // Validates all data structures inside this object. If not valid, returns false. + bool Validate() const; + + VkResult CheckCorruption(VmaAllocator hAllocator); + + // ppData can be null. + VkResult Map(VmaAllocator hAllocator, uint32_t count, void** ppData); + void Unmap(VmaAllocator hAllocator, uint32_t count); + + VkResult WriteMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); + VkResult ValidateMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); + + VkResult BindBufferMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkBuffer hBuffer); + VkResult BindImageMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkImage hImage); + +private: + uint32_t m_MemoryTypeIndex; + uint32_t m_Id; + VkDeviceMemory m_hMemory; + + /* + Protects access to m_hMemory so it's not used by multiple threads simultaneously, e.g. vkMapMemory, vkBindBufferMemory. + Also protects m_MapCount, m_pMappedData. + Allocations, deallocations, any change in m_pMetadata is protected by parent's VmaBlockVector::m_Mutex. + */ + VMA_MUTEX m_Mutex; + uint32_t m_MapCount; + void* m_pMappedData; +}; + +struct VmaPointerLess +{ + bool operator()(const void* lhs, const void* rhs) const + { + return lhs < rhs; + } +}; + +struct VmaDefragmentationMove +{ + size_t srcBlockIndex; + size_t dstBlockIndex; + VkDeviceSize srcOffset; + VkDeviceSize dstOffset; + VkDeviceSize size; +}; + +class VmaDefragmentationAlgorithm; + +/* +Sequence of VmaDeviceMemoryBlock. Represents memory blocks allocated for a specific +Vulkan memory type. + +Synchronized internally with a mutex. +*/ +struct VmaBlockVector +{ + VMA_CLASS_NO_COPY(VmaBlockVector) +public: + VmaBlockVector( + VmaAllocator hAllocator, + uint32_t memoryTypeIndex, + VkDeviceSize preferredBlockSize, + size_t minBlockCount, + size_t maxBlockCount, + VkDeviceSize bufferImageGranularity, + uint32_t frameInUseCount, + bool isCustomPool, + bool explicitBlockSize, + uint32_t algorithm); + ~VmaBlockVector(); + + VkResult CreateMinBlocks(); + + uint32_t GetMemoryTypeIndex() const { return m_MemoryTypeIndex; } + VkDeviceSize GetPreferredBlockSize() const { return m_PreferredBlockSize; } + VkDeviceSize GetBufferImageGranularity() const { return m_BufferImageGranularity; } + uint32_t GetFrameInUseCount() const { return m_FrameInUseCount; } + uint32_t GetAlgorithm() const { return m_Algorithm; } + + void GetPoolStats(VmaPoolStats* pStats); + + bool IsEmpty() const { return m_Blocks.empty(); } + bool IsCorruptionDetectionEnabled() const; + + VkResult Allocate( + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations); + + void Free( + VmaAllocation hAllocation); + + // Adds statistics of this BlockVector to pStats. + void AddStats(VmaStats* pStats); + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json); +#endif + + void MakePoolAllocationsLost( + uint32_t currentFrameIndex, + size_t* pLostAllocationCount); + VkResult CheckCorruption(); + + // Saves results in pCtx->res. + void Defragment( + class VmaBlockVectorDefragmentationContext* pCtx, + VmaDefragmentationStats* pStats, + VkDeviceSize& maxCpuBytesToMove, uint32_t& maxCpuAllocationsToMove, + VkDeviceSize& maxGpuBytesToMove, uint32_t& maxGpuAllocationsToMove, + VkCommandBuffer commandBuffer); + void DefragmentationEnd( + class VmaBlockVectorDefragmentationContext* pCtx, + VmaDefragmentationStats* pStats); + + //////////////////////////////////////////////////////////////////////////////// + // To be used only while the m_Mutex is locked. Used during defragmentation. + + size_t GetBlockCount() const { return m_Blocks.size(); } + VmaDeviceMemoryBlock* GetBlock(size_t index) const { return m_Blocks[index]; } + size_t CalcAllocationCount() const; + bool IsBufferImageGranularityConflictPossible() const; + +private: + friend class VmaDefragmentationAlgorithm_Generic; + + const VmaAllocator m_hAllocator; + const uint32_t m_MemoryTypeIndex; + const VkDeviceSize m_PreferredBlockSize; + const size_t m_MinBlockCount; + const size_t m_MaxBlockCount; + const VkDeviceSize m_BufferImageGranularity; + const uint32_t m_FrameInUseCount; + const bool m_IsCustomPool; + const bool m_ExplicitBlockSize; + const uint32_t m_Algorithm; + /* There can be at most one allocation that is completely empty - a + hysteresis to avoid pessimistic case of alternating creation and destruction + of a VkDeviceMemory. */ + bool m_HasEmptyBlock; + VMA_RW_MUTEX m_Mutex; + // Incrementally sorted by sumFreeSize, ascending. + VmaVector< VmaDeviceMemoryBlock*, VmaStlAllocator<VmaDeviceMemoryBlock*> > m_Blocks; + uint32_t m_NextBlockId; + + VkDeviceSize CalcMaxBlockSize() const; + + // Finds and removes given block from vector. + void Remove(VmaDeviceMemoryBlock* pBlock); + + // Performs single step in sorting m_Blocks. They may not be fully sorted + // after this call. + void IncrementallySortBlocks(); + + VkResult AllocatePage( + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation); + + // To be used only without CAN_MAKE_OTHER_LOST flag. + VkResult AllocateFromBlock( + VmaDeviceMemoryBlock* pBlock, + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + uint32_t strategy, + VmaAllocation* pAllocation); + + VkResult CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIndex); + + // Saves result to pCtx->res. + void ApplyDefragmentationMovesCpu( + class VmaBlockVectorDefragmentationContext* pDefragCtx, + const VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves); + // Saves result to pCtx->res. + void ApplyDefragmentationMovesGpu( + class VmaBlockVectorDefragmentationContext* pDefragCtx, + const VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkCommandBuffer commandBuffer); + + /* + Used during defragmentation. pDefragmentationStats is optional. It's in/out + - updated with new data. + */ + void FreeEmptyBlocks(VmaDefragmentationStats* pDefragmentationStats); +}; + +struct VmaPool_T +{ + VMA_CLASS_NO_COPY(VmaPool_T) +public: + VmaBlockVector m_BlockVector; + + VmaPool_T( + VmaAllocator hAllocator, + const VmaPoolCreateInfo& createInfo, + VkDeviceSize preferredBlockSize); + ~VmaPool_T(); + + uint32_t GetId() const { return m_Id; } + void SetId(uint32_t id) { VMA_ASSERT(m_Id == 0); m_Id = id; } + +#if VMA_STATS_STRING_ENABLED + //void PrintDetailedMap(class VmaStringBuilder& sb); +#endif + +private: + uint32_t m_Id; +}; + +/* +Performs defragmentation: + +- Updates `pBlockVector->m_pMetadata`. +- Updates allocations by calling ChangeBlockAllocation() or ChangeOffset(). +- Does not move actual data, only returns requested moves as `moves`. +*/ +class VmaDefragmentationAlgorithm +{ + VMA_CLASS_NO_COPY(VmaDefragmentationAlgorithm) +public: + VmaDefragmentationAlgorithm( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex) : + m_hAllocator(hAllocator), + m_pBlockVector(pBlockVector), + m_CurrentFrameIndex(currentFrameIndex) + { + } + virtual ~VmaDefragmentationAlgorithm() + { + } + + virtual void AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged) = 0; + virtual void AddAll() = 0; + + virtual VkResult Defragment( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove) = 0; + + virtual VkDeviceSize GetBytesMoved() const = 0; + virtual uint32_t GetAllocationsMoved() const = 0; + +protected: + VmaAllocator const m_hAllocator; + VmaBlockVector* const m_pBlockVector; + const uint32_t m_CurrentFrameIndex; + + struct AllocationInfo + { + VmaAllocation m_hAllocation; + VkBool32* m_pChanged; + + AllocationInfo() : + m_hAllocation(VK_NULL_HANDLE), + m_pChanged(VMA_NULL) + { + } + AllocationInfo(VmaAllocation hAlloc, VkBool32* pChanged) : + m_hAllocation(hAlloc), + m_pChanged(pChanged) + { + } + }; +}; + +class VmaDefragmentationAlgorithm_Generic : public VmaDefragmentationAlgorithm +{ + VMA_CLASS_NO_COPY(VmaDefragmentationAlgorithm_Generic) +public: + VmaDefragmentationAlgorithm_Generic( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex, + bool overlappingMoveSupported); + virtual ~VmaDefragmentationAlgorithm_Generic(); + + virtual void AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged); + virtual void AddAll() { m_AllAllocations = true; } + + virtual VkResult Defragment( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove); + + virtual VkDeviceSize GetBytesMoved() const { return m_BytesMoved; } + virtual uint32_t GetAllocationsMoved() const { return m_AllocationsMoved; } + +private: + uint32_t m_AllocationCount; + bool m_AllAllocations; + + VkDeviceSize m_BytesMoved; + uint32_t m_AllocationsMoved; + + struct AllocationInfoSizeGreater + { + bool operator()(const AllocationInfo& lhs, const AllocationInfo& rhs) const + { + return lhs.m_hAllocation->GetSize() > rhs.m_hAllocation->GetSize(); + } + }; + + struct AllocationInfoOffsetGreater + { + bool operator()(const AllocationInfo& lhs, const AllocationInfo& rhs) const + { + return lhs.m_hAllocation->GetOffset() > rhs.m_hAllocation->GetOffset(); + } + }; + + struct BlockInfo + { + size_t m_OriginalBlockIndex; + VmaDeviceMemoryBlock* m_pBlock; + bool m_HasNonMovableAllocations; + VmaVector< AllocationInfo, VmaStlAllocator<AllocationInfo> > m_Allocations; + + BlockInfo(const VkAllocationCallbacks* pAllocationCallbacks) : + m_OriginalBlockIndex(SIZE_MAX), + m_pBlock(VMA_NULL), + m_HasNonMovableAllocations(true), + m_Allocations(pAllocationCallbacks) + { + } + + void CalcHasNonMovableAllocations() + { + const size_t blockAllocCount = m_pBlock->m_pMetadata->GetAllocationCount(); + const size_t defragmentAllocCount = m_Allocations.size(); + m_HasNonMovableAllocations = blockAllocCount != defragmentAllocCount; + } + + void SortAllocationsBySizeDescending() + { + VMA_SORT(m_Allocations.begin(), m_Allocations.end(), AllocationInfoSizeGreater()); + } + + void SortAllocationsByOffsetDescending() + { + VMA_SORT(m_Allocations.begin(), m_Allocations.end(), AllocationInfoOffsetGreater()); + } + }; + + struct BlockPointerLess + { + bool operator()(const BlockInfo* pLhsBlockInfo, const VmaDeviceMemoryBlock* pRhsBlock) const + { + return pLhsBlockInfo->m_pBlock < pRhsBlock; + } + bool operator()(const BlockInfo* pLhsBlockInfo, const BlockInfo* pRhsBlockInfo) const + { + return pLhsBlockInfo->m_pBlock < pRhsBlockInfo->m_pBlock; + } + }; + + // 1. Blocks with some non-movable allocations go first. + // 2. Blocks with smaller sumFreeSize go first. + struct BlockInfoCompareMoveDestination + { + bool operator()(const BlockInfo* pLhsBlockInfo, const BlockInfo* pRhsBlockInfo) const + { + if(pLhsBlockInfo->m_HasNonMovableAllocations && !pRhsBlockInfo->m_HasNonMovableAllocations) + { + return true; + } + if(!pLhsBlockInfo->m_HasNonMovableAllocations && pRhsBlockInfo->m_HasNonMovableAllocations) + { + return false; + } + if(pLhsBlockInfo->m_pBlock->m_pMetadata->GetSumFreeSize() < pRhsBlockInfo->m_pBlock->m_pMetadata->GetSumFreeSize()) + { + return true; + } + return false; + } + }; + + typedef VmaVector< BlockInfo*, VmaStlAllocator<BlockInfo*> > BlockInfoVector; + BlockInfoVector m_Blocks; + + VkResult DefragmentRound( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove); + + size_t CalcBlocksWithNonMovableCount() const; + + static bool MoveMakesSense( + size_t dstBlockIndex, VkDeviceSize dstOffset, + size_t srcBlockIndex, VkDeviceSize srcOffset); +}; + +class VmaDefragmentationAlgorithm_Fast : public VmaDefragmentationAlgorithm +{ + VMA_CLASS_NO_COPY(VmaDefragmentationAlgorithm_Fast) +public: + VmaDefragmentationAlgorithm_Fast( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex, + bool overlappingMoveSupported); + virtual ~VmaDefragmentationAlgorithm_Fast(); + + virtual void AddAllocation(VmaAllocation /*hAlloc*/, VkBool32* /*pChanged*/) { ++m_AllocationCount; } + virtual void AddAll() { m_AllAllocations = true; } + + virtual VkResult Defragment( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove); + + virtual VkDeviceSize GetBytesMoved() const { return m_BytesMoved; } + virtual uint32_t GetAllocationsMoved() const { return m_AllocationsMoved; } + +private: + struct BlockInfo + { + size_t origBlockIndex; + }; + + class FreeSpaceDatabase + { + public: + FreeSpaceDatabase() + { + FreeSpace s = {}; + s.blockInfoIndex = SIZE_MAX; + for(size_t i = 0; i < MAX_COUNT; ++i) + { + m_FreeSpaces[i] = s; + } + } + + void Register(size_t blockInfoIndex, VkDeviceSize offset, VkDeviceSize size) + { + if(size < VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + return; + } + + // Find first invalid or the smallest structure. + size_t bestIndex = SIZE_MAX; + for(size_t i = 0; i < MAX_COUNT; ++i) + { + // Empty structure. + if(m_FreeSpaces[i].blockInfoIndex == SIZE_MAX) + { + bestIndex = i; + break; + } + if(m_FreeSpaces[i].size < size && + (bestIndex == SIZE_MAX || m_FreeSpaces[bestIndex].size > m_FreeSpaces[i].size)) + { + bestIndex = i; + } + } + + if(bestIndex != SIZE_MAX) + { + m_FreeSpaces[bestIndex].blockInfoIndex = blockInfoIndex; + m_FreeSpaces[bestIndex].offset = offset; + m_FreeSpaces[bestIndex].size = size; + } + } + + bool Fetch(VkDeviceSize alignment, VkDeviceSize size, + size_t& outBlockInfoIndex, VkDeviceSize& outDstOffset) + { + size_t bestIndex = SIZE_MAX; + VkDeviceSize bestFreeSpaceAfter = 0; + for(size_t i = 0; i < MAX_COUNT; ++i) + { + // Structure is valid. + if(m_FreeSpaces[i].blockInfoIndex != SIZE_MAX) + { + const VkDeviceSize dstOffset = VmaAlignUp(m_FreeSpaces[i].offset, alignment); + // Allocation fits into this structure. + if(dstOffset + size <= m_FreeSpaces[i].offset + m_FreeSpaces[i].size) + { + const VkDeviceSize freeSpaceAfter = (m_FreeSpaces[i].offset + m_FreeSpaces[i].size) - + (dstOffset + size); + if(bestIndex == SIZE_MAX || freeSpaceAfter > bestFreeSpaceAfter) + { + bestIndex = i; + bestFreeSpaceAfter = freeSpaceAfter; + } + } + } + } + + if(bestIndex != SIZE_MAX) + { + outBlockInfoIndex = m_FreeSpaces[bestIndex].blockInfoIndex; + outDstOffset = VmaAlignUp(m_FreeSpaces[bestIndex].offset, alignment); + + if(bestFreeSpaceAfter >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + // Leave this structure for remaining empty space. + const VkDeviceSize alignmentPlusSize = (outDstOffset - m_FreeSpaces[bestIndex].offset) + size; + m_FreeSpaces[bestIndex].offset += alignmentPlusSize; + m_FreeSpaces[bestIndex].size -= alignmentPlusSize; + } + else + { + // This structure becomes invalid. + m_FreeSpaces[bestIndex].blockInfoIndex = SIZE_MAX; + } + + return true; + } + + return false; + } + + private: + static const size_t MAX_COUNT = 4; + + struct FreeSpace + { + size_t blockInfoIndex; // SIZE_MAX means this structure is invalid. + VkDeviceSize offset; + VkDeviceSize size; + } m_FreeSpaces[MAX_COUNT]; + }; + + const bool m_OverlappingMoveSupported; + + uint32_t m_AllocationCount; + bool m_AllAllocations; + + VkDeviceSize m_BytesMoved; + uint32_t m_AllocationsMoved; + + VmaVector< BlockInfo, VmaStlAllocator<BlockInfo> > m_BlockInfos; + + void PreprocessMetadata(); + void PostprocessMetadata(); + void InsertSuballoc(VmaBlockMetadata_Generic* pMetadata, const VmaSuballocation& suballoc); +}; + +struct VmaBlockDefragmentationContext +{ + enum BLOCK_FLAG + { + BLOCK_FLAG_USED = 0x00000001, + }; + uint32_t flags; + VkBuffer hBuffer; + + VmaBlockDefragmentationContext() : + flags(0), + hBuffer(VK_NULL_HANDLE) + { + } +}; + +class VmaBlockVectorDefragmentationContext +{ + VMA_CLASS_NO_COPY(VmaBlockVectorDefragmentationContext) +public: + VkResult res; + bool mutexLocked; + VmaVector< VmaBlockDefragmentationContext, VmaStlAllocator<VmaBlockDefragmentationContext> > blockContexts; + + VmaBlockVectorDefragmentationContext( + VmaAllocator hAllocator, + VmaPool hCustomPool, // Optional. + VmaBlockVector* pBlockVector, + uint32_t currFrameIndex, + uint32_t flags); + ~VmaBlockVectorDefragmentationContext(); + + VmaPool GetCustomPool() const { return m_hCustomPool; } + VmaBlockVector* GetBlockVector() const { return m_pBlockVector; } + VmaDefragmentationAlgorithm* GetAlgorithm() const { return m_pAlgorithm; } + + void AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged); + void AddAll() { m_AllAllocations = true; } + + void Begin(bool overlappingMoveSupported); + +private: + const VmaAllocator m_hAllocator; + // Null if not from custom pool. + const VmaPool m_hCustomPool; + // Redundant, for convenience not to fetch from m_hCustomPool->m_BlockVector or m_hAllocator->m_pBlockVectors. + VmaBlockVector* const m_pBlockVector; + const uint32_t m_CurrFrameIndex; + /*const uint32_t m_AlgorithmFlags;*/ + // Owner of this object. + VmaDefragmentationAlgorithm* m_pAlgorithm; + + struct AllocInfo + { + VmaAllocation hAlloc; + VkBool32* pChanged; + }; + // Used between constructor and Begin. + VmaVector< AllocInfo, VmaStlAllocator<AllocInfo> > m_Allocations; + bool m_AllAllocations; +}; + +struct VmaDefragmentationContext_T +{ +private: + VMA_CLASS_NO_COPY(VmaDefragmentationContext_T) +public: + VmaDefragmentationContext_T( + VmaAllocator hAllocator, + uint32_t currFrameIndex, + uint32_t flags, + VmaDefragmentationStats* pStats); + ~VmaDefragmentationContext_T(); + + void AddPools(uint32_t poolCount, VmaPool* pPools); + void AddAllocations( + uint32_t allocationCount, + VmaAllocation* pAllocations, + VkBool32* pAllocationsChanged); + + /* + Returns: + - `VK_SUCCESS` if succeeded and object can be destroyed immediately. + - `VK_NOT_READY` if succeeded but the object must remain alive until vmaDefragmentationEnd(). + - Negative value if error occured and object can be destroyed immediately. + */ + VkResult Defragment( + VkDeviceSize maxCpuBytesToMove, uint32_t maxCpuAllocationsToMove, + VkDeviceSize maxGpuBytesToMove, uint32_t maxGpuAllocationsToMove, + VkCommandBuffer commandBuffer, VmaDefragmentationStats* pStats); + +private: + const VmaAllocator m_hAllocator; + const uint32_t m_CurrFrameIndex; + const uint32_t m_Flags; + VmaDefragmentationStats* const m_pStats; + // Owner of these objects. + VmaBlockVectorDefragmentationContext* m_DefaultPoolContexts[VK_MAX_MEMORY_TYPES]; + // Owner of these objects. + VmaVector< VmaBlockVectorDefragmentationContext*, VmaStlAllocator<VmaBlockVectorDefragmentationContext*> > m_CustomPoolContexts; +}; + +#if VMA_RECORDING_ENABLED + +class VmaRecorder +{ +public: + VmaRecorder(); + VkResult Init(const VmaRecordSettings& settings, bool useMutex); + void WriteConfiguration( + const VkPhysicalDeviceProperties& devProps, + const VkPhysicalDeviceMemoryProperties& memProps, + bool dedicatedAllocationExtensionEnabled); + ~VmaRecorder(); + + void RecordCreateAllocator(uint32_t frameIndex); + void RecordDestroyAllocator(uint32_t frameIndex); + void RecordCreatePool(uint32_t frameIndex, + const VmaPoolCreateInfo& createInfo, + VmaPool pool); + void RecordDestroyPool(uint32_t frameIndex, VmaPool pool); + void RecordAllocateMemory(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation); + void RecordAllocateMemoryPages(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + const VmaAllocationCreateInfo& createInfo, + uint64_t allocationCount, + const VmaAllocation* pAllocations); + void RecordAllocateMemoryForBuffer(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation); + void RecordAllocateMemoryForImage(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation); + void RecordFreeMemory(uint32_t frameIndex, + VmaAllocation allocation); + void RecordFreeMemoryPages(uint32_t frameIndex, + uint64_t allocationCount, + const VmaAllocation* pAllocations); + void RecordResizeAllocation( + uint32_t frameIndex, + VmaAllocation allocation, + VkDeviceSize newSize); + void RecordSetAllocationUserData(uint32_t frameIndex, + VmaAllocation allocation, + const void* pUserData); + void RecordCreateLostAllocation(uint32_t frameIndex, + VmaAllocation allocation); + void RecordMapMemory(uint32_t frameIndex, + VmaAllocation allocation); + void RecordUnmapMemory(uint32_t frameIndex, + VmaAllocation allocation); + void RecordFlushAllocation(uint32_t frameIndex, + VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size); + void RecordInvalidateAllocation(uint32_t frameIndex, + VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size); + void RecordCreateBuffer(uint32_t frameIndex, + const VkBufferCreateInfo& bufCreateInfo, + const VmaAllocationCreateInfo& allocCreateInfo, + VmaAllocation allocation); + void RecordCreateImage(uint32_t frameIndex, + const VkImageCreateInfo& imageCreateInfo, + const VmaAllocationCreateInfo& allocCreateInfo, + VmaAllocation allocation); + void RecordDestroyBuffer(uint32_t frameIndex, + VmaAllocation allocation); + void RecordDestroyImage(uint32_t frameIndex, + VmaAllocation allocation); + void RecordTouchAllocation(uint32_t frameIndex, + VmaAllocation allocation); + void RecordGetAllocationInfo(uint32_t frameIndex, + VmaAllocation allocation); + void RecordMakePoolAllocationsLost(uint32_t frameIndex, + VmaPool pool); + void RecordDefragmentationBegin(uint32_t frameIndex, + const VmaDefragmentationInfo2& info, + VmaDefragmentationContext ctx); + void RecordDefragmentationEnd(uint32_t frameIndex, + VmaDefragmentationContext ctx); + +private: + struct CallParams + { + uint32_t threadId; + double time; + }; + + class UserDataString + { + public: + UserDataString(VmaAllocationCreateFlags allocFlags, const void* pUserData); + const char* GetString() const { return m_Str; } + + private: + char m_PtrStr[17]; + const char* m_Str; + }; + + bool m_UseMutex; + VmaRecordFlags m_Flags; + FILE* m_File; + VMA_MUTEX m_FileMutex; + int64_t m_Freq; + int64_t m_StartCounter; + + void GetBasicParams(CallParams& outParams); + + // T must be a pointer type, e.g. VmaAllocation, VmaPool. + template<typename T> + void PrintPointerList(uint64_t count, const T* pItems) + { + if(count) + { + fprintf(m_File, "%p", pItems[0]); + for(uint64_t i = 1; i < count; ++i) + { + fprintf(m_File, " %p", pItems[i]); + } + } + } + + void PrintPointerList(uint64_t count, const VmaAllocation* pItems); + void Flush(); +}; + +#endif // #if VMA_RECORDING_ENABLED + +// Main allocator object. +struct VmaAllocator_T +{ + VMA_CLASS_NO_COPY(VmaAllocator_T) +public: + bool m_UseMutex; + bool m_UseKhrDedicatedAllocation; + VkDevice m_hDevice; + bool m_AllocationCallbacksSpecified; + VkAllocationCallbacks m_AllocationCallbacks; + VmaDeviceMemoryCallbacks m_DeviceMemoryCallbacks; + + // Number of bytes free out of limit, or VK_WHOLE_SIZE if no limit for that heap. + VkDeviceSize m_HeapSizeLimit[VK_MAX_MEMORY_HEAPS]; + VMA_MUTEX m_HeapSizeLimitMutex; + + VkPhysicalDeviceProperties m_PhysicalDeviceProperties; + VkPhysicalDeviceMemoryProperties m_MemProps; + + // Default pools. + VmaBlockVector* m_pBlockVectors[VK_MAX_MEMORY_TYPES]; + + // Each vector is sorted by memory (handle value). + typedef VmaVector< VmaAllocation, VmaStlAllocator<VmaAllocation> > AllocationVectorType; + AllocationVectorType* m_pDedicatedAllocations[VK_MAX_MEMORY_TYPES]; + VMA_RW_MUTEX m_DedicatedAllocationsMutex[VK_MAX_MEMORY_TYPES]; + + VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo); + VkResult Init(const VmaAllocatorCreateInfo* pCreateInfo); + ~VmaAllocator_T(); + + const VkAllocationCallbacks* GetAllocationCallbacks() const + { + return m_AllocationCallbacksSpecified ? &m_AllocationCallbacks : 0; + } + const VmaVulkanFunctions& GetVulkanFunctions() const + { + return m_VulkanFunctions; + } + + VkDeviceSize GetBufferImageGranularity() const + { + return VMA_MAX( + static_cast<VkDeviceSize>(VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY), + m_PhysicalDeviceProperties.limits.bufferImageGranularity); + } + + uint32_t GetMemoryHeapCount() const { return m_MemProps.memoryHeapCount; } + uint32_t GetMemoryTypeCount() const { return m_MemProps.memoryTypeCount; } + + uint32_t MemoryTypeIndexToHeapIndex(uint32_t memTypeIndex) const + { + VMA_ASSERT(memTypeIndex < m_MemProps.memoryTypeCount); + return m_MemProps.memoryTypes[memTypeIndex].heapIndex; + } + // True when specific memory type is HOST_VISIBLE but not HOST_COHERENT. + bool IsMemoryTypeNonCoherent(uint32_t memTypeIndex) const + { + return (m_MemProps.memoryTypes[memTypeIndex].propertyFlags & (VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT)) == + VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + } + // Minimum alignment for all allocations in specific memory type. + VkDeviceSize GetMemoryTypeMinAlignment(uint32_t memTypeIndex) const + { + return IsMemoryTypeNonCoherent(memTypeIndex) ? + VMA_MAX((VkDeviceSize)VMA_DEBUG_ALIGNMENT, m_PhysicalDeviceProperties.limits.nonCoherentAtomSize) : + (VkDeviceSize)VMA_DEBUG_ALIGNMENT; + } + + bool IsIntegratedGpu() const + { + return m_PhysicalDeviceProperties.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU; + } + +#if VMA_RECORDING_ENABLED + VmaRecorder* GetRecorder() const { return m_pRecorder; } +#endif + + void GetBufferMemoryRequirements( + VkBuffer hBuffer, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const; + void GetImageMemoryRequirements( + VkImage hImage, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const; + + // Main allocation function. + VkResult AllocateMemory( + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations); + + // Main deallocation function. + void FreeMemory( + size_t allocationCount, + const VmaAllocation* pAllocations); + + VkResult ResizeAllocation( + const VmaAllocation alloc, + VkDeviceSize newSize); + + void CalculateStats(VmaStats* pStats); + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json); +#endif + + VkResult DefragmentationBegin( + const VmaDefragmentationInfo2& info, + VmaDefragmentationStats* pStats, + VmaDefragmentationContext* pContext); + VkResult DefragmentationEnd( + VmaDefragmentationContext context); + + void GetAllocationInfo(VmaAllocation hAllocation, VmaAllocationInfo* pAllocationInfo); + bool TouchAllocation(VmaAllocation hAllocation); + + VkResult CreatePool(const VmaPoolCreateInfo* pCreateInfo, VmaPool* pPool); + void DestroyPool(VmaPool pool); + void GetPoolStats(VmaPool pool, VmaPoolStats* pPoolStats); + + void SetCurrentFrameIndex(uint32_t frameIndex); + uint32_t GetCurrentFrameIndex() const { return m_CurrentFrameIndex.load(); } + + void MakePoolAllocationsLost( + VmaPool hPool, + size_t* pLostAllocationCount); + VkResult CheckPoolCorruption(VmaPool hPool); + VkResult CheckCorruption(uint32_t memoryTypeBits); + + void CreateLostAllocation(VmaAllocation* pAllocation); + + VkResult AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory); + void FreeVulkanMemory(uint32_t memoryType, VkDeviceSize size, VkDeviceMemory hMemory); + + VkResult Map(VmaAllocation hAllocation, void** ppData); + void Unmap(VmaAllocation hAllocation); + + VkResult BindBufferMemory(VmaAllocation hAllocation, VkBuffer hBuffer); + VkResult BindImageMemory(VmaAllocation hAllocation, VkImage hImage); + + void FlushOrInvalidateAllocation( + VmaAllocation hAllocation, + VkDeviceSize offset, VkDeviceSize size, + VMA_CACHE_OPERATION op); + + void FillAllocation(const VmaAllocation hAllocation, uint8_t pattern); + +private: + VkDeviceSize m_PreferredLargeHeapBlockSize; + + VkPhysicalDevice m_PhysicalDevice; + VMA_ATOMIC_UINT32 m_CurrentFrameIndex; + + VMA_RW_MUTEX m_PoolsMutex; + // Protected by m_PoolsMutex. Sorted by pointer value. + VmaVector<VmaPool, VmaStlAllocator<VmaPool> > m_Pools; + uint32_t m_NextPoolId; + + VmaVulkanFunctions m_VulkanFunctions; + +#if VMA_RECORDING_ENABLED + VmaRecorder* m_pRecorder; +#endif + + void ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunctions); + + VkDeviceSize CalcPreferredBlockSize(uint32_t memTypeIndex); + + VkResult AllocateMemoryOfType( + VkDeviceSize size, + VkDeviceSize alignment, + bool dedicatedAllocation, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + const VmaAllocationCreateInfo& createInfo, + uint32_t memTypeIndex, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations); + + // Helper function only to be used inside AllocateDedicatedMemory. + VkResult AllocateDedicatedMemoryPage( + VkDeviceSize size, + VmaSuballocationType suballocType, + uint32_t memTypeIndex, + const VkMemoryAllocateInfo& allocInfo, + bool map, + bool isUserDataString, + void* pUserData, + VmaAllocation* pAllocation); + + // Allocates and registers new VkDeviceMemory specifically for dedicated allocations. + VkResult AllocateDedicatedMemory( + VkDeviceSize size, + VmaSuballocationType suballocType, + uint32_t memTypeIndex, + bool map, + bool isUserDataString, + void* pUserData, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + size_t allocationCount, + VmaAllocation* pAllocations); + + // Tries to free pMemory as Dedicated Memory. Returns true if found and freed. + void FreeDedicatedMemory(VmaAllocation allocation); +}; + +//////////////////////////////////////////////////////////////////////////////// +// Memory allocation #2 after VmaAllocator_T definition + +static void* VmaMalloc(VmaAllocator hAllocator, size_t size, size_t alignment) +{ + return VmaMalloc(&hAllocator->m_AllocationCallbacks, size, alignment); +} + +static void VmaFree(VmaAllocator hAllocator, void* ptr) +{ + VmaFree(&hAllocator->m_AllocationCallbacks, ptr); +} + +template<typename T> +static T* VmaAllocate(VmaAllocator hAllocator) +{ + return (T*)VmaMalloc(hAllocator, sizeof(T), VMA_ALIGN_OF(T)); +} + +template<typename T> +static T* VmaAllocateArray(VmaAllocator hAllocator, size_t count) +{ + return (T*)VmaMalloc(hAllocator, sizeof(T) * count, VMA_ALIGN_OF(T)); +} + +template<typename T> +static void vma_delete(VmaAllocator hAllocator, T* ptr) +{ + if(ptr != VMA_NULL) + { + ptr->~T(); + VmaFree(hAllocator, ptr); + } +} + +template<typename T> +static void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count) +{ + if(ptr != VMA_NULL) + { + for(size_t i = count; i--; ) + ptr[i].~T(); + VmaFree(hAllocator, ptr); + } +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaStringBuilder + +#if VMA_STATS_STRING_ENABLED + +class VmaStringBuilder +{ +public: + VmaStringBuilder(VmaAllocator alloc) : m_Data(VmaStlAllocator<char>(alloc->GetAllocationCallbacks())) { } + size_t GetLength() const { return m_Data.size(); } + const char* GetData() const { return m_Data.data(); } + + void Add(char ch) { m_Data.push_back(ch); } + void Add(const char* pStr); + void AddNewLine() { Add('\n'); } + void AddNumber(uint32_t num); + void AddNumber(uint64_t num); + void AddPointer(const void* ptr); + +private: + VmaVector< char, VmaStlAllocator<char> > m_Data; +}; + +void VmaStringBuilder::Add(const char* pStr) +{ + const size_t strLen = strlen(pStr); + if(strLen > 0) + { + const size_t oldCount = m_Data.size(); + m_Data.resize(oldCount + strLen); + memcpy(m_Data.data() + oldCount, pStr, strLen); + } +} + +void VmaStringBuilder::AddNumber(uint32_t num) +{ + char buf[11]; + VmaUint32ToStr(buf, sizeof(buf), num); + Add(buf); +} + +void VmaStringBuilder::AddNumber(uint64_t num) +{ + char buf[21]; + VmaUint64ToStr(buf, sizeof(buf), num); + Add(buf); +} + +void VmaStringBuilder::AddPointer(const void* ptr) +{ + char buf[21]; + VmaPtrToStr(buf, sizeof(buf), ptr); + Add(buf); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +//////////////////////////////////////////////////////////////////////////////// +// VmaJsonWriter + +#if VMA_STATS_STRING_ENABLED + +class VmaJsonWriter +{ + VMA_CLASS_NO_COPY(VmaJsonWriter) +public: + VmaJsonWriter(const VkAllocationCallbacks* pAllocationCallbacks, VmaStringBuilder& sb); + ~VmaJsonWriter(); + + void BeginObject(bool singleLine = false); + void EndObject(); + + void BeginArray(bool singleLine = false); + void EndArray(); + + void WriteString(const char* pStr); + void BeginString(const char* pStr = VMA_NULL); + void ContinueString(const char* pStr); + void ContinueString(uint32_t n); + void ContinueString(uint64_t n); + void ContinueString_Pointer(const void* ptr); + void EndString(const char* pStr = VMA_NULL); + + void WriteNumber(uint32_t n); + void WriteNumber(uint64_t n); + void WriteBool(bool b); + void WriteNull(); + +private: + static const char* const INDENT; + + enum COLLECTION_TYPE + { + COLLECTION_TYPE_OBJECT, + COLLECTION_TYPE_ARRAY, + }; + struct StackItem + { + COLLECTION_TYPE type; + uint32_t valueCount; + bool singleLineMode; + }; + + VmaStringBuilder& m_SB; + VmaVector< StackItem, VmaStlAllocator<StackItem> > m_Stack; + bool m_InsideString; + + void BeginValue(bool isString); + void WriteIndent(bool oneLess = false); +}; + +const char* const VmaJsonWriter::INDENT = " "; + +VmaJsonWriter::VmaJsonWriter(const VkAllocationCallbacks* pAllocationCallbacks, VmaStringBuilder& sb) : + m_SB(sb), + m_Stack(VmaStlAllocator<StackItem>(pAllocationCallbacks)), + m_InsideString(false) +{ +} + +VmaJsonWriter::~VmaJsonWriter() +{ + VMA_ASSERT(!m_InsideString); + VMA_ASSERT(m_Stack.empty()); +} + +void VmaJsonWriter::BeginObject(bool singleLine) +{ + VMA_ASSERT(!m_InsideString); + + BeginValue(false); + m_SB.Add('{'); + + StackItem item; + item.type = COLLECTION_TYPE_OBJECT; + item.valueCount = 0; + item.singleLineMode = singleLine; + m_Stack.push_back(item); +} + +void VmaJsonWriter::EndObject() +{ + VMA_ASSERT(!m_InsideString); + + WriteIndent(true); + m_SB.Add('}'); + + VMA_ASSERT(!m_Stack.empty() && m_Stack.back().type == COLLECTION_TYPE_OBJECT); + m_Stack.pop_back(); +} + +void VmaJsonWriter::BeginArray(bool singleLine) +{ + VMA_ASSERT(!m_InsideString); + + BeginValue(false); + m_SB.Add('['); + + StackItem item; + item.type = COLLECTION_TYPE_ARRAY; + item.valueCount = 0; + item.singleLineMode = singleLine; + m_Stack.push_back(item); +} + +void VmaJsonWriter::EndArray() +{ + VMA_ASSERT(!m_InsideString); + + WriteIndent(true); + m_SB.Add(']'); + + VMA_ASSERT(!m_Stack.empty() && m_Stack.back().type == COLLECTION_TYPE_ARRAY); + m_Stack.pop_back(); +} + +void VmaJsonWriter::WriteString(const char* pStr) +{ + BeginString(pStr); + EndString(); +} + +void VmaJsonWriter::BeginString(const char* pStr) +{ + VMA_ASSERT(!m_InsideString); + + BeginValue(true); + m_SB.Add('"'); + m_InsideString = true; + if(pStr != VMA_NULL && pStr[0] != '\0') + { + ContinueString(pStr); + } +} + +void VmaJsonWriter::ContinueString(const char* pStr) +{ + VMA_ASSERT(m_InsideString); + + const size_t strLen = strlen(pStr); + for(size_t i = 0; i < strLen; ++i) + { + char ch = pStr[i]; + if(ch == '\\') + { + m_SB.Add("\\\\"); + } + else if(ch == '"') + { + m_SB.Add("\\\""); + } + else if(ch >= 32) + { + m_SB.Add(ch); + } + else switch(ch) + { + case '\b': + m_SB.Add("\\b"); + break; + case '\f': + m_SB.Add("\\f"); + break; + case '\n': + m_SB.Add("\\n"); + break; + case '\r': + m_SB.Add("\\r"); + break; + case '\t': + m_SB.Add("\\t"); + break; + default: + VMA_ASSERT(0 && "Character not currently supported."); + break; + } + } +} + +void VmaJsonWriter::ContinueString(uint32_t n) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::ContinueString(uint64_t n) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::ContinueString_Pointer(const void* ptr) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddPointer(ptr); +} + +void VmaJsonWriter::EndString(const char* pStr) +{ + VMA_ASSERT(m_InsideString); + if(pStr != VMA_NULL && pStr[0] != '\0') + { + ContinueString(pStr); + } + m_SB.Add('"'); + m_InsideString = false; +} + +void VmaJsonWriter::WriteNumber(uint32_t n) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::WriteNumber(uint64_t n) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::WriteBool(bool b) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.Add(b ? "true" : "false"); +} + +void VmaJsonWriter::WriteNull() +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.Add("null"); +} + +void VmaJsonWriter::BeginValue(bool isString) +{ + if(!m_Stack.empty()) + { + StackItem& currItem = m_Stack.back(); + if(currItem.type == COLLECTION_TYPE_OBJECT && + currItem.valueCount % 2 == 0) + { + (void) isString; + VMA_ASSERT(isString); + } + + if(currItem.type == COLLECTION_TYPE_OBJECT && + currItem.valueCount % 2 != 0) + { + m_SB.Add(": "); + } + else if(currItem.valueCount > 0) + { + m_SB.Add(", "); + WriteIndent(); + } + else + { + WriteIndent(); + } + ++currItem.valueCount; + } +} + +void VmaJsonWriter::WriteIndent(bool oneLess) +{ + if(!m_Stack.empty() && !m_Stack.back().singleLineMode) + { + m_SB.AddNewLine(); + + size_t count = m_Stack.size(); + if(count > 0 && oneLess) + { + --count; + } + for(size_t i = 0; i < count; ++i) + { + m_SB.Add(INDENT); + } + } +} + +#endif // #if VMA_STATS_STRING_ENABLED + +//////////////////////////////////////////////////////////////////////////////// + +void VmaAllocation_T::SetUserData(VmaAllocator hAllocator, void* pUserData) +{ + if(IsUserDataString()) + { + VMA_ASSERT(pUserData == VMA_NULL || pUserData != m_pUserData); + + FreeUserDataString(hAllocator); + + if(pUserData != VMA_NULL) + { + const char* const newStrSrc = (char*)pUserData; + const size_t newStrLen = strlen(newStrSrc); + char* const newStrDst = vma_new_array(hAllocator, char, newStrLen + 1); + memcpy(newStrDst, newStrSrc, newStrLen + 1); + m_pUserData = newStrDst; + } + } + else + { + m_pUserData = pUserData; + } +} + +void VmaAllocation_T::ChangeBlockAllocation( + VmaAllocator hAllocator, + VmaDeviceMemoryBlock* block, + VkDeviceSize offset) +{ + VMA_ASSERT(block != VMA_NULL); + VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); + + // Move mapping reference counter from old block to new block. + if(block != m_BlockAllocation.m_Block) + { + uint32_t mapRefCount = m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP; + if(IsPersistentMap()) + ++mapRefCount; + m_BlockAllocation.m_Block->Unmap(hAllocator, mapRefCount); + block->Map(hAllocator, mapRefCount, VMA_NULL); + } + + m_BlockAllocation.m_Block = block; + m_BlockAllocation.m_Offset = offset; +} + +void VmaAllocation_T::ChangeSize(VkDeviceSize newSize) +{ + VMA_ASSERT(newSize > 0); + m_Size = newSize; +} + +void VmaAllocation_T::ChangeOffset(VkDeviceSize newOffset) +{ + VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); + m_BlockAllocation.m_Offset = newOffset; +} + +VkDeviceSize VmaAllocation_T::GetOffset() const +{ + switch(m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Offset; + case ALLOCATION_TYPE_DEDICATED: + return 0; + default: + VMA_ASSERT(0); + return 0; + } +} + +VkDeviceMemory VmaAllocation_T::GetMemory() const +{ + switch(m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->GetDeviceMemory(); + case ALLOCATION_TYPE_DEDICATED: + return m_DedicatedAllocation.m_hMemory; + default: + VMA_ASSERT(0); + return VK_NULL_HANDLE; + } +} + +uint32_t VmaAllocation_T::GetMemoryTypeIndex() const +{ + switch(m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->GetMemoryTypeIndex(); + case ALLOCATION_TYPE_DEDICATED: + return m_DedicatedAllocation.m_MemoryTypeIndex; + default: + VMA_ASSERT(0); + return UINT32_MAX; + } +} + +void* VmaAllocation_T::GetMappedData() const +{ + switch(m_Type) + { + case ALLOCATION_TYPE_BLOCK: + if(m_MapCount != 0) + { + void* pBlockData = m_BlockAllocation.m_Block->GetMappedData(); + VMA_ASSERT(pBlockData != VMA_NULL); + return (char*)pBlockData + m_BlockAllocation.m_Offset; + } + else + { + return VMA_NULL; + } + break; + case ALLOCATION_TYPE_DEDICATED: + VMA_ASSERT((m_DedicatedAllocation.m_pMappedData != VMA_NULL) == (m_MapCount != 0)); + return m_DedicatedAllocation.m_pMappedData; + default: + VMA_ASSERT(0); + return VMA_NULL; + } +} + +bool VmaAllocation_T::CanBecomeLost() const +{ + switch(m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_CanBecomeLost; + case ALLOCATION_TYPE_DEDICATED: + return false; + default: + VMA_ASSERT(0); + return false; + } +} + +VmaPool VmaAllocation_T::GetPool() const +{ + VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); + return m_BlockAllocation.m_hPool; +} + +bool VmaAllocation_T::MakeLost(uint32_t currentFrameIndex, uint32_t frameInUseCount) +{ + VMA_ASSERT(CanBecomeLost()); + + /* + Warning: This is a carefully designed algorithm. + Do not modify unless you really know what you're doing :) + */ + uint32_t localLastUseFrameIndex = GetLastUseFrameIndex(); + for(;;) + { + if(localLastUseFrameIndex == VMA_FRAME_INDEX_LOST) + { + VMA_ASSERT(0); + return false; + } + else if(localLastUseFrameIndex + frameInUseCount >= currentFrameIndex) + { + return false; + } + else // Last use time earlier than current time. + { + if(CompareExchangeLastUseFrameIndex(localLastUseFrameIndex, VMA_FRAME_INDEX_LOST)) + { + // Setting hAllocation.LastUseFrameIndex atomic to VMA_FRAME_INDEX_LOST is enough to mark it as LOST. + // Calling code just needs to unregister this allocation in owning VmaDeviceMemoryBlock. + return true; + } + } + } +} + +#if VMA_STATS_STRING_ENABLED + +// Correspond to values of enum VmaSuballocationType. +static const char* VMA_SUBALLOCATION_TYPE_NAMES[] = { + "FREE", + "UNKNOWN", + "BUFFER", + "IMAGE_UNKNOWN", + "IMAGE_LINEAR", + "IMAGE_OPTIMAL", +}; + +void VmaAllocation_T::PrintParameters(class VmaJsonWriter& json) const +{ + json.WriteString("Type"); + json.WriteString(VMA_SUBALLOCATION_TYPE_NAMES[m_SuballocationType]); + + json.WriteString("Size"); + json.WriteNumber(m_Size); + + if(m_pUserData != VMA_NULL) + { + json.WriteString("UserData"); + if(IsUserDataString()) + { + json.WriteString((const char*)m_pUserData); + } + else + { + json.BeginString(); + json.ContinueString_Pointer(m_pUserData); + json.EndString(); + } + } + + json.WriteString("CreationFrameIndex"); + json.WriteNumber(m_CreationFrameIndex); + + json.WriteString("LastUseFrameIndex"); + json.WriteNumber(GetLastUseFrameIndex()); + + if(m_BufferImageUsage != 0) + { + json.WriteString("Usage"); + json.WriteNumber(m_BufferImageUsage); + } +} + +#endif + +void VmaAllocation_T::FreeUserDataString(VmaAllocator hAllocator) +{ + VMA_ASSERT(IsUserDataString()); + if(m_pUserData != VMA_NULL) + { + char* const oldStr = (char*)m_pUserData; + const size_t oldStrLen = strlen(oldStr); + vma_delete_array(hAllocator, oldStr, oldStrLen + 1); + m_pUserData = VMA_NULL; + } +} + +void VmaAllocation_T::BlockAllocMap() +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_BLOCK); + + if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) < 0x7F) + { + ++m_MapCount; + } + else + { + VMA_ASSERT(0 && "Allocation mapped too many times simultaneously."); + } +} + +void VmaAllocation_T::BlockAllocUnmap() +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_BLOCK); + + if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) != 0) + { + --m_MapCount; + } + else + { + VMA_ASSERT(0 && "Unmapping allocation not previously mapped."); + } +} + +VkResult VmaAllocation_T::DedicatedAllocMap(VmaAllocator hAllocator, void** ppData) +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); + + if(m_MapCount != 0) + { + if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) < 0x7F) + { + VMA_ASSERT(m_DedicatedAllocation.m_pMappedData != VMA_NULL); + *ppData = m_DedicatedAllocation.m_pMappedData; + ++m_MapCount; + return VK_SUCCESS; + } + else + { + VMA_ASSERT(0 && "Dedicated allocation mapped too many times simultaneously."); + return VK_ERROR_MEMORY_MAP_FAILED; + } + } + else + { + VkResult result = (*hAllocator->GetVulkanFunctions().vkMapMemory)( + hAllocator->m_hDevice, + m_DedicatedAllocation.m_hMemory, + 0, // offset + VK_WHOLE_SIZE, + 0, // flags + ppData); + if(result == VK_SUCCESS) + { + m_DedicatedAllocation.m_pMappedData = *ppData; + m_MapCount = 1; + } + return result; + } +} + +void VmaAllocation_T::DedicatedAllocUnmap(VmaAllocator hAllocator) +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); + + if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) != 0) + { + --m_MapCount; + if(m_MapCount == 0) + { + m_DedicatedAllocation.m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)( + hAllocator->m_hDevice, + m_DedicatedAllocation.m_hMemory); + } + } + else + { + VMA_ASSERT(0 && "Unmapping dedicated allocation not previously mapped."); + } +} + +#if VMA_STATS_STRING_ENABLED + +static void VmaPrintStatInfo(VmaJsonWriter& json, const VmaStatInfo& stat) +{ + json.BeginObject(); + + json.WriteString("Blocks"); + json.WriteNumber(stat.blockCount); + + json.WriteString("Allocations"); + json.WriteNumber(stat.allocationCount); + + json.WriteString("UnusedRanges"); + json.WriteNumber(stat.unusedRangeCount); + + json.WriteString("UsedBytes"); + json.WriteNumber(stat.usedBytes); + + json.WriteString("UnusedBytes"); + json.WriteNumber(stat.unusedBytes); + + if(stat.allocationCount > 1) + { + json.WriteString("AllocationSize"); + json.BeginObject(true); + json.WriteString("Min"); + json.WriteNumber(stat.allocationSizeMin); + json.WriteString("Avg"); + json.WriteNumber(stat.allocationSizeAvg); + json.WriteString("Max"); + json.WriteNumber(stat.allocationSizeMax); + json.EndObject(); + } + + if(stat.unusedRangeCount > 1) + { + json.WriteString("UnusedRangeSize"); + json.BeginObject(true); + json.WriteString("Min"); + json.WriteNumber(stat.unusedRangeSizeMin); + json.WriteString("Avg"); + json.WriteNumber(stat.unusedRangeSizeAvg); + json.WriteString("Max"); + json.WriteNumber(stat.unusedRangeSizeMax); + json.EndObject(); + } + + json.EndObject(); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +struct VmaSuballocationItemSizeLess +{ + bool operator()( + const VmaSuballocationList::iterator lhs, + const VmaSuballocationList::iterator rhs) const + { + return lhs->size < rhs->size; + } + bool operator()( + const VmaSuballocationList::iterator lhs, + VkDeviceSize rhsSize) const + { + return lhs->size < rhsSize; + } +}; + + +//////////////////////////////////////////////////////////////////////////////// +// class VmaBlockMetadata + +VmaBlockMetadata::VmaBlockMetadata(VmaAllocator hAllocator) : + m_Size(0), + m_pAllocationCallbacks(hAllocator->GetAllocationCallbacks()) +{ +} + +#if VMA_STATS_STRING_ENABLED + +void VmaBlockMetadata::PrintDetailedMap_Begin(class VmaJsonWriter& json, + VkDeviceSize unusedBytes, + size_t allocationCount, + size_t unusedRangeCount) const +{ + json.BeginObject(); + + json.WriteString("TotalBytes"); + json.WriteNumber(GetSize()); + + json.WriteString("UnusedBytes"); + json.WriteNumber(unusedBytes); + + json.WriteString("Allocations"); + json.WriteNumber((uint64_t)allocationCount); + + json.WriteString("UnusedRanges"); + json.WriteNumber((uint64_t)unusedRangeCount); + + json.WriteString("Suballocations"); + json.BeginArray(); +} + +void VmaBlockMetadata::PrintDetailedMap_Allocation(class VmaJsonWriter& json, + VkDeviceSize offset, + VmaAllocation hAllocation) const +{ + json.BeginObject(true); + + json.WriteString("Offset"); + json.WriteNumber(offset); + + hAllocation->PrintParameters(json); + + json.EndObject(); +} + +void VmaBlockMetadata::PrintDetailedMap_UnusedRange(class VmaJsonWriter& json, + VkDeviceSize offset, + VkDeviceSize size) const +{ + json.BeginObject(true); + + json.WriteString("Offset"); + json.WriteNumber(offset); + + json.WriteString("Type"); + json.WriteString(VMA_SUBALLOCATION_TYPE_NAMES[VMA_SUBALLOCATION_TYPE_FREE]); + + json.WriteString("Size"); + json.WriteNumber(size); + + json.EndObject(); +} + +void VmaBlockMetadata::PrintDetailedMap_End(class VmaJsonWriter& json) const +{ + json.EndArray(); + json.EndObject(); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +//////////////////////////////////////////////////////////////////////////////// +// class VmaBlockMetadata_Generic + +VmaBlockMetadata_Generic::VmaBlockMetadata_Generic(VmaAllocator hAllocator) : + VmaBlockMetadata(hAllocator), + m_FreeCount(0), + m_SumFreeSize(0), + m_Suballocations(VmaStlAllocator<VmaSuballocation>(hAllocator->GetAllocationCallbacks())), + m_FreeSuballocationsBySize(VmaStlAllocator<VmaSuballocationList::iterator>(hAllocator->GetAllocationCallbacks())) +{ +} + +VmaBlockMetadata_Generic::~VmaBlockMetadata_Generic() +{ +} + +void VmaBlockMetadata_Generic::Init(VkDeviceSize size) +{ + VmaBlockMetadata::Init(size); + + m_FreeCount = 1; + m_SumFreeSize = size; + + VmaSuballocation suballoc = {}; + suballoc.offset = 0; + suballoc.size = size; + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + suballoc.hAllocation = VK_NULL_HANDLE; + + VMA_ASSERT(size > VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER); + m_Suballocations.push_back(suballoc); + VmaSuballocationList::iterator suballocItem = m_Suballocations.end(); + --suballocItem; + m_FreeSuballocationsBySize.push_back(suballocItem); +} + +bool VmaBlockMetadata_Generic::Validate() const +{ + VMA_VALIDATE(!m_Suballocations.empty()); + + // Expected offset of new suballocation as calculated from previous ones. + VkDeviceSize calculatedOffset = 0; + // Expected number of free suballocations as calculated from traversing their list. + uint32_t calculatedFreeCount = 0; + // Expected sum size of free suballocations as calculated from traversing their list. + VkDeviceSize calculatedSumFreeSize = 0; + // Expected number of free suballocations that should be registered in + // m_FreeSuballocationsBySize calculated from traversing their list. + size_t freeSuballocationsToRegister = 0; + // True if previous visited suballocation was free. + bool prevFree = false; + + for(VmaSuballocationList::const_iterator suballocItem = m_Suballocations.cbegin(); + suballocItem != m_Suballocations.cend(); + ++suballocItem) + { + const VmaSuballocation& subAlloc = *suballocItem; + + // Actual offset of this suballocation doesn't match expected one. + VMA_VALIDATE(subAlloc.offset == calculatedOffset); + + const bool currFree = (subAlloc.type == VMA_SUBALLOCATION_TYPE_FREE); + // Two adjacent free suballocations are invalid. They should be merged. + VMA_VALIDATE(!prevFree || !currFree); + + VMA_VALIDATE(currFree == (subAlloc.hAllocation == VK_NULL_HANDLE)); + + if(currFree) + { + calculatedSumFreeSize += subAlloc.size; + ++calculatedFreeCount; + if(subAlloc.size >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + ++freeSuballocationsToRegister; + } + + // Margin required between allocations - every free space must be at least that large. +#if VMA_DEBUG_MARGIN + VMA_VALIDATE(subAlloc.size >= VMA_DEBUG_MARGIN); +#endif + } + else + { + VMA_VALIDATE(subAlloc.hAllocation->GetOffset() == subAlloc.offset); + VMA_VALIDATE(subAlloc.hAllocation->GetSize() == subAlloc.size); + + // Margin required between allocations - previous allocation must be free. + VMA_VALIDATE(VMA_DEBUG_MARGIN == 0 || prevFree); + } + + calculatedOffset += subAlloc.size; + prevFree = currFree; + } + + // Number of free suballocations registered in m_FreeSuballocationsBySize doesn't + // match expected one. + VMA_VALIDATE(m_FreeSuballocationsBySize.size() == freeSuballocationsToRegister); + + VkDeviceSize lastSize = 0; + for(size_t i = 0; i < m_FreeSuballocationsBySize.size(); ++i) + { + VmaSuballocationList::iterator suballocItem = m_FreeSuballocationsBySize[i]; + + // Only free suballocations can be registered in m_FreeSuballocationsBySize. + VMA_VALIDATE(suballocItem->type == VMA_SUBALLOCATION_TYPE_FREE); + // They must be sorted by size ascending. + VMA_VALIDATE(suballocItem->size >= lastSize); + + lastSize = suballocItem->size; + } + + // Check if totals match calculacted values. + VMA_VALIDATE(ValidateFreeSuballocationList()); + VMA_VALIDATE(calculatedOffset == GetSize()); + VMA_VALIDATE(calculatedSumFreeSize == m_SumFreeSize); + VMA_VALIDATE(calculatedFreeCount == m_FreeCount); + + return true; +} + +VkDeviceSize VmaBlockMetadata_Generic::GetUnusedRangeSizeMax() const +{ + if(!m_FreeSuballocationsBySize.empty()) + { + return m_FreeSuballocationsBySize.back()->size; + } + else + { + return 0; + } +} + +bool VmaBlockMetadata_Generic::IsEmpty() const +{ + return (m_Suballocations.size() == 1) && (m_FreeCount == 1); +} + +void VmaBlockMetadata_Generic::CalcAllocationStatInfo(VmaStatInfo& outInfo) const +{ + outInfo.blockCount = 1; + + const uint32_t rangeCount = (uint32_t)m_Suballocations.size(); + outInfo.allocationCount = rangeCount - m_FreeCount; + outInfo.unusedRangeCount = m_FreeCount; + + outInfo.unusedBytes = m_SumFreeSize; + outInfo.usedBytes = GetSize() - outInfo.unusedBytes; + + outInfo.allocationSizeMin = UINT64_MAX; + outInfo.allocationSizeMax = 0; + outInfo.unusedRangeSizeMin = UINT64_MAX; + outInfo.unusedRangeSizeMax = 0; + + for(VmaSuballocationList::const_iterator suballocItem = m_Suballocations.cbegin(); + suballocItem != m_Suballocations.cend(); + ++suballocItem) + { + const VmaSuballocation& suballoc = *suballocItem; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + { + outInfo.allocationSizeMin = VMA_MIN(outInfo.allocationSizeMin, suballoc.size); + outInfo.allocationSizeMax = VMA_MAX(outInfo.allocationSizeMax, suballoc.size); + } + else + { + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, suballoc.size); + outInfo.unusedRangeSizeMax = VMA_MAX(outInfo.unusedRangeSizeMax, suballoc.size); + } + } +} + +void VmaBlockMetadata_Generic::AddPoolStats(VmaPoolStats& inoutStats) const +{ + const uint32_t rangeCount = (uint32_t)m_Suballocations.size(); + + inoutStats.size += GetSize(); + inoutStats.unusedSize += m_SumFreeSize; + inoutStats.allocationCount += rangeCount - m_FreeCount; + inoutStats.unusedRangeCount += m_FreeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, GetUnusedRangeSizeMax()); +} + +#if VMA_STATS_STRING_ENABLED + +void VmaBlockMetadata_Generic::PrintDetailedMap(class VmaJsonWriter& json) const +{ + PrintDetailedMap_Begin(json, + m_SumFreeSize, // unusedBytes + m_Suballocations.size() - (size_t)m_FreeCount, // allocationCount + m_FreeCount); // unusedRangeCount + + size_t i = 0; + for(VmaSuballocationList::const_iterator suballocItem = m_Suballocations.cbegin(); + suballocItem != m_Suballocations.cend(); + ++suballocItem, ++i) + { + if(suballocItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + PrintDetailedMap_UnusedRange(json, suballocItem->offset, suballocItem->size); + } + else + { + PrintDetailedMap_Allocation(json, suballocItem->offset, suballocItem->hAllocation); + } + } + + PrintDetailedMap_End(json); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Generic::CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(!upperAddress); + (void) upperAddress; + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(pAllocationRequest != VMA_NULL); + VMA_HEAVY_ASSERT(Validate()); + + // There is not enough total free space in this block to fullfill the request: Early return. + if(canMakeOtherLost == false && + m_SumFreeSize < allocSize + 2 * VMA_DEBUG_MARGIN) + { + return false; + } + + // New algorithm, efficiently searching freeSuballocationsBySize. + const size_t freeSuballocCount = m_FreeSuballocationsBySize.size(); + if(freeSuballocCount > 0) + { + if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT) + { + // Find first free suballocation with size not less than allocSize + 2 * VMA_DEBUG_MARGIN. + VmaSuballocationList::iterator* const it = VmaBinaryFindFirstNotLess( + m_FreeSuballocationsBySize.data(), + m_FreeSuballocationsBySize.data() + freeSuballocCount, + allocSize + 2 * VMA_DEBUG_MARGIN, + VmaSuballocationItemSizeLess()); + size_t index = it - m_FreeSuballocationsBySize.data(); + for(; index < freeSuballocCount; ++index) + { + if(CheckAllocation( + currentFrameIndex, + frameInUseCount, + bufferImageGranularity, + allocSize, + allocAlignment, + allocType, + m_FreeSuballocationsBySize[index], + false, // canMakeOtherLost + &pAllocationRequest->offset, + &pAllocationRequest->itemsToMakeLostCount, + &pAllocationRequest->sumFreeSize, + &pAllocationRequest->sumItemSize)) + { + pAllocationRequest->item = m_FreeSuballocationsBySize[index]; + return true; + } + } + } + else if(strategy == VMA_ALLOCATION_INTERNAL_STRATEGY_MIN_OFFSET) + { + for(VmaSuballocationList::iterator it = m_Suballocations.begin(); + it != m_Suballocations.end(); + ++it) + { + if(it->type == VMA_SUBALLOCATION_TYPE_FREE && CheckAllocation( + currentFrameIndex, + frameInUseCount, + bufferImageGranularity, + allocSize, + allocAlignment, + allocType, + it, + false, // canMakeOtherLost + &pAllocationRequest->offset, + &pAllocationRequest->itemsToMakeLostCount, + &pAllocationRequest->sumFreeSize, + &pAllocationRequest->sumItemSize)) + { + pAllocationRequest->item = it; + return true; + } + } + } + else // WORST_FIT, FIRST_FIT + { + // Search staring from biggest suballocations. + for(size_t index = freeSuballocCount; index--; ) + { + if(CheckAllocation( + currentFrameIndex, + frameInUseCount, + bufferImageGranularity, + allocSize, + allocAlignment, + allocType, + m_FreeSuballocationsBySize[index], + false, // canMakeOtherLost + &pAllocationRequest->offset, + &pAllocationRequest->itemsToMakeLostCount, + &pAllocationRequest->sumFreeSize, + &pAllocationRequest->sumItemSize)) + { + pAllocationRequest->item = m_FreeSuballocationsBySize[index]; + return true; + } + } + } + } + + if(canMakeOtherLost) + { + // Brute-force algorithm. TODO: Come up with something better. + + pAllocationRequest->sumFreeSize = VK_WHOLE_SIZE; + pAllocationRequest->sumItemSize = VK_WHOLE_SIZE; + + VmaAllocationRequest tmpAllocRequest = {}; + for(VmaSuballocationList::iterator suballocIt = m_Suballocations.begin(); + suballocIt != m_Suballocations.end(); + ++suballocIt) + { + if(suballocIt->type == VMA_SUBALLOCATION_TYPE_FREE || + suballocIt->hAllocation->CanBecomeLost()) + { + if(CheckAllocation( + currentFrameIndex, + frameInUseCount, + bufferImageGranularity, + allocSize, + allocAlignment, + allocType, + suballocIt, + canMakeOtherLost, + &tmpAllocRequest.offset, + &tmpAllocRequest.itemsToMakeLostCount, + &tmpAllocRequest.sumFreeSize, + &tmpAllocRequest.sumItemSize)) + { + tmpAllocRequest.item = suballocIt; + + if(tmpAllocRequest.CalcCost() < pAllocationRequest->CalcCost() || + strategy == VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT) + { + *pAllocationRequest = tmpAllocRequest; + } + } + } + } + + if(pAllocationRequest->sumItemSize != VK_WHOLE_SIZE) + { + return true; + } + } + + return false; +} + +bool VmaBlockMetadata_Generic::MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest) +{ + while(pAllocationRequest->itemsToMakeLostCount > 0) + { + if(pAllocationRequest->item->type == VMA_SUBALLOCATION_TYPE_FREE) + { + ++pAllocationRequest->item; + } + VMA_ASSERT(pAllocationRequest->item != m_Suballocations.end()); + VMA_ASSERT(pAllocationRequest->item->hAllocation != VK_NULL_HANDLE); + VMA_ASSERT(pAllocationRequest->item->hAllocation->CanBecomeLost()); + if(pAllocationRequest->item->hAllocation->MakeLost(currentFrameIndex, frameInUseCount)) + { + pAllocationRequest->item = FreeSuballocation(pAllocationRequest->item); + --pAllocationRequest->itemsToMakeLostCount; + } + else + { + return false; + } + } + + VMA_HEAVY_ASSERT(Validate()); + VMA_ASSERT(pAllocationRequest->item != m_Suballocations.end()); + VMA_ASSERT(pAllocationRequest->item->type == VMA_SUBALLOCATION_TYPE_FREE); + + return true; +} + +uint32_t VmaBlockMetadata_Generic::MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount) +{ + uint32_t lostAllocationCount = 0; + for(VmaSuballocationList::iterator it = m_Suballocations.begin(); + it != m_Suballocations.end(); + ++it) + { + if(it->type != VMA_SUBALLOCATION_TYPE_FREE && + it->hAllocation->CanBecomeLost() && + it->hAllocation->MakeLost(currentFrameIndex, frameInUseCount)) + { + it = FreeSuballocation(it); + ++lostAllocationCount; + } + } + return lostAllocationCount; +} + +VkResult VmaBlockMetadata_Generic::CheckCorruption(const void* pBlockData) +{ + for(VmaSuballocationList::iterator it = m_Suballocations.begin(); + it != m_Suballocations.end(); + ++it) + { + if(it->type != VMA_SUBALLOCATION_TYPE_FREE) + { + if(!VmaValidateMagicValue(pBlockData, it->offset - VMA_DEBUG_MARGIN)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED BEFORE VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + if(!VmaValidateMagicValue(pBlockData, it->offset + it->size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + } + } + + return VK_SUCCESS; +} + +void VmaBlockMetadata_Generic::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation) +{ + VMA_ASSERT(!upperAddress); + (void) upperAddress; + VMA_ASSERT(request.item != m_Suballocations.end()); + VmaSuballocation& suballoc = *request.item; + // Given suballocation is a free block. + VMA_ASSERT(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + // Given offset is inside this suballocation. + VMA_ASSERT(request.offset >= suballoc.offset); + const VkDeviceSize paddingBegin = request.offset - suballoc.offset; + VMA_ASSERT(suballoc.size >= paddingBegin + allocSize); + const VkDeviceSize paddingEnd = suballoc.size - paddingBegin - allocSize; + + // Unregister this free suballocation from m_FreeSuballocationsBySize and update + // it to become used. + UnregisterFreeSuballocation(request.item); + + suballoc.offset = request.offset; + suballoc.size = allocSize; + suballoc.type = type; + suballoc.hAllocation = hAllocation; + + // If there are any free bytes remaining at the end, insert new free suballocation after current one. + if(paddingEnd) + { + VmaSuballocation paddingSuballoc = {}; + paddingSuballoc.offset = request.offset + allocSize; + paddingSuballoc.size = paddingEnd; + paddingSuballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + VmaSuballocationList::iterator next = request.item; + ++next; + const VmaSuballocationList::iterator paddingEndItem = + m_Suballocations.insert(next, paddingSuballoc); + RegisterFreeSuballocation(paddingEndItem); + } + + // If there are any free bytes remaining at the beginning, insert new free suballocation before current one. + if(paddingBegin) + { + VmaSuballocation paddingSuballoc = {}; + paddingSuballoc.offset = request.offset - paddingBegin; + paddingSuballoc.size = paddingBegin; + paddingSuballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + const VmaSuballocationList::iterator paddingBeginItem = + m_Suballocations.insert(request.item, paddingSuballoc); + RegisterFreeSuballocation(paddingBeginItem); + } + + // Update totals. + m_FreeCount = m_FreeCount - 1; + if(paddingBegin > 0) + { + ++m_FreeCount; + } + if(paddingEnd > 0) + { + ++m_FreeCount; + } + m_SumFreeSize -= allocSize; +} + +void VmaBlockMetadata_Generic::Free(const VmaAllocation allocation) +{ + for(VmaSuballocationList::iterator suballocItem = m_Suballocations.begin(); + suballocItem != m_Suballocations.end(); + ++suballocItem) + { + VmaSuballocation& suballoc = *suballocItem; + if(suballoc.hAllocation == allocation) + { + FreeSuballocation(suballocItem); + VMA_HEAVY_ASSERT(Validate()); + return; + } + } + VMA_ASSERT(0 && "Not found!"); +} + +void VmaBlockMetadata_Generic::FreeAtOffset(VkDeviceSize offset) +{ + for(VmaSuballocationList::iterator suballocItem = m_Suballocations.begin(); + suballocItem != m_Suballocations.end(); + ++suballocItem) + { + VmaSuballocation& suballoc = *suballocItem; + if(suballoc.offset == offset) + { + FreeSuballocation(suballocItem); + return; + } + } + VMA_ASSERT(0 && "Not found!"); +} + +bool VmaBlockMetadata_Generic::ResizeAllocation(const VmaAllocation alloc, VkDeviceSize newSize) +{ + typedef VmaSuballocationList::iterator iter_type; + for(iter_type suballocItem = m_Suballocations.begin(); + suballocItem != m_Suballocations.end(); + ++suballocItem) + { + VmaSuballocation& suballoc = *suballocItem; + if(suballoc.hAllocation == alloc) + { + iter_type nextItem = suballocItem; + ++nextItem; + + // Should have been ensured on higher level. + VMA_ASSERT(newSize != alloc->GetSize() && newSize > 0); + + // Shrinking. + if(newSize < alloc->GetSize()) + { + const VkDeviceSize sizeDiff = suballoc.size - newSize; + + // There is next item. + if(nextItem != m_Suballocations.end()) + { + // Next item is free. + if(nextItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + // Grow this next item backward. + UnregisterFreeSuballocation(nextItem); + nextItem->offset -= sizeDiff; + nextItem->size += sizeDiff; + RegisterFreeSuballocation(nextItem); + } + // Next item is not free. + else + { + // Create free item after current one. + VmaSuballocation newFreeSuballoc; + newFreeSuballoc.hAllocation = VK_NULL_HANDLE; + newFreeSuballoc.offset = suballoc.offset + newSize; + newFreeSuballoc.size = sizeDiff; + newFreeSuballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + iter_type newFreeSuballocIt = m_Suballocations.insert(nextItem, newFreeSuballoc); + RegisterFreeSuballocation(newFreeSuballocIt); + + ++m_FreeCount; + } + } + // This is the last item. + else + { + // Create free item at the end. + VmaSuballocation newFreeSuballoc; + newFreeSuballoc.hAllocation = VK_NULL_HANDLE; + newFreeSuballoc.offset = suballoc.offset + newSize; + newFreeSuballoc.size = sizeDiff; + newFreeSuballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + m_Suballocations.push_back(newFreeSuballoc); + + iter_type newFreeSuballocIt = m_Suballocations.end(); + RegisterFreeSuballocation(--newFreeSuballocIt); + + ++m_FreeCount; + } + + suballoc.size = newSize; + m_SumFreeSize += sizeDiff; + } + // Growing. + else + { + const VkDeviceSize sizeDiff = newSize - suballoc.size; + + // There is next item. + if(nextItem != m_Suballocations.end()) + { + // Next item is free. + if(nextItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + // There is not enough free space, including margin. + if(nextItem->size < sizeDiff + VMA_DEBUG_MARGIN) + { + return false; + } + + // There is more free space than required. + if(nextItem->size > sizeDiff) + { + // Move and shrink this next item. + UnregisterFreeSuballocation(nextItem); + nextItem->offset += sizeDiff; + nextItem->size -= sizeDiff; + RegisterFreeSuballocation(nextItem); + } + // There is exactly the amount of free space required. + else + { + // Remove this next free item. + UnregisterFreeSuballocation(nextItem); + m_Suballocations.erase(nextItem); + --m_FreeCount; + } + } + // Next item is not free - there is no space to grow. + else + { + return false; + } + } + // This is the last item - there is no space to grow. + else + { + return false; + } + + suballoc.size = newSize; + m_SumFreeSize -= sizeDiff; + } + + // We cannot call Validate() here because alloc object is updated to new size outside of this call. + return true; + } + } + VMA_ASSERT(0 && "Not found!"); + return false; +} + +bool VmaBlockMetadata_Generic::ValidateFreeSuballocationList() const +{ + VkDeviceSize lastSize = 0; + for(size_t i = 0, count = m_FreeSuballocationsBySize.size(); i < count; ++i) + { + const VmaSuballocationList::iterator it = m_FreeSuballocationsBySize[i]; + + VMA_VALIDATE(it->type == VMA_SUBALLOCATION_TYPE_FREE); + VMA_VALIDATE(it->size >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER); + VMA_VALIDATE(it->size >= lastSize); + lastSize = it->size; + } + return true; +} + +bool VmaBlockMetadata_Generic::CheckAllocation( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + VmaSuballocationList::const_iterator suballocItem, + bool canMakeOtherLost, + VkDeviceSize* pOffset, + size_t* itemsToMakeLostCount, + VkDeviceSize* pSumFreeSize, + VkDeviceSize* pSumItemSize) const +{ + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(suballocItem != m_Suballocations.cend()); + VMA_ASSERT(pOffset != VMA_NULL); + + *itemsToMakeLostCount = 0; + *pSumFreeSize = 0; + *pSumItemSize = 0; + + if(canMakeOtherLost) + { + if(suballocItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + *pSumFreeSize = suballocItem->size; + } + else + { + if(suballocItem->hAllocation->CanBecomeLost() && + suballocItem->hAllocation->GetLastUseFrameIndex() + frameInUseCount < currentFrameIndex) + { + ++*itemsToMakeLostCount; + *pSumItemSize = suballocItem->size; + } + else + { + return false; + } + } + + // Remaining size is too small for this request: Early return. + if(GetSize() - suballocItem->offset < allocSize) + { + return false; + } + + // Start from offset equal to beginning of this suballocation. + *pOffset = suballocItem->offset; + + // Apply VMA_DEBUG_MARGIN at the beginning. + if(VMA_DEBUG_MARGIN > 0) + { + *pOffset += VMA_DEBUG_MARGIN; + } + + // Apply alignment. + *pOffset = VmaAlignUp(*pOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if(bufferImageGranularity > 1) + { + bool bufferImageGranularityConflict = false; + VmaSuballocationList::const_iterator prevSuballocItem = suballocItem; + while(prevSuballocItem != m_Suballocations.cbegin()) + { + --prevSuballocItem; + const VmaSuballocation& prevSuballoc = *prevSuballocItem; + if(VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, *pOffset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(prevSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if(bufferImageGranularityConflict) + { + *pOffset = VmaAlignUp(*pOffset, bufferImageGranularity); + } + } + + // Now that we have final *pOffset, check if we are past suballocItem. + // If yes, return false - this function should be called for another suballocItem as starting point. + if(*pOffset >= suballocItem->offset + suballocItem->size) + { + return false; + } + + // Calculate padding at the beginning based on current offset. + const VkDeviceSize paddingBegin = *pOffset - suballocItem->offset; + + // Calculate required margin at the end. + const VkDeviceSize requiredEndMargin = VMA_DEBUG_MARGIN; + + const VkDeviceSize totalSize = paddingBegin + allocSize + requiredEndMargin; + // Another early return check. + if(suballocItem->offset + totalSize > GetSize()) + { + return false; + } + + // Advance lastSuballocItem until desired size is reached. + // Update itemsToMakeLostCount. + VmaSuballocationList::const_iterator lastSuballocItem = suballocItem; + if(totalSize > suballocItem->size) + { + VkDeviceSize remainingSize = totalSize - suballocItem->size; + while(remainingSize > 0) + { + ++lastSuballocItem; + if(lastSuballocItem == m_Suballocations.cend()) + { + return false; + } + if(lastSuballocItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + *pSumFreeSize += lastSuballocItem->size; + } + else + { + VMA_ASSERT(lastSuballocItem->hAllocation != VK_NULL_HANDLE); + if(lastSuballocItem->hAllocation->CanBecomeLost() && + lastSuballocItem->hAllocation->GetLastUseFrameIndex() + frameInUseCount < currentFrameIndex) + { + ++*itemsToMakeLostCount; + *pSumItemSize += lastSuballocItem->size; + } + else + { + return false; + } + } + remainingSize = (lastSuballocItem->size < remainingSize) ? + remainingSize - lastSuballocItem->size : 0; + } + } + + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, we must mark more allocations lost or fail. + if(bufferImageGranularity > 1) + { + VmaSuballocationList::const_iterator nextSuballocItem = lastSuballocItem; + ++nextSuballocItem; + while(nextSuballocItem != m_Suballocations.cend()) + { + const VmaSuballocation& nextSuballoc = *nextSuballocItem; + if(VmaBlocksOnSamePage(*pOffset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(allocType, nextSuballoc.type)) + { + VMA_ASSERT(nextSuballoc.hAllocation != VK_NULL_HANDLE); + if(nextSuballoc.hAllocation->CanBecomeLost() && + nextSuballoc.hAllocation->GetLastUseFrameIndex() + frameInUseCount < currentFrameIndex) + { + ++*itemsToMakeLostCount; + } + else + { + return false; + } + } + } + else + { + // Already on next page. + break; + } + ++nextSuballocItem; + } + } + } + else + { + const VmaSuballocation& suballoc = *suballocItem; + VMA_ASSERT(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + + *pSumFreeSize = suballoc.size; + + // Size of this suballocation is too small for this request: Early return. + if(suballoc.size < allocSize) + { + return false; + } + + // Start from offset equal to beginning of this suballocation. + *pOffset = suballoc.offset; + + // Apply VMA_DEBUG_MARGIN at the beginning. + if(VMA_DEBUG_MARGIN > 0) + { + *pOffset += VMA_DEBUG_MARGIN; + } + + // Apply alignment. + *pOffset = VmaAlignUp(*pOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if(bufferImageGranularity > 1) + { + bool bufferImageGranularityConflict = false; + VmaSuballocationList::const_iterator prevSuballocItem = suballocItem; + while(prevSuballocItem != m_Suballocations.cbegin()) + { + --prevSuballocItem; + const VmaSuballocation& prevSuballoc = *prevSuballocItem; + if(VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, *pOffset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(prevSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if(bufferImageGranularityConflict) + { + *pOffset = VmaAlignUp(*pOffset, bufferImageGranularity); + } + } + + // Calculate padding at the beginning based on current offset. + const VkDeviceSize paddingBegin = *pOffset - suballoc.offset; + + // Calculate required margin at the end. + const VkDeviceSize requiredEndMargin = VMA_DEBUG_MARGIN; + + // Fail if requested size plus margin before and after is bigger than size of this suballocation. + if(paddingBegin + allocSize + requiredEndMargin > suballoc.size) + { + return false; + } + + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if(bufferImageGranularity > 1) + { + VmaSuballocationList::const_iterator nextSuballocItem = suballocItem; + ++nextSuballocItem; + while(nextSuballocItem != m_Suballocations.cend()) + { + const VmaSuballocation& nextSuballoc = *nextSuballocItem; + if(VmaBlocksOnSamePage(*pOffset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(allocType, nextSuballoc.type)) + { + return false; + } + } + else + { + // Already on next page. + break; + } + ++nextSuballocItem; + } + } + } + + // All tests passed: Success. pOffset is already filled. + return true; +} + +void VmaBlockMetadata_Generic::MergeFreeWithNext(VmaSuballocationList::iterator item) +{ + VMA_ASSERT(item != m_Suballocations.end()); + VMA_ASSERT(item->type == VMA_SUBALLOCATION_TYPE_FREE); + + VmaSuballocationList::iterator nextItem = item; + ++nextItem; + VMA_ASSERT(nextItem != m_Suballocations.end()); + VMA_ASSERT(nextItem->type == VMA_SUBALLOCATION_TYPE_FREE); + + item->size += nextItem->size; + --m_FreeCount; + m_Suballocations.erase(nextItem); +} + +VmaSuballocationList::iterator VmaBlockMetadata_Generic::FreeSuballocation(VmaSuballocationList::iterator suballocItem) +{ + // Change this suballocation to be marked as free. + VmaSuballocation& suballoc = *suballocItem; + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + suballoc.hAllocation = VK_NULL_HANDLE; + + // Update totals. + ++m_FreeCount; + m_SumFreeSize += suballoc.size; + + // Merge with previous and/or next suballocation if it's also free. + bool mergeWithNext = false; + bool mergeWithPrev = false; + + VmaSuballocationList::iterator nextItem = suballocItem; + ++nextItem; + if((nextItem != m_Suballocations.end()) && (nextItem->type == VMA_SUBALLOCATION_TYPE_FREE)) + { + mergeWithNext = true; + } + + VmaSuballocationList::iterator prevItem = suballocItem; + if(suballocItem != m_Suballocations.begin()) + { + --prevItem; + if(prevItem->type == VMA_SUBALLOCATION_TYPE_FREE) + { + mergeWithPrev = true; + } + } + + if(mergeWithNext) + { + UnregisterFreeSuballocation(nextItem); + MergeFreeWithNext(suballocItem); + } + + if(mergeWithPrev) + { + UnregisterFreeSuballocation(prevItem); + MergeFreeWithNext(prevItem); + RegisterFreeSuballocation(prevItem); + return prevItem; + } + else + { + RegisterFreeSuballocation(suballocItem); + return suballocItem; + } +} + +void VmaBlockMetadata_Generic::RegisterFreeSuballocation(VmaSuballocationList::iterator item) +{ + VMA_ASSERT(item->type == VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(item->size > 0); + + // You may want to enable this validation at the beginning or at the end of + // this function, depending on what do you want to check. + VMA_HEAVY_ASSERT(ValidateFreeSuballocationList()); + + if(item->size >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + if(m_FreeSuballocationsBySize.empty()) + { + m_FreeSuballocationsBySize.push_back(item); + } + else + { + VmaVectorInsertSorted<VmaSuballocationItemSizeLess>(m_FreeSuballocationsBySize, item); + } + } + + //VMA_HEAVY_ASSERT(ValidateFreeSuballocationList()); +} + + +void VmaBlockMetadata_Generic::UnregisterFreeSuballocation(VmaSuballocationList::iterator item) +{ + VMA_ASSERT(item->type == VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(item->size > 0); + + // You may want to enable this validation at the beginning or at the end of + // this function, depending on what do you want to check. + VMA_HEAVY_ASSERT(ValidateFreeSuballocationList()); + + if(item->size >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + VmaSuballocationList::iterator* const it = VmaBinaryFindFirstNotLess( + m_FreeSuballocationsBySize.data(), + m_FreeSuballocationsBySize.data() + m_FreeSuballocationsBySize.size(), + item, + VmaSuballocationItemSizeLess()); + for(size_t index = it - m_FreeSuballocationsBySize.data(); + index < m_FreeSuballocationsBySize.size(); + ++index) + { + if(m_FreeSuballocationsBySize[index] == item) + { + VmaVectorRemove(m_FreeSuballocationsBySize, index); + return; + } + VMA_ASSERT((m_FreeSuballocationsBySize[index]->size == item->size) && "Not found."); + } + VMA_ASSERT(0 && "Not found."); + } + + //VMA_HEAVY_ASSERT(ValidateFreeSuballocationList()); +} + +bool VmaBlockMetadata_Generic::IsBufferImageGranularityConflictPossible( + VkDeviceSize bufferImageGranularity, + VmaSuballocationType& inOutPrevSuballocType) const +{ + if(bufferImageGranularity == 1 || IsEmpty()) + { + return false; + } + + VkDeviceSize minAlignment = VK_WHOLE_SIZE; + bool typeConflictFound = false; + for(VmaSuballocationList::const_iterator it = m_Suballocations.cbegin(); + it != m_Suballocations.cend(); + ++it) + { + const VmaSuballocationType suballocType = it->type; + if(suballocType != VMA_SUBALLOCATION_TYPE_FREE) + { + minAlignment = VMA_MIN(minAlignment, it->hAllocation->GetAlignment()); + if(VmaIsBufferImageGranularityConflict(inOutPrevSuballocType, suballocType)) + { + typeConflictFound = true; + } + inOutPrevSuballocType = suballocType; + } + } + + return typeConflictFound || minAlignment >= bufferImageGranularity; +} + +//////////////////////////////////////////////////////////////////////////////// +// class VmaBlockMetadata_Linear + +VmaBlockMetadata_Linear::VmaBlockMetadata_Linear(VmaAllocator hAllocator) : + VmaBlockMetadata(hAllocator), + m_SumFreeSize(0), + m_Suballocations0(VmaStlAllocator<VmaSuballocation>(hAllocator->GetAllocationCallbacks())), + m_Suballocations1(VmaStlAllocator<VmaSuballocation>(hAllocator->GetAllocationCallbacks())), + m_1stVectorIndex(0), + m_2ndVectorMode(SECOND_VECTOR_EMPTY), + m_1stNullItemsBeginCount(0), + m_1stNullItemsMiddleCount(0), + m_2ndNullItemsCount(0) +{ +} + +VmaBlockMetadata_Linear::~VmaBlockMetadata_Linear() +{ +} + +void VmaBlockMetadata_Linear::Init(VkDeviceSize size) +{ + VmaBlockMetadata::Init(size); + m_SumFreeSize = size; +} + +bool VmaBlockMetadata_Linear::Validate() const +{ + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + VMA_VALIDATE(suballocations2nd.empty() == (m_2ndVectorMode == SECOND_VECTOR_EMPTY)); + VMA_VALIDATE(!suballocations1st.empty() || + suballocations2nd.empty() || + m_2ndVectorMode != SECOND_VECTOR_RING_BUFFER); + + if(!suballocations1st.empty()) + { + // Null item at the beginning should be accounted into m_1stNullItemsBeginCount. + VMA_VALIDATE(suballocations1st[m_1stNullItemsBeginCount].hAllocation != VK_NULL_HANDLE); + // Null item at the end should be just pop_back(). + VMA_VALIDATE(suballocations1st.back().hAllocation != VK_NULL_HANDLE); + } + if(!suballocations2nd.empty()) + { + // Null item at the end should be just pop_back(). + VMA_VALIDATE(suballocations2nd.back().hAllocation != VK_NULL_HANDLE); + } + + VMA_VALIDATE(m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount <= suballocations1st.size()); + VMA_VALIDATE(m_2ndNullItemsCount <= suballocations2nd.size()); + + VkDeviceSize sumUsedSize = 0; + const size_t suballoc1stCount = suballocations1st.size(); + VkDeviceSize offset = VMA_DEBUG_MARGIN; + + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + const size_t suballoc2ndCount = suballocations2nd.size(); + size_t nullItem2ndCount = 0; + for(size_t i = 0; i < suballoc2ndCount; ++i) + { + const VmaSuballocation& suballoc = suballocations2nd[i]; + const bool currFree = (suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + + VMA_VALIDATE(currFree == (suballoc.hAllocation == VK_NULL_HANDLE)); + VMA_VALIDATE(suballoc.offset >= offset); + + if(!currFree) + { + VMA_VALIDATE(suballoc.hAllocation->GetOffset() == suballoc.offset); + VMA_VALIDATE(suballoc.hAllocation->GetSize() == suballoc.size); + sumUsedSize += suballoc.size; + } + else + { + ++nullItem2ndCount; + } + + offset = suballoc.offset + suballoc.size + VMA_DEBUG_MARGIN; + } + + VMA_VALIDATE(nullItem2ndCount == m_2ndNullItemsCount); + } + + for(size_t i = 0; i < m_1stNullItemsBeginCount; ++i) + { + const VmaSuballocation& suballoc = suballocations1st[i]; + VMA_VALIDATE(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE && + suballoc.hAllocation == VK_NULL_HANDLE); + } + + size_t nullItem1stCount = m_1stNullItemsBeginCount; + + for(size_t i = m_1stNullItemsBeginCount; i < suballoc1stCount; ++i) + { + const VmaSuballocation& suballoc = suballocations1st[i]; + const bool currFree = (suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + + VMA_VALIDATE(currFree == (suballoc.hAllocation == VK_NULL_HANDLE)); + VMA_VALIDATE(suballoc.offset >= offset); + VMA_VALIDATE(i >= m_1stNullItemsBeginCount || currFree); + + if(!currFree) + { + VMA_VALIDATE(suballoc.hAllocation->GetOffset() == suballoc.offset); + VMA_VALIDATE(suballoc.hAllocation->GetSize() == suballoc.size); + sumUsedSize += suballoc.size; + } + else + { + ++nullItem1stCount; + } + + offset = suballoc.offset + suballoc.size + VMA_DEBUG_MARGIN; + } + VMA_VALIDATE(nullItem1stCount == m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount); + + if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + const size_t suballoc2ndCount = suballocations2nd.size(); + size_t nullItem2ndCount = 0; + for(size_t i = suballoc2ndCount; i--; ) + { + const VmaSuballocation& suballoc = suballocations2nd[i]; + const bool currFree = (suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + + VMA_VALIDATE(currFree == (suballoc.hAllocation == VK_NULL_HANDLE)); + VMA_VALIDATE(suballoc.offset >= offset); + + if(!currFree) + { + VMA_VALIDATE(suballoc.hAllocation->GetOffset() == suballoc.offset); + VMA_VALIDATE(suballoc.hAllocation->GetSize() == suballoc.size); + sumUsedSize += suballoc.size; + } + else + { + ++nullItem2ndCount; + } + + offset = suballoc.offset + suballoc.size + VMA_DEBUG_MARGIN; + } + + VMA_VALIDATE(nullItem2ndCount == m_2ndNullItemsCount); + } + + VMA_VALIDATE(offset <= GetSize()); + VMA_VALIDATE(m_SumFreeSize == GetSize() - sumUsedSize); + + return true; +} + +size_t VmaBlockMetadata_Linear::GetAllocationCount() const +{ + return AccessSuballocations1st().size() - (m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount) + + AccessSuballocations2nd().size() - m_2ndNullItemsCount; +} + +VkDeviceSize VmaBlockMetadata_Linear::GetUnusedRangeSizeMax() const +{ + const VkDeviceSize size = GetSize(); + + /* + We don't consider gaps inside allocation vectors with freed allocations because + they are not suitable for reuse in linear allocator. We consider only space that + is available for new allocations. + */ + if(IsEmpty()) + { + return size; + } + + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + + switch(m_2ndVectorMode) + { + case SECOND_VECTOR_EMPTY: + /* + Available space is after end of 1st, as well as before beginning of 1st (which + whould make it a ring buffer). + */ + { + const size_t suballocations1stCount = suballocations1st.size(); + VMA_ASSERT(suballocations1stCount > m_1stNullItemsBeginCount); + const VmaSuballocation& firstSuballoc = suballocations1st[m_1stNullItemsBeginCount]; + const VmaSuballocation& lastSuballoc = suballocations1st[suballocations1stCount - 1]; + return VMA_MAX( + firstSuballoc.offset, + size - (lastSuballoc.offset + lastSuballoc.size)); + } + break; + + case SECOND_VECTOR_RING_BUFFER: + /* + Available space is only between end of 2nd and beginning of 1st. + */ + { + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + const VmaSuballocation& lastSuballoc2nd = suballocations2nd.back(); + const VmaSuballocation& firstSuballoc1st = suballocations1st[m_1stNullItemsBeginCount]; + return firstSuballoc1st.offset - (lastSuballoc2nd.offset + lastSuballoc2nd.size); + } + break; + + case SECOND_VECTOR_DOUBLE_STACK: + /* + Available space is only between end of 1st and top of 2nd. + */ + { + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + const VmaSuballocation& topSuballoc2nd = suballocations2nd.back(); + const VmaSuballocation& lastSuballoc1st = suballocations1st.back(); + return topSuballoc2nd.offset - (lastSuballoc1st.offset + lastSuballoc1st.size); + } + break; + + default: + VMA_ASSERT(0); + return 0; + } +} + +void VmaBlockMetadata_Linear::CalcAllocationStatInfo(VmaStatInfo& outInfo) const +{ + const VkDeviceSize size = GetSize(); + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + const size_t suballoc1stCount = suballocations1st.size(); + const size_t suballoc2ndCount = suballocations2nd.size(); + + outInfo.blockCount = 1; + outInfo.allocationCount = (uint32_t)GetAllocationCount(); + outInfo.unusedRangeCount = 0; + outInfo.usedBytes = 0; + outInfo.allocationSizeMin = UINT64_MAX; + outInfo.allocationSizeMax = 0; + outInfo.unusedRangeSizeMin = UINT64_MAX; + outInfo.unusedRangeSizeMax = 0; + + VkDeviceSize lastOffset = 0; + + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + const VkDeviceSize freeSpace2ndTo1stEnd = suballocations1st[m_1stNullItemsBeginCount].offset; + size_t nextAlloc2ndIndex = 0; + while(lastOffset < freeSpace2ndTo1stEnd) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc2ndIndex < suballoc2ndCount && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex < suballoc2ndCount) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + outInfo.usedBytes += suballoc.size; + outInfo.allocationSizeMin = VMA_MIN(outInfo.allocationSizeMin, suballoc.size); + outInfo.allocationSizeMax = VMA_MIN(outInfo.allocationSizeMax, suballoc.size); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc2ndIndex; + } + // We are at the end. + else + { + // There is free space from lastOffset to freeSpace2ndTo1stEnd. + if(lastOffset < freeSpace2ndTo1stEnd) + { + const VkDeviceSize unusedRangeSize = freeSpace2ndTo1stEnd - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace2ndTo1stEnd; + } + } + } + + size_t nextAlloc1stIndex = m_1stNullItemsBeginCount; + const VkDeviceSize freeSpace1stTo2ndEnd = + m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK ? suballocations2nd.back().offset : size; + while(lastOffset < freeSpace1stTo2ndEnd) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc1stIndex < suballoc1stCount && + suballocations1st[nextAlloc1stIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc1stIndex; + } + + // Found non-null allocation. + if(nextAlloc1stIndex < suballoc1stCount) + { + const VmaSuballocation& suballoc = suballocations1st[nextAlloc1stIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + outInfo.usedBytes += suballoc.size; + outInfo.allocationSizeMin = VMA_MIN(outInfo.allocationSizeMin, suballoc.size); + outInfo.allocationSizeMax = VMA_MIN(outInfo.allocationSizeMax, suballoc.size); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc1stIndex; + } + // We are at the end. + else + { + // There is free space from lastOffset to freeSpace1stTo2ndEnd. + if(lastOffset < freeSpace1stTo2ndEnd) + { + const VkDeviceSize unusedRangeSize = freeSpace1stTo2ndEnd - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace1stTo2ndEnd; + } + } + + if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + size_t nextAlloc2ndIndex = suballocations2nd.size() - 1; + while(lastOffset < size) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc2ndIndex != SIZE_MAX && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + --nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex != SIZE_MAX) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + outInfo.usedBytes += suballoc.size; + outInfo.allocationSizeMin = VMA_MIN(outInfo.allocationSizeMin, suballoc.size); + outInfo.allocationSizeMax = VMA_MIN(outInfo.allocationSizeMax, suballoc.size); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + --nextAlloc2ndIndex; + } + // We are at the end. + else + { + // There is free space from lastOffset to size. + if(lastOffset < size) + { + const VkDeviceSize unusedRangeSize = size - lastOffset; + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusedRangeSize); + outInfo.unusedRangeSizeMax = VMA_MIN(outInfo.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = size; + } + } + } + + outInfo.unusedBytes = size - outInfo.usedBytes; +} + +void VmaBlockMetadata_Linear::AddPoolStats(VmaPoolStats& inoutStats) const +{ + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + const VkDeviceSize size = GetSize(); + const size_t suballoc1stCount = suballocations1st.size(); + const size_t suballoc2ndCount = suballocations2nd.size(); + + inoutStats.size += size; + + VkDeviceSize lastOffset = 0; + + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + const VkDeviceSize freeSpace2ndTo1stEnd = suballocations1st[m_1stNullItemsBeginCount].offset; + size_t nextAlloc2ndIndex = m_1stNullItemsBeginCount; + while(lastOffset < freeSpace2ndTo1stEnd) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex < suballoc2ndCount && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex < suballoc2ndCount) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < freeSpace2ndTo1stEnd) + { + // There is free space from lastOffset to freeSpace2ndTo1stEnd. + const VkDeviceSize unusedRangeSize = freeSpace2ndTo1stEnd - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace2ndTo1stEnd; + } + } + } + + size_t nextAlloc1stIndex = m_1stNullItemsBeginCount; + const VkDeviceSize freeSpace1stTo2ndEnd = + m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK ? suballocations2nd.back().offset : size; + while(lastOffset < freeSpace1stTo2ndEnd) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc1stIndex < suballoc1stCount && + suballocations1st[nextAlloc1stIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc1stIndex; + } + + // Found non-null allocation. + if(nextAlloc1stIndex < suballoc1stCount) + { + const VmaSuballocation& suballoc = suballocations1st[nextAlloc1stIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc1stIndex; + } + // We are at the end. + else + { + if(lastOffset < freeSpace1stTo2ndEnd) + { + // There is free space from lastOffset to freeSpace1stTo2ndEnd. + const VkDeviceSize unusedRangeSize = freeSpace1stTo2ndEnd - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace1stTo2ndEnd; + } + } + + if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + size_t nextAlloc2ndIndex = suballocations2nd.size() - 1; + while(lastOffset < size) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex != SIZE_MAX && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + --nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex != SIZE_MAX) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + --nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < size) + { + // There is free space from lastOffset to size. + const VkDeviceSize unusedRangeSize = size - lastOffset; + inoutStats.unusedSize += unusedRangeSize; + ++inoutStats.unusedRangeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, unusedRangeSize); + } + + // End of loop. + lastOffset = size; + } + } + } +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata_Linear::PrintDetailedMap(class VmaJsonWriter& json) const +{ + const VkDeviceSize size = GetSize(); + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + const size_t suballoc1stCount = suballocations1st.size(); + const size_t suballoc2ndCount = suballocations2nd.size(); + + // FIRST PASS + + size_t unusedRangeCount = 0; + VkDeviceSize usedBytes = 0; + + VkDeviceSize lastOffset = 0; + + size_t alloc2ndCount = 0; + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + const VkDeviceSize freeSpace2ndTo1stEnd = suballocations1st[m_1stNullItemsBeginCount].offset; + size_t nextAlloc2ndIndex = 0; + while(lastOffset < freeSpace2ndTo1stEnd) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex < suballoc2ndCount && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex < suballoc2ndCount) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + ++unusedRangeCount; + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++alloc2ndCount; + usedBytes += suballoc.size; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < freeSpace2ndTo1stEnd) + { + // There is free space from lastOffset to freeSpace2ndTo1stEnd. + ++unusedRangeCount; + } + + // End of loop. + lastOffset = freeSpace2ndTo1stEnd; + } + } + } + + size_t nextAlloc1stIndex = m_1stNullItemsBeginCount; + size_t alloc1stCount = 0; + const VkDeviceSize freeSpace1stTo2ndEnd = + m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK ? suballocations2nd.back().offset : size; + while(lastOffset < freeSpace1stTo2ndEnd) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc1stIndex < suballoc1stCount && + suballocations1st[nextAlloc1stIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc1stIndex; + } + + // Found non-null allocation. + if(nextAlloc1stIndex < suballoc1stCount) + { + const VmaSuballocation& suballoc = suballocations1st[nextAlloc1stIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + ++unusedRangeCount; + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++alloc1stCount; + usedBytes += suballoc.size; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc1stIndex; + } + // We are at the end. + else + { + if(lastOffset < size) + { + // There is free space from lastOffset to freeSpace1stTo2ndEnd. + ++unusedRangeCount; + } + + // End of loop. + lastOffset = freeSpace1stTo2ndEnd; + } + } + + if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + size_t nextAlloc2ndIndex = suballocations2nd.size() - 1; + while(lastOffset < size) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex != SIZE_MAX && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + --nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex != SIZE_MAX) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + ++unusedRangeCount; + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++alloc2ndCount; + usedBytes += suballoc.size; + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + --nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < size) + { + // There is free space from lastOffset to size. + ++unusedRangeCount; + } + + // End of loop. + lastOffset = size; + } + } + } + + const VkDeviceSize unusedBytes = size - usedBytes; + PrintDetailedMap_Begin(json, unusedBytes, alloc1stCount + alloc2ndCount, unusedRangeCount); + + // SECOND PASS + lastOffset = 0; + + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + const VkDeviceSize freeSpace2ndTo1stEnd = suballocations1st[m_1stNullItemsBeginCount].offset; + size_t nextAlloc2ndIndex = 0; + while(lastOffset < freeSpace2ndTo1stEnd) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex < suballoc2ndCount && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex < suballoc2ndCount) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + PrintDetailedMap_Allocation(json, suballoc.offset, suballoc.hAllocation); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < freeSpace2ndTo1stEnd) + { + // There is free space from lastOffset to freeSpace2ndTo1stEnd. + const VkDeviceSize unusedRangeSize = freeSpace2ndTo1stEnd - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace2ndTo1stEnd; + } + } + } + + nextAlloc1stIndex = m_1stNullItemsBeginCount; + while(lastOffset < freeSpace1stTo2ndEnd) + { + // Find next non-null allocation or move nextAllocIndex to the end. + while(nextAlloc1stIndex < suballoc1stCount && + suballocations1st[nextAlloc1stIndex].hAllocation == VK_NULL_HANDLE) + { + ++nextAlloc1stIndex; + } + + // Found non-null allocation. + if(nextAlloc1stIndex < suballoc1stCount) + { + const VmaSuballocation& suballoc = suballocations1st[nextAlloc1stIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + PrintDetailedMap_Allocation(json, suballoc.offset, suballoc.hAllocation); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc1stIndex; + } + // We are at the end. + else + { + if(lastOffset < freeSpace1stTo2ndEnd) + { + // There is free space from lastOffset to freeSpace1stTo2ndEnd. + const VkDeviceSize unusedRangeSize = freeSpace1stTo2ndEnd - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // End of loop. + lastOffset = freeSpace1stTo2ndEnd; + } + } + + if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + size_t nextAlloc2ndIndex = suballocations2nd.size() - 1; + while(lastOffset < size) + { + // Find next non-null allocation or move nextAlloc2ndIndex to the end. + while(nextAlloc2ndIndex != SIZE_MAX && + suballocations2nd[nextAlloc2ndIndex].hAllocation == VK_NULL_HANDLE) + { + --nextAlloc2ndIndex; + } + + // Found non-null allocation. + if(nextAlloc2ndIndex != SIZE_MAX) + { + const VmaSuballocation& suballoc = suballocations2nd[nextAlloc2ndIndex]; + + // 1. Process free space before this allocation. + if(lastOffset < suballoc.offset) + { + // There is free space from lastOffset to suballoc.offset. + const VkDeviceSize unusedRangeSize = suballoc.offset - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + PrintDetailedMap_Allocation(json, suballoc.offset, suballoc.hAllocation); + + // 3. Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + --nextAlloc2ndIndex; + } + // We are at the end. + else + { + if(lastOffset < size) + { + // There is free space from lastOffset to size. + const VkDeviceSize unusedRangeSize = size - lastOffset; + PrintDetailedMap_UnusedRange(json, lastOffset, unusedRangeSize); + } + + // End of loop. + lastOffset = size; + } + } + } + + PrintDetailedMap_End(json); +} +#endif // #if VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Linear::CreateAllocationRequest( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool canMakeOtherLost, + uint32_t /*strategy*/, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(pAllocationRequest != VMA_NULL); + VMA_HEAVY_ASSERT(Validate()); + + const VkDeviceSize size = GetSize(); + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + if(upperAddress) + { + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + VMA_ASSERT(0 && "Trying to use pool with linear algorithm as double stack, while it is already being used as ring buffer."); + return false; + } + + // Try to allocate before 2nd.back(), or end of block if 2nd.empty(). + if(allocSize > size) + { + return false; + } + VkDeviceSize resultBaseOffset = size - allocSize; + if(!suballocations2nd.empty()) + { + const VmaSuballocation& lastSuballoc = suballocations2nd.back(); + resultBaseOffset = lastSuballoc.offset - allocSize; + if(allocSize > lastSuballoc.offset) + { + return false; + } + } + + // Start from offset equal to end of free space. + VkDeviceSize resultOffset = resultBaseOffset; + + // Apply VMA_DEBUG_MARGIN at the end. + if(VMA_DEBUG_MARGIN > 0) + { +#if VMA_DEBUG_MARGIN + if(resultOffset < VMA_DEBUG_MARGIN) + { + return false; + } +#endif + resultOffset -= VMA_DEBUG_MARGIN; + } + + // Apply alignment. + resultOffset = VmaAlignDown(resultOffset, allocAlignment); + + // Check next suballocations from 2nd for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if(bufferImageGranularity > 1 && !suballocations2nd.empty()) + { + bool bufferImageGranularityConflict = false; + for(size_t nextSuballocIndex = suballocations2nd.size(); nextSuballocIndex--; ) + { + const VmaSuballocation& nextSuballoc = suballocations2nd[nextSuballocIndex]; + if(VmaBlocksOnSamePage(resultOffset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(nextSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if(bufferImageGranularityConflict) + { + resultOffset = VmaAlignDown(resultOffset, bufferImageGranularity); + } + } + + // There is enough free space. + const VkDeviceSize endOf1st = !suballocations1st.empty() ? + suballocations1st.back().offset + suballocations1st.back().size : + 0; + if(endOf1st + VMA_DEBUG_MARGIN <= resultOffset) + { + // Check previous suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if(bufferImageGranularity > 1) + { + for(size_t prevSuballocIndex = suballocations1st.size(); prevSuballocIndex--; ) + { + const VmaSuballocation& prevSuballoc = suballocations1st[prevSuballocIndex]; + if(VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, resultOffset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(allocType, prevSuballoc.type)) + { + return false; + } + } + else + { + // Already on next page. + break; + } + } + } + + // All tests passed: Success. + pAllocationRequest->offset = resultOffset; + pAllocationRequest->sumFreeSize = resultBaseOffset + allocSize - endOf1st; + pAllocationRequest->sumItemSize = 0; + // pAllocationRequest->item unused. + pAllocationRequest->itemsToMakeLostCount = 0; + return true; + } + } + else // !upperAddress + { + if(m_2ndVectorMode == SECOND_VECTOR_EMPTY || m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + // Try to allocate at the end of 1st vector. + + VkDeviceSize resultBaseOffset = 0; + if(!suballocations1st.empty()) + { + const VmaSuballocation& lastSuballoc = suballocations1st.back(); + resultBaseOffset = lastSuballoc.offset + lastSuballoc.size; + } + + // Start from offset equal to beginning of free space. + VkDeviceSize resultOffset = resultBaseOffset; + + // Apply VMA_DEBUG_MARGIN at the beginning. + if(VMA_DEBUG_MARGIN > 0) + { + resultOffset += VMA_DEBUG_MARGIN; + } + + // Apply alignment. + resultOffset = VmaAlignUp(resultOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if(bufferImageGranularity > 1 && !suballocations1st.empty()) + { + bool bufferImageGranularityConflict = false; + for(size_t prevSuballocIndex = suballocations1st.size(); prevSuballocIndex--; ) + { + const VmaSuballocation& prevSuballoc = suballocations1st[prevSuballocIndex]; + if(VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, resultOffset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(prevSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if(bufferImageGranularityConflict) + { + resultOffset = VmaAlignUp(resultOffset, bufferImageGranularity); + } + } + + const VkDeviceSize freeSpaceEnd = m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK ? + suballocations2nd.back().offset : size; + + // There is enough free space at the end after alignment. + if(resultOffset + allocSize + VMA_DEBUG_MARGIN <= freeSpaceEnd) + { + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if(bufferImageGranularity > 1 && m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + for(size_t nextSuballocIndex = suballocations2nd.size(); nextSuballocIndex--; ) + { + const VmaSuballocation& nextSuballoc = suballocations2nd[nextSuballocIndex]; + if(VmaBlocksOnSamePage(resultOffset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(allocType, nextSuballoc.type)) + { + return false; + } + } + else + { + // Already on previous page. + break; + } + } + } + + // All tests passed: Success. + pAllocationRequest->offset = resultOffset; + pAllocationRequest->sumFreeSize = freeSpaceEnd - resultBaseOffset; + pAllocationRequest->sumItemSize = 0; + // pAllocationRequest->item unused. + pAllocationRequest->itemsToMakeLostCount = 0; + return true; + } + } + + // Wrap-around to end of 2nd vector. Try to allocate there, watching for the + // beginning of 1st vector as the end of free space. + if(m_2ndVectorMode == SECOND_VECTOR_EMPTY || m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + VMA_ASSERT(!suballocations1st.empty()); + + VkDeviceSize resultBaseOffset = 0; + if(!suballocations2nd.empty()) + { + const VmaSuballocation& lastSuballoc = suballocations2nd.back(); + resultBaseOffset = lastSuballoc.offset + lastSuballoc.size; + } + + // Start from offset equal to beginning of free space. + VkDeviceSize resultOffset = resultBaseOffset; + + // Apply VMA_DEBUG_MARGIN at the beginning. + if(VMA_DEBUG_MARGIN > 0) + { + resultOffset += VMA_DEBUG_MARGIN; + } + + // Apply alignment. + resultOffset = VmaAlignUp(resultOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if(bufferImageGranularity > 1 && !suballocations2nd.empty()) + { + bool bufferImageGranularityConflict = false; + for(size_t prevSuballocIndex = suballocations2nd.size(); prevSuballocIndex--; ) + { + const VmaSuballocation& prevSuballoc = suballocations2nd[prevSuballocIndex]; + if(VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, resultOffset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(prevSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if(bufferImageGranularityConflict) + { + resultOffset = VmaAlignUp(resultOffset, bufferImageGranularity); + } + } + + pAllocationRequest->itemsToMakeLostCount = 0; + pAllocationRequest->sumItemSize = 0; + size_t index1st = m_1stNullItemsBeginCount; + + if(canMakeOtherLost) + { + while(index1st < suballocations1st.size() && + resultOffset + allocSize + VMA_DEBUG_MARGIN > suballocations1st[index1st].offset) + { + // Next colliding allocation at the beginning of 1st vector found. Try to make it lost. + const VmaSuballocation& suballoc = suballocations1st[index1st]; + if(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE) + { + // No problem. + } + else + { + VMA_ASSERT(suballoc.hAllocation != VK_NULL_HANDLE); + if(suballoc.hAllocation->CanBecomeLost() && + suballoc.hAllocation->GetLastUseFrameIndex() + frameInUseCount < currentFrameIndex) + { + ++pAllocationRequest->itemsToMakeLostCount; + pAllocationRequest->sumItemSize += suballoc.size; + } + else + { + return false; + } + } + ++index1st; + } + + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, we must mark more allocations lost or fail. + if(bufferImageGranularity > 1) + { + while(index1st < suballocations1st.size()) + { + const VmaSuballocation& suballoc = suballocations1st[index1st]; + if(VmaBlocksOnSamePage(resultOffset, allocSize, suballoc.offset, bufferImageGranularity)) + { + if(suballoc.hAllocation != VK_NULL_HANDLE) + { + // Not checking actual VmaIsBufferImageGranularityConflict(allocType, suballoc.type). + if(suballoc.hAllocation->CanBecomeLost() && + suballoc.hAllocation->GetLastUseFrameIndex() + frameInUseCount < currentFrameIndex) + { + ++pAllocationRequest->itemsToMakeLostCount; + pAllocationRequest->sumItemSize += suballoc.size; + } + else + { + return false; + } + } + } + else + { + // Already on next page. + break; + } + ++index1st; + } + } + } + + // There is enough free space at the end after alignment. + if((index1st == suballocations1st.size() && resultOffset + allocSize + VMA_DEBUG_MARGIN < size) || + (index1st < suballocations1st.size() && resultOffset + allocSize + VMA_DEBUG_MARGIN <= suballocations1st[index1st].offset)) + { + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if(bufferImageGranularity > 1) + { + for(size_t nextSuballocIndex = index1st; + nextSuballocIndex < suballocations1st.size(); + nextSuballocIndex++) + { + const VmaSuballocation& nextSuballoc = suballocations1st[nextSuballocIndex]; + if(VmaBlocksOnSamePage(resultOffset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if(VmaIsBufferImageGranularityConflict(allocType, nextSuballoc.type)) + { + return false; + } + } + else + { + // Already on next page. + break; + } + } + } + + // All tests passed: Success. + pAllocationRequest->offset = resultOffset; + pAllocationRequest->sumFreeSize = + (index1st < suballocations1st.size() ? suballocations1st[index1st].offset : size) + - resultBaseOffset + - pAllocationRequest->sumItemSize; + // pAllocationRequest->item unused. + return true; + } + } + } + + return false; +} + +bool VmaBlockMetadata_Linear::MakeRequestedAllocationsLost( + uint32_t currentFrameIndex, + uint32_t frameInUseCount, + VmaAllocationRequest* pAllocationRequest) +{ + if(pAllocationRequest->itemsToMakeLostCount == 0) + { + return true; + } + + VMA_ASSERT(m_2ndVectorMode == SECOND_VECTOR_EMPTY || m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER); + + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + size_t index1st = m_1stNullItemsBeginCount; + size_t madeLostCount = 0; + while(madeLostCount < pAllocationRequest->itemsToMakeLostCount) + { + VMA_ASSERT(index1st < suballocations1st.size()); + VmaSuballocation& suballoc = suballocations1st[index1st]; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + { + VMA_ASSERT(suballoc.hAllocation != VK_NULL_HANDLE); + VMA_ASSERT(suballoc.hAllocation->CanBecomeLost()); + if(suballoc.hAllocation->MakeLost(currentFrameIndex, frameInUseCount)) + { + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + suballoc.hAllocation = VK_NULL_HANDLE; + m_SumFreeSize += suballoc.size; + ++m_1stNullItemsMiddleCount; + ++madeLostCount; + } + else + { + return false; + } + } + ++index1st; + } + + CleanupAfterFree(); + //VMA_HEAVY_ASSERT(Validate()); // Already called by ClanupAfterFree(). + + return true; +} + +uint32_t VmaBlockMetadata_Linear::MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount) +{ + uint32_t lostAllocationCount = 0; + + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + for(size_t i = m_1stNullItemsBeginCount, count = suballocations1st.size(); i < count; ++i) + { + VmaSuballocation& suballoc = suballocations1st[i]; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE && + suballoc.hAllocation->CanBecomeLost() && + suballoc.hAllocation->MakeLost(currentFrameIndex, frameInUseCount)) + { + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + suballoc.hAllocation = VK_NULL_HANDLE; + ++m_1stNullItemsMiddleCount; + m_SumFreeSize += suballoc.size; + ++lostAllocationCount; + } + } + + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + for(size_t i = 0, count = suballocations2nd.size(); i < count; ++i) + { + VmaSuballocation& suballoc = suballocations2nd[i]; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE && + suballoc.hAllocation->CanBecomeLost() && + suballoc.hAllocation->MakeLost(currentFrameIndex, frameInUseCount)) + { + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + suballoc.hAllocation = VK_NULL_HANDLE; + ++m_2ndNullItemsCount; + ++lostAllocationCount; + } + } + + if(lostAllocationCount) + { + CleanupAfterFree(); + } + + return lostAllocationCount; +} + +VkResult VmaBlockMetadata_Linear::CheckCorruption(const void* pBlockData) +{ + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + for(size_t i = m_1stNullItemsBeginCount, count = suballocations1st.size(); i < count; ++i) + { + const VmaSuballocation& suballoc = suballocations1st[i]; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + { + if(!VmaValidateMagicValue(pBlockData, suballoc.offset - VMA_DEBUG_MARGIN)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED BEFORE VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + if(!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + } + } + + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + for(size_t i = 0, count = suballocations2nd.size(); i < count; ++i) + { + const VmaSuballocation& suballoc = suballocations2nd[i]; + if(suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + { + if(!VmaValidateMagicValue(pBlockData, suballoc.offset - VMA_DEBUG_MARGIN)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED BEFORE VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + if(!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_VALIDATION_FAILED_EXT; + } + } + } + + return VK_SUCCESS; +} + +void VmaBlockMetadata_Linear::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + VkDeviceSize allocSize, + bool upperAddress, + VmaAllocation hAllocation) +{ + const VmaSuballocation newSuballoc = { request.offset, allocSize, hAllocation, type }; + + if(upperAddress) + { + VMA_ASSERT(m_2ndVectorMode != SECOND_VECTOR_RING_BUFFER && + "CRITICAL ERROR: Trying to use linear allocator as double stack while it was already used as ring buffer."); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + suballocations2nd.push_back(newSuballoc); + m_2ndVectorMode = SECOND_VECTOR_DOUBLE_STACK; + } + else + { + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + + // First allocation. + if(suballocations1st.empty()) + { + suballocations1st.push_back(newSuballoc); + } + else + { + // New allocation at the end of 1st vector. + if(request.offset >= suballocations1st.back().offset + suballocations1st.back().size) + { + // Check if it fits before the end of the block. + VMA_ASSERT(request.offset + allocSize <= GetSize()); + suballocations1st.push_back(newSuballoc); + } + // New allocation at the end of 2-part ring buffer, so before first allocation from 1st vector. + else if(request.offset + allocSize <= suballocations1st[m_1stNullItemsBeginCount].offset) + { + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + switch(m_2ndVectorMode) + { + case SECOND_VECTOR_EMPTY: + // First allocation from second part ring buffer. + VMA_ASSERT(suballocations2nd.empty()); + m_2ndVectorMode = SECOND_VECTOR_RING_BUFFER; + break; + case SECOND_VECTOR_RING_BUFFER: + // 2-part ring buffer is already started. + VMA_ASSERT(!suballocations2nd.empty()); + break; + case SECOND_VECTOR_DOUBLE_STACK: + VMA_ASSERT(0 && "CRITICAL ERROR: Trying to use linear allocator as ring buffer while it was already used as double stack."); + break; + default: + VMA_ASSERT(0); + } + + suballocations2nd.push_back(newSuballoc); + } + else + { + VMA_ASSERT(0 && "CRITICAL INTERNAL ERROR."); + } + } + } + + m_SumFreeSize -= newSuballoc.size; +} + +void VmaBlockMetadata_Linear::Free(const VmaAllocation allocation) +{ + FreeAtOffset(allocation->GetOffset()); +} + +void VmaBlockMetadata_Linear::FreeAtOffset(VkDeviceSize offset) +{ + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + if(!suballocations1st.empty()) + { + // First allocation: Mark it as next empty at the beginning. + VmaSuballocation& firstSuballoc = suballocations1st[m_1stNullItemsBeginCount]; + if(firstSuballoc.offset == offset) + { + firstSuballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + firstSuballoc.hAllocation = VK_NULL_HANDLE; + m_SumFreeSize += firstSuballoc.size; + ++m_1stNullItemsBeginCount; + CleanupAfterFree(); + return; + } + } + + // Last allocation in 2-part ring buffer or top of upper stack (same logic). + if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER || + m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK) + { + VmaSuballocation& lastSuballoc = suballocations2nd.back(); + if(lastSuballoc.offset == offset) + { + m_SumFreeSize += lastSuballoc.size; + suballocations2nd.pop_back(); + CleanupAfterFree(); + return; + } + } + // Last allocation in 1st vector. + else if(m_2ndVectorMode == SECOND_VECTOR_EMPTY) + { + VmaSuballocation& lastSuballoc = suballocations1st.back(); + if(lastSuballoc.offset == offset) + { + m_SumFreeSize += lastSuballoc.size; + suballocations1st.pop_back(); + CleanupAfterFree(); + return; + } + } + + // Item from the middle of 1st vector. + { + VmaSuballocation refSuballoc; + refSuballoc.offset = offset; + // Rest of members stays uninitialized intentionally for better performance. + SuballocationVectorType::iterator it = VmaVectorFindSorted<VmaSuballocationOffsetLess>( + suballocations1st.begin() + m_1stNullItemsBeginCount, + suballocations1st.end(), + refSuballoc); + if(it != suballocations1st.end()) + { + it->type = VMA_SUBALLOCATION_TYPE_FREE; + it->hAllocation = VK_NULL_HANDLE; + ++m_1stNullItemsMiddleCount; + m_SumFreeSize += it->size; + CleanupAfterFree(); + return; + } + } + + if(m_2ndVectorMode != SECOND_VECTOR_EMPTY) + { + // Item from the middle of 2nd vector. + VmaSuballocation refSuballoc; + refSuballoc.offset = offset; + // Rest of members stays uninitialized intentionally for better performance. + SuballocationVectorType::iterator it = m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER ? + VmaVectorFindSorted<VmaSuballocationOffsetLess>(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc) : + VmaVectorFindSorted<VmaSuballocationOffsetGreater>(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc); + if(it != suballocations2nd.end()) + { + it->type = VMA_SUBALLOCATION_TYPE_FREE; + it->hAllocation = VK_NULL_HANDLE; + ++m_2ndNullItemsCount; + m_SumFreeSize += it->size; + CleanupAfterFree(); + return; + } + } + + VMA_ASSERT(0 && "Allocation to free not found in linear allocator!"); +} + +bool VmaBlockMetadata_Linear::ShouldCompact1st() const +{ + const size_t nullItemCount = m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount; + const size_t suballocCount = AccessSuballocations1st().size(); + return suballocCount > 32 && nullItemCount * 2 >= (suballocCount - nullItemCount) * 3; +} + +void VmaBlockMetadata_Linear::CleanupAfterFree() +{ + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + if(IsEmpty()) + { + suballocations1st.clear(); + suballocations2nd.clear(); + m_1stNullItemsBeginCount = 0; + m_1stNullItemsMiddleCount = 0; + m_2ndNullItemsCount = 0; + m_2ndVectorMode = SECOND_VECTOR_EMPTY; + } + else + { + const size_t suballoc1stCount = suballocations1st.size(); + const size_t nullItem1stCount = m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount; + VMA_ASSERT(nullItem1stCount <= suballoc1stCount); + + // Find more null items at the beginning of 1st vector. + while(m_1stNullItemsBeginCount < suballoc1stCount && + suballocations1st[m_1stNullItemsBeginCount].hAllocation == VK_NULL_HANDLE) + { + ++m_1stNullItemsBeginCount; + --m_1stNullItemsMiddleCount; + } + + // Find more null items at the end of 1st vector. + while(m_1stNullItemsMiddleCount > 0 && + suballocations1st.back().hAllocation == VK_NULL_HANDLE) + { + --m_1stNullItemsMiddleCount; + suballocations1st.pop_back(); + } + + // Find more null items at the end of 2nd vector. + while(m_2ndNullItemsCount > 0 && + suballocations2nd.back().hAllocation == VK_NULL_HANDLE) + { + --m_2ndNullItemsCount; + suballocations2nd.pop_back(); + } + + if(ShouldCompact1st()) + { + const size_t nonNullItemCount = suballoc1stCount - nullItem1stCount; + size_t srcIndex = m_1stNullItemsBeginCount; + for(size_t dstIndex = 0; dstIndex < nonNullItemCount; ++dstIndex) + { + while(suballocations1st[srcIndex].hAllocation == VK_NULL_HANDLE) + { + ++srcIndex; + } + if(dstIndex != srcIndex) + { + suballocations1st[dstIndex] = suballocations1st[srcIndex]; + } + ++srcIndex; + } + suballocations1st.resize(nonNullItemCount); + m_1stNullItemsBeginCount = 0; + m_1stNullItemsMiddleCount = 0; + } + + // 2nd vector became empty. + if(suballocations2nd.empty()) + { + m_2ndVectorMode = SECOND_VECTOR_EMPTY; + } + + // 1st vector became empty. + if(suballocations1st.size() - m_1stNullItemsBeginCount == 0) + { + suballocations1st.clear(); + m_1stNullItemsBeginCount = 0; + + if(!suballocations2nd.empty() && m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) + { + // Swap 1st with 2nd. Now 2nd is empty. + m_2ndVectorMode = SECOND_VECTOR_EMPTY; + m_1stNullItemsMiddleCount = m_2ndNullItemsCount; + while(m_1stNullItemsBeginCount < suballocations2nd.size() && + suballocations2nd[m_1stNullItemsBeginCount].hAllocation == VK_NULL_HANDLE) + { + ++m_1stNullItemsBeginCount; + --m_1stNullItemsMiddleCount; + } + m_2ndNullItemsCount = 0; + m_1stVectorIndex ^= 1; + } + } + } + + VMA_HEAVY_ASSERT(Validate()); +} + + +//////////////////////////////////////////////////////////////////////////////// +// class VmaBlockMetadata_Buddy + +VmaBlockMetadata_Buddy::VmaBlockMetadata_Buddy(VmaAllocator hAllocator) : + VmaBlockMetadata(hAllocator), + m_Root(VMA_NULL), + m_AllocationCount(0), + m_FreeCount(1), + m_SumFreeSize(0) +{ + memset(m_FreeList, 0, sizeof(m_FreeList)); +} + +VmaBlockMetadata_Buddy::~VmaBlockMetadata_Buddy() +{ + DeleteNode(m_Root); +} + +void VmaBlockMetadata_Buddy::Init(VkDeviceSize size) +{ + VmaBlockMetadata::Init(size); + + m_UsableSize = VmaPrevPow2(size); + m_SumFreeSize = m_UsableSize; + + // Calculate m_LevelCount. + m_LevelCount = 1; + while(m_LevelCount < MAX_LEVELS && + LevelToNodeSize(m_LevelCount) >= MIN_NODE_SIZE) + { + ++m_LevelCount; + } + + Node* rootNode = vma_new(GetAllocationCallbacks(), Node)(); + rootNode->offset = 0; + rootNode->type = Node::TYPE_FREE; + rootNode->parent = VMA_NULL; + rootNode->buddy = VMA_NULL; + + m_Root = rootNode; + AddToFreeListFront(0, rootNode); +} + +bool VmaBlockMetadata_Buddy::Validate() const +{ + // Validate tree. + ValidationContext ctx; + if(!ValidateNode(ctx, VMA_NULL, m_Root, 0, LevelToNodeSize(0))) + { + VMA_VALIDATE(false && "ValidateNode failed."); + } + VMA_VALIDATE(m_AllocationCount == ctx.calculatedAllocationCount); + VMA_VALIDATE(m_SumFreeSize == ctx.calculatedSumFreeSize); + + // Validate free node lists. + for(uint32_t level = 0; level < m_LevelCount; ++level) + { + VMA_VALIDATE(m_FreeList[level].front == VMA_NULL || + m_FreeList[level].front->free.prev == VMA_NULL); + + for(Node* node = m_FreeList[level].front; + node != VMA_NULL; + node = node->free.next) + { + VMA_VALIDATE(node->type == Node::TYPE_FREE); + + if(node->free.next == VMA_NULL) + { + VMA_VALIDATE(m_FreeList[level].back == node); + } + else + { + VMA_VALIDATE(node->free.next->free.prev == node); + } + } + } + + // Validate that free lists ar higher levels are empty. + for(uint32_t level = m_LevelCount; level < MAX_LEVELS; ++level) + { + VMA_VALIDATE(m_FreeList[level].front == VMA_NULL && m_FreeList[level].back == VMA_NULL); + } + + return true; +} + +VkDeviceSize VmaBlockMetadata_Buddy::GetUnusedRangeSizeMax() const +{ + for(uint32_t level = 0; level < m_LevelCount; ++level) + { + if(m_FreeList[level].front != VMA_NULL) + { + return LevelToNodeSize(level); + } + } + return 0; +} + +void VmaBlockMetadata_Buddy::CalcAllocationStatInfo(VmaStatInfo& outInfo) const +{ + const VkDeviceSize unusableSize = GetUnusableSize(); + + outInfo.blockCount = 1; + + outInfo.allocationCount = outInfo.unusedRangeCount = 0; + outInfo.usedBytes = outInfo.unusedBytes = 0; + + outInfo.allocationSizeMax = outInfo.unusedRangeSizeMax = 0; + outInfo.allocationSizeMin = outInfo.unusedRangeSizeMin = UINT64_MAX; + outInfo.allocationSizeAvg = outInfo.unusedRangeSizeAvg = 0; // Unused. + + CalcAllocationStatInfoNode(outInfo, m_Root, LevelToNodeSize(0)); + + if(unusableSize > 0) + { + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusableSize; + outInfo.unusedRangeSizeMax = VMA_MAX(outInfo.unusedRangeSizeMax, unusableSize); + outInfo.unusedRangeSizeMin = VMA_MIN(outInfo.unusedRangeSizeMin, unusableSize); + } +} + +void VmaBlockMetadata_Buddy::AddPoolStats(VmaPoolStats& inoutStats) const +{ + const VkDeviceSize unusableSize = GetUnusableSize(); + + inoutStats.size += GetSize(); + inoutStats.unusedSize += m_SumFreeSize + unusableSize; + inoutStats.allocationCount += m_AllocationCount; + inoutStats.unusedRangeCount += m_FreeCount; + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, GetUnusedRangeSizeMax()); + + if(unusableSize > 0) + { + ++inoutStats.unusedRangeCount; + // Not updating inoutStats.unusedRangeSizeMax with unusableSize because this space is not available for allocations. + } +} + +#if VMA_STATS_STRING_ENABLED + +void VmaBlockMetadata_Buddy::PrintDetailedMap(class VmaJsonWriter& json) const +{ + // TODO optimize + VmaStatInfo stat; + CalcAllocationStatInfo(stat); + + PrintDetailedMap_Begin( + json, + stat.unusedBytes, + stat.allocationCount, + stat.unusedRangeCount); + + PrintDetailedMapNode(json, m_Root, LevelToNodeSize(0)); + + const VkDeviceSize unusableSize = GetUnusableSize(); + if(unusableSize > 0) + { + PrintDetailedMap_UnusedRange(json, + m_UsableSize, // offset + unusableSize); // size + } + + PrintDetailedMap_End(json); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Buddy::CreateAllocationRequest( + uint32_t /*currentFrameIndex*/, + uint32_t /*frameInUseCount*/, + VkDeviceSize bufferImageGranularity, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + bool /*canMakeOtherLost*/, + uint32_t /*strategy*/, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(!upperAddress && "VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT can be used only with linear algorithm."); + (void) upperAddress; + + // Simple way to respect bufferImageGranularity. May be optimized some day. + // Whenever it might be an OPTIMAL image... + if(allocType == VMA_SUBALLOCATION_TYPE_UNKNOWN || + allocType == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + allocType == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL) + { + allocAlignment = VMA_MAX(allocAlignment, bufferImageGranularity); + allocSize = VMA_MAX(allocSize, bufferImageGranularity); + } + + if(allocSize > m_UsableSize) + { + return false; + } + + const uint32_t targetLevel = AllocSizeToLevel(allocSize); + for(uint32_t level = targetLevel + 1; level--; ) + { + for(Node* freeNode = m_FreeList[level].front; + freeNode != VMA_NULL; + freeNode = freeNode->free.next) + { + if(freeNode->offset % allocAlignment == 0) + { + pAllocationRequest->offset = freeNode->offset; + pAllocationRequest->sumFreeSize = LevelToNodeSize(level); + pAllocationRequest->sumItemSize = 0; + pAllocationRequest->itemsToMakeLostCount = 0; + pAllocationRequest->customData = (void*)(uintptr_t)level; + return true; + } + } + } + + return false; +} + +bool VmaBlockMetadata_Buddy::MakeRequestedAllocationsLost( + uint32_t /*currentFrameIndex*/, + uint32_t /*frameInUseCount*/, + VmaAllocationRequest* pAllocationRequest) +{ + /* + Lost allocations are not supported in buddy allocator at the moment. + Support might be added in the future. + */ + return pAllocationRequest->itemsToMakeLostCount == 0; +} + +uint32_t VmaBlockMetadata_Buddy::MakeAllocationsLost(uint32_t /*currentFrameIndex*/, uint32_t /*frameInUseCount*/) +{ + /* + Lost allocations are not supported in buddy allocator at the moment. + Support might be added in the future. + */ + return 0; +} + +void VmaBlockMetadata_Buddy::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType /*type*/, + VkDeviceSize allocSize, + bool /*upperAddress*/, + VmaAllocation hAllocation) +{ + const uint32_t targetLevel = AllocSizeToLevel(allocSize); + uint32_t currLevel = (uint32_t)(uintptr_t)request.customData; + + Node* currNode = m_FreeList[currLevel].front; + VMA_ASSERT(currNode != VMA_NULL && currNode->type == Node::TYPE_FREE); + while(currNode->offset != request.offset) + { + currNode = currNode->free.next; + VMA_ASSERT(currNode != VMA_NULL && currNode->type == Node::TYPE_FREE); + } + + // Go down, splitting free nodes. + while(currLevel < targetLevel) + { + // currNode is already first free node at currLevel. + // Remove it from list of free nodes at this currLevel. + RemoveFromFreeList(currLevel, currNode); + + const uint32_t childrenLevel = currLevel + 1; + + // Create two free sub-nodes. + Node* leftChild = vma_new(GetAllocationCallbacks(), Node)(); + Node* rightChild = vma_new(GetAllocationCallbacks(), Node)(); + + leftChild->offset = currNode->offset; + leftChild->type = Node::TYPE_FREE; + leftChild->parent = currNode; + leftChild->buddy = rightChild; + + rightChild->offset = currNode->offset + LevelToNodeSize(childrenLevel); + rightChild->type = Node::TYPE_FREE; + rightChild->parent = currNode; + rightChild->buddy = leftChild; + + // Convert current currNode to split type. + currNode->type = Node::TYPE_SPLIT; + currNode->split.leftChild = leftChild; + + // Add child nodes to free list. Order is important! + AddToFreeListFront(childrenLevel, rightChild); + AddToFreeListFront(childrenLevel, leftChild); + + ++m_FreeCount; + //m_SumFreeSize -= LevelToNodeSize(currLevel) % 2; // Useful only when level node sizes can be non power of 2. + ++currLevel; + currNode = m_FreeList[currLevel].front; + + /* + We can be sure that currNode, as left child of node previously split, + also fullfills the alignment requirement. + */ + } + + // Remove from free list. + VMA_ASSERT(currLevel == targetLevel && + currNode != VMA_NULL && + currNode->type == Node::TYPE_FREE); + RemoveFromFreeList(currLevel, currNode); + + // Convert to allocation node. + currNode->type = Node::TYPE_ALLOCATION; + currNode->allocation.alloc = hAllocation; + + ++m_AllocationCount; + --m_FreeCount; + m_SumFreeSize -= allocSize; +} + +void VmaBlockMetadata_Buddy::DeleteNode(Node* node) +{ + if(node->type == Node::TYPE_SPLIT) + { + DeleteNode(node->split.leftChild->buddy); + DeleteNode(node->split.leftChild); + } + + vma_delete(GetAllocationCallbacks(), node); +} + +bool VmaBlockMetadata_Buddy::ValidateNode(ValidationContext& ctx, const Node* parent, const Node* curr, uint32_t level, VkDeviceSize levelNodeSize) const +{ + VMA_VALIDATE(level < m_LevelCount); + VMA_VALIDATE(curr->parent == parent); + VMA_VALIDATE((curr->buddy == VMA_NULL) == (parent == VMA_NULL)); + VMA_VALIDATE(curr->buddy == VMA_NULL || curr->buddy->buddy == curr); + switch(curr->type) + { + case Node::TYPE_FREE: + // curr->free.prev, next are validated separately. + ctx.calculatedSumFreeSize += levelNodeSize; + ++ctx.calculatedFreeCount; + break; + case Node::TYPE_ALLOCATION: + ++ctx.calculatedAllocationCount; + ctx.calculatedSumFreeSize += levelNodeSize - curr->allocation.alloc->GetSize(); + VMA_VALIDATE(curr->allocation.alloc != VK_NULL_HANDLE); + break; + case Node::TYPE_SPLIT: + { + const uint32_t childrenLevel = level + 1; + const VkDeviceSize childrenLevelNodeSize = levelNodeSize / 2; + const Node* const leftChild = curr->split.leftChild; + VMA_VALIDATE(leftChild != VMA_NULL); + VMA_VALIDATE(leftChild->offset == curr->offset); + if(!ValidateNode(ctx, curr, leftChild, childrenLevel, childrenLevelNodeSize)) + { + VMA_VALIDATE(false && "ValidateNode for left child failed."); + } + const Node* const rightChild = leftChild->buddy; + VMA_VALIDATE(rightChild->offset == curr->offset + childrenLevelNodeSize); + if(!ValidateNode(ctx, curr, rightChild, childrenLevel, childrenLevelNodeSize)) + { + VMA_VALIDATE(false && "ValidateNode for right child failed."); + } + } + break; + default: + return false; + } + + return true; +} + +uint32_t VmaBlockMetadata_Buddy::AllocSizeToLevel(VkDeviceSize allocSize) const +{ + // I know this could be optimized somehow e.g. by using std::log2p1 from C++20. + uint32_t level = 0; + VkDeviceSize currLevelNodeSize = m_UsableSize; + VkDeviceSize nextLevelNodeSize = currLevelNodeSize >> 1; + while(allocSize <= nextLevelNodeSize && level + 1 < m_LevelCount) + { + ++level; + currLevelNodeSize = nextLevelNodeSize; + nextLevelNodeSize = currLevelNodeSize >> 1; + } + return level; +} + +void VmaBlockMetadata_Buddy::FreeAtOffset(VmaAllocation alloc, VkDeviceSize offset) +{ + // Find node and level. + Node* node = m_Root; + VkDeviceSize nodeOffset = 0; + uint32_t level = 0; + VkDeviceSize levelNodeSize = LevelToNodeSize(0); + while(node->type == Node::TYPE_SPLIT) + { + const VkDeviceSize nextLevelSize = levelNodeSize >> 1; + if(offset < nodeOffset + nextLevelSize) + { + node = node->split.leftChild; + } + else + { + node = node->split.leftChild->buddy; + nodeOffset += nextLevelSize; + } + ++level; + levelNodeSize = nextLevelSize; + } + + VMA_ASSERT(node != VMA_NULL && node->type == Node::TYPE_ALLOCATION); + VMA_ASSERT(alloc == VK_NULL_HANDLE || node->allocation.alloc == alloc); + + ++m_FreeCount; + --m_AllocationCount; + m_SumFreeSize += alloc->GetSize(); + + node->type = Node::TYPE_FREE; + + // Join free nodes if possible. + while(level > 0 && node->buddy->type == Node::TYPE_FREE) + { + RemoveFromFreeList(level, node->buddy); + Node* const parent = node->parent; + + vma_delete(GetAllocationCallbacks(), node->buddy); + vma_delete(GetAllocationCallbacks(), node); + parent->type = Node::TYPE_FREE; + + node = parent; + --level; + //m_SumFreeSize += LevelToNodeSize(level) % 2; // Useful only when level node sizes can be non power of 2. + --m_FreeCount; + } + + AddToFreeListFront(level, node); +} + +void VmaBlockMetadata_Buddy::CalcAllocationStatInfoNode(VmaStatInfo& outInfo, const Node* node, VkDeviceSize levelNodeSize) const +{ + switch(node->type) + { + case Node::TYPE_FREE: + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += levelNodeSize; + outInfo.unusedRangeSizeMax = VMA_MAX(outInfo.unusedRangeSizeMax, levelNodeSize); + outInfo.unusedRangeSizeMin = VMA_MAX(outInfo.unusedRangeSizeMin, levelNodeSize); + break; + case Node::TYPE_ALLOCATION: + { + const VkDeviceSize allocSize = node->allocation.alloc->GetSize(); + ++outInfo.allocationCount; + outInfo.usedBytes += allocSize; + outInfo.allocationSizeMax = VMA_MAX(outInfo.allocationSizeMax, allocSize); + outInfo.allocationSizeMin = VMA_MAX(outInfo.allocationSizeMin, allocSize); + + const VkDeviceSize unusedRangeSize = levelNodeSize - allocSize; + if(unusedRangeSize > 0) + { + ++outInfo.unusedRangeCount; + outInfo.unusedBytes += unusedRangeSize; + outInfo.unusedRangeSizeMax = VMA_MAX(outInfo.unusedRangeSizeMax, unusedRangeSize); + outInfo.unusedRangeSizeMin = VMA_MAX(outInfo.unusedRangeSizeMin, unusedRangeSize); + } + } + break; + case Node::TYPE_SPLIT: + { + const VkDeviceSize childrenNodeSize = levelNodeSize / 2; + const Node* const leftChild = node->split.leftChild; + CalcAllocationStatInfoNode(outInfo, leftChild, childrenNodeSize); + const Node* const rightChild = leftChild->buddy; + CalcAllocationStatInfoNode(outInfo, rightChild, childrenNodeSize); + } + break; + default: + VMA_ASSERT(0); + } +} + +void VmaBlockMetadata_Buddy::AddToFreeListFront(uint32_t level, Node* node) +{ + VMA_ASSERT(node->type == Node::TYPE_FREE); + + // List is empty. + Node* const frontNode = m_FreeList[level].front; + if(frontNode == VMA_NULL) + { + VMA_ASSERT(m_FreeList[level].back == VMA_NULL); + node->free.prev = node->free.next = VMA_NULL; + m_FreeList[level].front = m_FreeList[level].back = node; + } + else + { + VMA_ASSERT(frontNode->free.prev == VMA_NULL); + node->free.prev = VMA_NULL; + node->free.next = frontNode; + frontNode->free.prev = node; + m_FreeList[level].front = node; + } +} + +void VmaBlockMetadata_Buddy::RemoveFromFreeList(uint32_t level, Node* node) +{ + VMA_ASSERT(m_FreeList[level].front != VMA_NULL); + + // It is at the front. + if(node->free.prev == VMA_NULL) + { + VMA_ASSERT(m_FreeList[level].front == node); + m_FreeList[level].front = node->free.next; + } + else + { + Node* const prevFreeNode = node->free.prev; + VMA_ASSERT(prevFreeNode->free.next == node); + prevFreeNode->free.next = node->free.next; + } + + // It is at the back. + if(node->free.next == VMA_NULL) + { + VMA_ASSERT(m_FreeList[level].back == node); + m_FreeList[level].back = node->free.prev; + } + else + { + Node* const nextFreeNode = node->free.next; + VMA_ASSERT(nextFreeNode->free.prev == node); + nextFreeNode->free.prev = node->free.prev; + } +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata_Buddy::PrintDetailedMapNode(class VmaJsonWriter& json, const Node* node, VkDeviceSize levelNodeSize) const +{ + switch(node->type) + { + case Node::TYPE_FREE: + PrintDetailedMap_UnusedRange(json, node->offset, levelNodeSize); + break; + case Node::TYPE_ALLOCATION: + { + PrintDetailedMap_Allocation(json, node->offset, node->allocation.alloc); + const VkDeviceSize allocSize = node->allocation.alloc->GetSize(); + if(allocSize < levelNodeSize) + { + PrintDetailedMap_UnusedRange(json, node->offset + allocSize, levelNodeSize - allocSize); + } + } + break; + case Node::TYPE_SPLIT: + { + const VkDeviceSize childrenNodeSize = levelNodeSize / 2; + const Node* const leftChild = node->split.leftChild; + PrintDetailedMapNode(json, leftChild, childrenNodeSize); + const Node* const rightChild = leftChild->buddy; + PrintDetailedMapNode(json, rightChild, childrenNodeSize); + } + break; + default: + VMA_ASSERT(0); + } +} +#endif // #if VMA_STATS_STRING_ENABLED + + +//////////////////////////////////////////////////////////////////////////////// +// class VmaDeviceMemoryBlock + +VmaDeviceMemoryBlock::VmaDeviceMemoryBlock(VmaAllocator /*hAllocator*/) : + m_pMetadata(VMA_NULL), + m_MemoryTypeIndex(UINT32_MAX), + m_Id(0), + m_hMemory(VK_NULL_HANDLE), + m_MapCount(0), + m_pMappedData(VMA_NULL) +{ +} + +void VmaDeviceMemoryBlock::Init( + VmaAllocator hAllocator, + uint32_t newMemoryTypeIndex, + VkDeviceMemory newMemory, + VkDeviceSize newSize, + uint32_t id, + uint32_t algorithm) +{ + VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); + + m_MemoryTypeIndex = newMemoryTypeIndex; + m_Id = id; + m_hMemory = newMemory; + + switch(algorithm) + { + case VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_Linear)(hAllocator); + break; + case VMA_POOL_CREATE_BUDDY_ALGORITHM_BIT: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_Buddy)(hAllocator); + break; + default: + VMA_ASSERT(0); + // Fall-through. + case 0: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_Generic)(hAllocator); + } + m_pMetadata->Init(newSize); +} + +void VmaDeviceMemoryBlock::Destroy(VmaAllocator allocator) +{ + // This is the most important assert in the entire library. + // Hitting it means you have some memory leak - unreleased VmaAllocation objects. + VMA_ASSERT(m_pMetadata->IsEmpty() && "Some allocations were not freed before destruction of this memory block!"); + + VMA_ASSERT(m_hMemory != VK_NULL_HANDLE); + allocator->FreeVulkanMemory(m_MemoryTypeIndex, m_pMetadata->GetSize(), m_hMemory); + m_hMemory = VK_NULL_HANDLE; + + vma_delete(allocator, m_pMetadata); + m_pMetadata = VMA_NULL; +} + +bool VmaDeviceMemoryBlock::Validate() const +{ + VMA_VALIDATE((m_hMemory != VK_NULL_HANDLE) && + (m_pMetadata->GetSize() != 0)); + + return m_pMetadata->Validate(); +} + +VkResult VmaDeviceMemoryBlock::CheckCorruption(VmaAllocator hAllocator) +{ + void* pData = nullptr; + VkResult res = Map(hAllocator, 1, &pData); + if(res != VK_SUCCESS) + { + return res; + } + + res = m_pMetadata->CheckCorruption(pData); + + Unmap(hAllocator, 1); + + return res; +} + +VkResult VmaDeviceMemoryBlock::Map(VmaAllocator hAllocator, uint32_t count, void** ppData) +{ + if(count == 0) + { + return VK_SUCCESS; + } + + VmaMutexLock lock(m_Mutex, hAllocator->m_UseMutex); + if(m_MapCount != 0) + { + m_MapCount += count; + VMA_ASSERT(m_pMappedData != VMA_NULL); + if(ppData != VMA_NULL) + { + *ppData = m_pMappedData; + } + return VK_SUCCESS; + } + else + { + VkResult result = (*hAllocator->GetVulkanFunctions().vkMapMemory)( + hAllocator->m_hDevice, + m_hMemory, + 0, // offset + VK_WHOLE_SIZE, + 0, // flags + &m_pMappedData); + if(result == VK_SUCCESS) + { + if(ppData != VMA_NULL) + { + *ppData = m_pMappedData; + } + m_MapCount = count; + } + return result; + } +} + +void VmaDeviceMemoryBlock::Unmap(VmaAllocator hAllocator, uint32_t count) +{ + if(count == 0) + { + return; + } + + VmaMutexLock lock(m_Mutex, hAllocator->m_UseMutex); + if(m_MapCount >= count) + { + m_MapCount -= count; + if(m_MapCount == 0) + { + m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)(hAllocator->m_hDevice, m_hMemory); + } + } + else + { + VMA_ASSERT(0 && "VkDeviceMemory block is being unmapped while it was not previously mapped."); + } +} + +VkResult VmaDeviceMemoryBlock::WriteMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize) +{ + VMA_ASSERT(VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_MARGIN % 4 == 0 && VMA_DEBUG_DETECT_CORRUPTION); + VMA_ASSERT(allocOffset >= VMA_DEBUG_MARGIN); + + void* pData; + VkResult res = Map(hAllocator, 1, &pData); + if(res != VK_SUCCESS) + { + return res; + } + + VmaWriteMagicValue(pData, allocOffset - VMA_DEBUG_MARGIN); + VmaWriteMagicValue(pData, allocOffset + allocSize); + + Unmap(hAllocator, 1); + + return VK_SUCCESS; +} + +VkResult VmaDeviceMemoryBlock::ValidateMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize) +{ + VMA_ASSERT(VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_MARGIN % 4 == 0 && VMA_DEBUG_DETECT_CORRUPTION); + VMA_ASSERT(allocOffset >= VMA_DEBUG_MARGIN); + + void* pData; + VkResult res = Map(hAllocator, 1, &pData); + if(res != VK_SUCCESS) + { + return res; + } + + if(!VmaValidateMagicValue(pData, allocOffset - VMA_DEBUG_MARGIN)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED BEFORE FREED ALLOCATION!"); + } + else if(!VmaValidateMagicValue(pData, allocOffset + allocSize)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER FREED ALLOCATION!"); + } + + Unmap(hAllocator, 1); + + return VK_SUCCESS; +} + +VkResult VmaDeviceMemoryBlock::BindBufferMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkBuffer hBuffer) +{ + VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && + hAllocation->GetBlock() == this); + // This lock is important so that we don't call vkBind... and/or vkMap... simultaneously on the same VkDeviceMemory from multiple threads. + VmaMutexLock lock(m_Mutex, hAllocator->m_UseMutex); + return hAllocator->GetVulkanFunctions().vkBindBufferMemory( + hAllocator->m_hDevice, + hBuffer, + m_hMemory, + hAllocation->GetOffset()); +} + +VkResult VmaDeviceMemoryBlock::BindImageMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkImage hImage) +{ + VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && + hAllocation->GetBlock() == this); + // This lock is important so that we don't call vkBind... and/or vkMap... simultaneously on the same VkDeviceMemory from multiple threads. + VmaMutexLock lock(m_Mutex, hAllocator->m_UseMutex); + return hAllocator->GetVulkanFunctions().vkBindImageMemory( + hAllocator->m_hDevice, + hImage, + m_hMemory, + hAllocation->GetOffset()); +} + +static void InitStatInfo(VmaStatInfo& outInfo) +{ + memset(&outInfo, 0, sizeof(outInfo)); + outInfo.allocationSizeMin = UINT64_MAX; + outInfo.unusedRangeSizeMin = UINT64_MAX; +} + +// Adds statistics srcInfo into inoutInfo, like: inoutInfo += srcInfo. +static void VmaAddStatInfo(VmaStatInfo& inoutInfo, const VmaStatInfo& srcInfo) +{ + inoutInfo.blockCount += srcInfo.blockCount; + inoutInfo.allocationCount += srcInfo.allocationCount; + inoutInfo.unusedRangeCount += srcInfo.unusedRangeCount; + inoutInfo.usedBytes += srcInfo.usedBytes; + inoutInfo.unusedBytes += srcInfo.unusedBytes; + inoutInfo.allocationSizeMin = VMA_MIN(inoutInfo.allocationSizeMin, srcInfo.allocationSizeMin); + inoutInfo.allocationSizeMax = VMA_MAX(inoutInfo.allocationSizeMax, srcInfo.allocationSizeMax); + inoutInfo.unusedRangeSizeMin = VMA_MIN(inoutInfo.unusedRangeSizeMin, srcInfo.unusedRangeSizeMin); + inoutInfo.unusedRangeSizeMax = VMA_MAX(inoutInfo.unusedRangeSizeMax, srcInfo.unusedRangeSizeMax); +} + +static void VmaPostprocessCalcStatInfo(VmaStatInfo& inoutInfo) +{ + inoutInfo.allocationSizeAvg = (inoutInfo.allocationCount > 0) ? + VmaRoundDiv<VkDeviceSize>(inoutInfo.usedBytes, inoutInfo.allocationCount) : 0; + inoutInfo.unusedRangeSizeAvg = (inoutInfo.unusedRangeCount > 0) ? + VmaRoundDiv<VkDeviceSize>(inoutInfo.unusedBytes, inoutInfo.unusedRangeCount) : 0; +} + +VmaPool_T::VmaPool_T( + VmaAllocator hAllocator, + const VmaPoolCreateInfo& createInfo, + VkDeviceSize preferredBlockSize) : + m_BlockVector( + hAllocator, + createInfo.memoryTypeIndex, + createInfo.blockSize != 0 ? createInfo.blockSize : preferredBlockSize, + createInfo.minBlockCount, + createInfo.maxBlockCount, + (createInfo.flags & VMA_POOL_CREATE_IGNORE_BUFFER_IMAGE_GRANULARITY_BIT) != 0 ? 1 : hAllocator->GetBufferImageGranularity(), + createInfo.frameInUseCount, + true, // isCustomPool + createInfo.blockSize != 0, // explicitBlockSize + createInfo.flags & VMA_POOL_CREATE_ALGORITHM_MASK), // algorithm + m_Id(0) +{ +} + +VmaPool_T::~VmaPool_T() +{ +} + +#if VMA_STATS_STRING_ENABLED + +#endif // #if VMA_STATS_STRING_ENABLED + +VmaBlockVector::VmaBlockVector( + VmaAllocator hAllocator, + uint32_t memoryTypeIndex, + VkDeviceSize preferredBlockSize, + size_t minBlockCount, + size_t maxBlockCount, + VkDeviceSize bufferImageGranularity, + uint32_t frameInUseCount, + bool isCustomPool, + bool explicitBlockSize, + uint32_t algorithm) : + m_hAllocator(hAllocator), + m_MemoryTypeIndex(memoryTypeIndex), + m_PreferredBlockSize(preferredBlockSize), + m_MinBlockCount(minBlockCount), + m_MaxBlockCount(maxBlockCount), + m_BufferImageGranularity(bufferImageGranularity), + m_FrameInUseCount(frameInUseCount), + m_IsCustomPool(isCustomPool), + m_ExplicitBlockSize(explicitBlockSize), + m_Algorithm(algorithm), + m_HasEmptyBlock(false), + m_Blocks(VmaStlAllocator<VmaDeviceMemoryBlock*>(hAllocator->GetAllocationCallbacks())), + m_NextBlockId(0) +{ +} + +VmaBlockVector::~VmaBlockVector() +{ + for(size_t i = m_Blocks.size(); i--; ) + { + m_Blocks[i]->Destroy(m_hAllocator); + vma_delete(m_hAllocator, m_Blocks[i]); + } +} + +VkResult VmaBlockVector::CreateMinBlocks() +{ + for(size_t i = 0; i < m_MinBlockCount; ++i) + { + VkResult res = CreateBlock(m_PreferredBlockSize, VMA_NULL); + if(res != VK_SUCCESS) + { + return res; + } + } + return VK_SUCCESS; +} + +void VmaBlockVector::GetPoolStats(VmaPoolStats* pStats) +{ + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + const size_t blockCount = m_Blocks.size(); + + pStats->size = 0; + pStats->unusedSize = 0; + pStats->allocationCount = 0; + pStats->unusedRangeCount = 0; + pStats->unusedRangeSizeMax = 0; + pStats->blockCount = blockCount; + + for(uint32_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + { + const VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pBlock); + VMA_HEAVY_ASSERT(pBlock->Validate()); + pBlock->m_pMetadata->AddPoolStats(*pStats); + } +} + +bool VmaBlockVector::IsCorruptionDetectionEnabled() const +{ + const uint32_t requiredMemFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; + return (VMA_DEBUG_DETECT_CORRUPTION != 0) && + (VMA_DEBUG_MARGIN > 0) && + (m_hAllocator->m_MemProps.memoryTypes[m_MemoryTypeIndex].propertyFlags & requiredMemFlags) == requiredMemFlags; +} + +static const uint32_t VMA_ALLOCATION_TRY_COUNT = 32; + +VkResult VmaBlockVector::Allocate( + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + size_t allocIndex; + VkResult res = VK_SUCCESS; + + { + VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); + for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + res = AllocatePage( + hCurrentPool, + currentFrameIndex, + size, + alignment, + createInfo, + suballocType, + pAllocations + allocIndex); + if(res != VK_SUCCESS) + { + break; + } + } + } + + if(res != VK_SUCCESS) + { + // Free all already created allocations. + while(allocIndex--) + { + Free(pAllocations[allocIndex]); + } + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + } + + return res; +} + +VkResult VmaBlockVector::AllocatePage( + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation) +{ + const bool isUpperAddress = (createInfo.flags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0; + bool canMakeOtherLost = (createInfo.flags & VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT) != 0; + const bool mapped = (createInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0; + const bool isUserDataString = (createInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0; + const bool canCreateNewBlock = + ((createInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) == 0) && + (m_Blocks.size() < m_MaxBlockCount); + uint32_t strategy = createInfo.flags & VMA_ALLOCATION_CREATE_STRATEGY_MASK; + + // If linearAlgorithm is used, canMakeOtherLost is available only when used as ring buffer. + // Which in turn is available only when maxBlockCount = 1. + if(m_Algorithm == VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT && m_MaxBlockCount > 1) + { + canMakeOtherLost = false; + } + + // Upper address can only be used with linear allocator and within single memory block. + if(isUpperAddress && + (m_Algorithm != VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT || m_MaxBlockCount > 1)) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + // Validate strategy. + switch(strategy) + { + case 0: + strategy = VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT; + break; + case VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT: + case VMA_ALLOCATION_CREATE_STRATEGY_WORST_FIT_BIT: + case VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT: + break; + default: + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + // Early reject: requested allocation size is larger that maximum block size for this block vector. + if(size + 2 * VMA_DEBUG_MARGIN > m_PreferredBlockSize) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + + /* + Under certain condition, this whole section can be skipped for optimization, so + we move on directly to trying to allocate with canMakeOtherLost. That's the case + e.g. for custom pools with linear algorithm. + */ + if(!canMakeOtherLost || canCreateNewBlock) + { + // 1. Search existing allocations. Try to allocate without making other allocations lost. + VmaAllocationCreateFlags allocFlagsCopy = createInfo.flags; + allocFlagsCopy &= ~VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT; + + if(m_Algorithm == VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) + { + // Use only last block. + if(!m_Blocks.empty()) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks.back(); + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock( + pCurrBlock, + hCurrentPool, + currentFrameIndex, + size, + alignment, + allocFlagsCopy, + createInfo.pUserData, + suballocType, + strategy, + pAllocation); + if(res == VK_SUCCESS) + { + VMA_DEBUG_LOG(" Returned from last block #%u", (uint32_t)(m_Blocks.size() - 1)); + return VK_SUCCESS; + } + } + } + else + { + if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT) + { + // Forward order in m_Blocks - prefer blocks with smallest amount of free space. + for(size_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex ) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock( + pCurrBlock, + hCurrentPool, + currentFrameIndex, + size, + alignment, + allocFlagsCopy, + createInfo.pUserData, + suballocType, + strategy, + pAllocation); + if(res == VK_SUCCESS) + { + VMA_DEBUG_LOG(" Returned from existing block #%u", (uint32_t)blockIndex); + return VK_SUCCESS; + } + } + } + else // WORST_FIT, FIRST_FIT + { + // Backward order in m_Blocks - prefer blocks with largest amount of free space. + for(size_t blockIndex = m_Blocks.size(); blockIndex--; ) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock( + pCurrBlock, + hCurrentPool, + currentFrameIndex, + size, + alignment, + allocFlagsCopy, + createInfo.pUserData, + suballocType, + strategy, + pAllocation); + if(res == VK_SUCCESS) + { + VMA_DEBUG_LOG(" Returned from existing block #%u", (uint32_t)blockIndex); + return VK_SUCCESS; + } + } + } + } + + // 2. Try to create new block. + if(canCreateNewBlock) + { + // Calculate optimal size for new block. + VkDeviceSize newBlockSize = m_PreferredBlockSize; + uint32_t newBlockSizeShift = 0; + const uint32_t NEW_BLOCK_SIZE_SHIFT_MAX = 3; + + if(!m_ExplicitBlockSize) + { + // Allocate 1/8, 1/4, 1/2 as first blocks. + const VkDeviceSize maxExistingBlockSize = CalcMaxBlockSize(); + for(uint32_t i = 0; i < NEW_BLOCK_SIZE_SHIFT_MAX; ++i) + { + const VkDeviceSize smallerNewBlockSize = newBlockSize / 2; + if(smallerNewBlockSize > maxExistingBlockSize && smallerNewBlockSize >= size * 2) + { + newBlockSize = smallerNewBlockSize; + ++newBlockSizeShift; + } + else + { + break; + } + } + } + + size_t newBlockIndex = 0; + VkResult res = CreateBlock(newBlockSize, &newBlockIndex); + // Allocation of this size failed? Try 1/2, 1/4, 1/8 of m_PreferredBlockSize. + if(!m_ExplicitBlockSize) + { + while(res < 0 && newBlockSizeShift < NEW_BLOCK_SIZE_SHIFT_MAX) + { + const VkDeviceSize smallerNewBlockSize = newBlockSize / 2; + if(smallerNewBlockSize >= size) + { + newBlockSize = smallerNewBlockSize; + ++newBlockSizeShift; + res = CreateBlock(newBlockSize, &newBlockIndex); + } + else + { + break; + } + } + } + + if(res == VK_SUCCESS) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[newBlockIndex]; + VMA_ASSERT(pBlock->m_pMetadata->GetSize() >= size); + + res = AllocateFromBlock( + pBlock, + hCurrentPool, + currentFrameIndex, + size, + alignment, + allocFlagsCopy, + createInfo.pUserData, + suballocType, + strategy, + pAllocation); + if(res == VK_SUCCESS) + { + VMA_DEBUG_LOG(" Created new block Size=%llu", newBlockSize); + return VK_SUCCESS; + } + else + { + // Allocation from new block failed, possibly due to VMA_DEBUG_MARGIN or alignment. + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + } + } + } + + // 3. Try to allocate from existing blocks with making other allocations lost. + if(canMakeOtherLost) + { + uint32_t tryIndex = 0; + for(; tryIndex < VMA_ALLOCATION_TRY_COUNT; ++tryIndex) + { + VmaDeviceMemoryBlock* pBestRequestBlock = VMA_NULL; + VmaAllocationRequest bestRequest = {}; + VkDeviceSize bestRequestCost = VK_WHOLE_SIZE; + + // 1. Search existing allocations. + if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT) + { + // Forward order in m_Blocks - prefer blocks with smallest amount of free space. + for(size_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex ) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pCurrBlock); + VmaAllocationRequest currRequest = {}; + if(pCurrBlock->m_pMetadata->CreateAllocationRequest( + currentFrameIndex, + m_FrameInUseCount, + m_BufferImageGranularity, + size, + alignment, + (createInfo.flags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0, + suballocType, + canMakeOtherLost, + strategy, + &currRequest)) + { + const VkDeviceSize currRequestCost = currRequest.CalcCost(); + if(pBestRequestBlock == VMA_NULL || + currRequestCost < bestRequestCost) + { + pBestRequestBlock = pCurrBlock; + bestRequest = currRequest; + bestRequestCost = currRequestCost; + + if(bestRequestCost == 0) + { + break; + } + } + } + } + } + else // WORST_FIT, FIRST_FIT + { + // Backward order in m_Blocks - prefer blocks with largest amount of free space. + for(size_t blockIndex = m_Blocks.size(); blockIndex--; ) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pCurrBlock); + VmaAllocationRequest currRequest = {}; + if(pCurrBlock->m_pMetadata->CreateAllocationRequest( + currentFrameIndex, + m_FrameInUseCount, + m_BufferImageGranularity, + size, + alignment, + (createInfo.flags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0, + suballocType, + canMakeOtherLost, + strategy, + &currRequest)) + { + const VkDeviceSize currRequestCost = currRequest.CalcCost(); + if(pBestRequestBlock == VMA_NULL || + currRequestCost < bestRequestCost || + strategy == VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT) + { + pBestRequestBlock = pCurrBlock; + bestRequest = currRequest; + bestRequestCost = currRequestCost; + + if(bestRequestCost == 0 || + strategy == VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT) + { + break; + } + } + } + } + } + + if(pBestRequestBlock != VMA_NULL) + { + if(mapped) + { + VkResult res = pBestRequestBlock->Map(m_hAllocator, 1, VMA_NULL); + if(res != VK_SUCCESS) + { + return res; + } + } + + if(pBestRequestBlock->m_pMetadata->MakeRequestedAllocationsLost( + currentFrameIndex, + m_FrameInUseCount, + &bestRequest)) + { + // We no longer have an empty Allocation. + if(pBestRequestBlock->m_pMetadata->IsEmpty()) + { + m_HasEmptyBlock = false; + } + // Allocate from this pBlock. + *pAllocation = vma_new(m_hAllocator, VmaAllocation_T)(currentFrameIndex, isUserDataString); + pBestRequestBlock->m_pMetadata->Alloc(bestRequest, suballocType, size, isUpperAddress, *pAllocation); + (*pAllocation)->InitBlockAllocation( + hCurrentPool, + pBestRequestBlock, + bestRequest.offset, + alignment, + size, + suballocType, + mapped, + (createInfo.flags & VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT) != 0); + VMA_HEAVY_ASSERT(pBestRequestBlock->Validate()); + VMA_DEBUG_LOG(" Returned from existing allocation #%u", (uint32_t)blockIndex); + (*pAllocation)->SetUserData(m_hAllocator, createInfo.pUserData); + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) + { + m_hAllocator->FillAllocation(*pAllocation, VMA_ALLOCATION_FILL_PATTERN_CREATED); + } + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBestRequestBlock->WriteMagicValueAroundAllocation(m_hAllocator, bestRequest.offset, size); + (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); + } + return VK_SUCCESS; + } + // else: Some allocations must have been touched while we are here. Next try. + } + else + { + // Could not find place in any of the blocks - break outer loop. + break; + } + } + /* Maximum number of tries exceeded - a very unlike event when many other + threads are simultaneously touching allocations making it impossible to make + lost at the same time as we try to allocate. */ + if(tryIndex == VMA_ALLOCATION_TRY_COUNT) + { + return VK_ERROR_TOO_MANY_OBJECTS; + } + } + + return VK_ERROR_OUT_OF_DEVICE_MEMORY; +} + +void VmaBlockVector::Free( + VmaAllocation hAllocation) +{ + VmaDeviceMemoryBlock* pBlockToDelete = VMA_NULL; + + // Scope for lock. + { + VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); + + VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); + + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBlock->ValidateMagicValueAroundAllocation(m_hAllocator, hAllocation->GetOffset(), hAllocation->GetSize()); + (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to validate magic value."); + } + + if(hAllocation->IsPersistentMap()) + { + pBlock->Unmap(m_hAllocator, 1); + } + + pBlock->m_pMetadata->Free(hAllocation); + VMA_HEAVY_ASSERT(pBlock->Validate()); + + VMA_DEBUG_LOG(" Freed from MemoryTypeIndex=%u", memTypeIndex); + + // pBlock became empty after this deallocation. + if(pBlock->m_pMetadata->IsEmpty()) + { + // Already has empty Allocation. We don't want to have two, so delete this one. + if(m_HasEmptyBlock && m_Blocks.size() > m_MinBlockCount) + { + pBlockToDelete = pBlock; + Remove(pBlock); + } + // We now have first empty block. + else + { + m_HasEmptyBlock = true; + } + } + // pBlock didn't become empty, but we have another empty block - find and free that one. + // (This is optional, heuristics.) + else if(m_HasEmptyBlock) + { + VmaDeviceMemoryBlock* pLastBlock = m_Blocks.back(); + if(pLastBlock->m_pMetadata->IsEmpty() && m_Blocks.size() > m_MinBlockCount) + { + pBlockToDelete = pLastBlock; + m_Blocks.pop_back(); + m_HasEmptyBlock = false; + } + } + + IncrementallySortBlocks(); + } + + // Destruction of a free Allocation. Deferred until this point, outside of mutex + // lock, for performance reason. + if(pBlockToDelete != VMA_NULL) + { + VMA_DEBUG_LOG(" Deleted empty allocation"); + pBlockToDelete->Destroy(m_hAllocator); + vma_delete(m_hAllocator, pBlockToDelete); + } +} + +VkDeviceSize VmaBlockVector::CalcMaxBlockSize() const +{ + VkDeviceSize result = 0; + for(size_t i = m_Blocks.size(); i--; ) + { + result = VMA_MAX(result, m_Blocks[i]->m_pMetadata->GetSize()); + if(result >= m_PreferredBlockSize) + { + break; + } + } + return result; +} + +void VmaBlockVector::Remove(VmaDeviceMemoryBlock* pBlock) +{ + for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + if(m_Blocks[blockIndex] == pBlock) + { + VmaVectorRemove(m_Blocks, blockIndex); + return; + } + } + VMA_ASSERT(0); +} + +void VmaBlockVector::IncrementallySortBlocks() +{ + if(m_Algorithm != VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) + { + // Bubble sort only until first swap. + for(size_t i = 1; i < m_Blocks.size(); ++i) + { + if(m_Blocks[i - 1]->m_pMetadata->GetSumFreeSize() > m_Blocks[i]->m_pMetadata->GetSumFreeSize()) + { + VMA_SWAP(m_Blocks[i - 1], m_Blocks[i]); + return; + } + } + } +} + +VkResult VmaBlockVector::AllocateFromBlock( + VmaDeviceMemoryBlock* pBlock, + VmaPool hCurrentPool, + uint32_t currentFrameIndex, + VkDeviceSize size, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + uint32_t strategy, + VmaAllocation* pAllocation) +{ + VMA_ASSERT((allocFlags & VMA_ALLOCATION_CREATE_CAN_MAKE_OTHER_LOST_BIT) == 0); + const bool isUpperAddress = (allocFlags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0; + const bool mapped = (allocFlags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0; + const bool isUserDataString = (allocFlags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0; + + VmaAllocationRequest currRequest = {}; + if(pBlock->m_pMetadata->CreateAllocationRequest( + currentFrameIndex, + m_FrameInUseCount, + m_BufferImageGranularity, + size, + alignment, + isUpperAddress, + suballocType, + false, // canMakeOtherLost + strategy, + &currRequest)) + { + // Allocate from pCurrBlock. + VMA_ASSERT(currRequest.itemsToMakeLostCount == 0); + + if(mapped) + { + VkResult res = pBlock->Map(m_hAllocator, 1, VMA_NULL); + if(res != VK_SUCCESS) + { + return res; + } + } + + // We no longer have an empty Allocation. + if(pBlock->m_pMetadata->IsEmpty()) + { + m_HasEmptyBlock = false; + } + + *pAllocation = vma_new(m_hAllocator, VmaAllocation_T)(currentFrameIndex, isUserDataString); + pBlock->m_pMetadata->Alloc(currRequest, suballocType, size, isUpperAddress, *pAllocation); + (*pAllocation)->InitBlockAllocation( + hCurrentPool, + pBlock, + currRequest.offset, + alignment, + size, + suballocType, + mapped, + (allocFlags & VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT) != 0); + VMA_HEAVY_ASSERT(pBlock->Validate()); + (*pAllocation)->SetUserData(m_hAllocator, pUserData); + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) + { + m_hAllocator->FillAllocation(*pAllocation, VMA_ALLOCATION_FILL_PATTERN_CREATED); + } + if(IsCorruptionDetectionEnabled()) + { + VkResult res = pBlock->WriteMagicValueAroundAllocation(m_hAllocator, currRequest.offset, size); + (void) res; + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); + } + return VK_SUCCESS; + } + return VK_ERROR_OUT_OF_DEVICE_MEMORY; +} + +VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIndex) +{ + VkMemoryAllocateInfo allocInfo = {}; + allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + allocInfo.memoryTypeIndex = m_MemoryTypeIndex; + allocInfo.allocationSize = blockSize; + VkDeviceMemory mem = VK_NULL_HANDLE; + VkResult res = m_hAllocator->AllocateVulkanMemory(&allocInfo, &mem); + if(res < 0) + { + return res; + } + + // New VkDeviceMemory successfully created. + + // Create new Allocation for it. + VmaDeviceMemoryBlock* const pBlock = vma_new(m_hAllocator, VmaDeviceMemoryBlock)(m_hAllocator); + pBlock->Init( + m_hAllocator, + m_MemoryTypeIndex, + mem, + allocInfo.allocationSize, + m_NextBlockId++, + m_Algorithm); + + m_Blocks.push_back(pBlock); + if(pNewBlockIndex != VMA_NULL) + { + *pNewBlockIndex = m_Blocks.size() - 1; + } + + return VK_SUCCESS; +} + +void VmaBlockVector::ApplyDefragmentationMovesCpu( + class VmaBlockVectorDefragmentationContext* pDefragCtx, + const VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves) +{ + const size_t blockCount = m_Blocks.size(); + const bool isNonCoherent = m_hAllocator->IsMemoryTypeNonCoherent(m_MemoryTypeIndex); + + enum BLOCK_FLAG + { + BLOCK_FLAG_USED = 0x00000001, + BLOCK_FLAG_MAPPED_FOR_DEFRAGMENTATION = 0x00000002, + }; + + struct BlockInfo + { + uint32_t flags; + void* pMappedData; + }; + VmaVector< BlockInfo, VmaStlAllocator<BlockInfo> > + blockInfo(blockCount, VmaStlAllocator<BlockInfo>(m_hAllocator->GetAllocationCallbacks())); + memset(blockInfo.data(), 0, blockCount * sizeof(BlockInfo)); + + // Go over all moves. Mark blocks that are used with BLOCK_FLAG_USED. + const size_t moveCount = moves.size(); + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { + const VmaDefragmentationMove& move = moves[moveIndex]; + blockInfo[move.srcBlockIndex].flags |= BLOCK_FLAG_USED; + blockInfo[move.dstBlockIndex].flags |= BLOCK_FLAG_USED; + } + + VMA_ASSERT(pDefragCtx->res == VK_SUCCESS); + + // Go over all blocks. Get mapped pointer or map if necessary. + for(size_t blockIndex = 0; pDefragCtx->res == VK_SUCCESS && blockIndex < blockCount; ++blockIndex) + { + BlockInfo& currBlockInfo = blockInfo[blockIndex]; + VmaDeviceMemoryBlock* pBlock = m_Blocks[blockIndex]; + if((currBlockInfo.flags & BLOCK_FLAG_USED) != 0) + { + currBlockInfo.pMappedData = pBlock->GetMappedData(); + // It is not originally mapped - map it. + if(currBlockInfo.pMappedData == VMA_NULL) + { + pDefragCtx->res = pBlock->Map(m_hAllocator, 1, &currBlockInfo.pMappedData); + if(pDefragCtx->res == VK_SUCCESS) + { + currBlockInfo.flags |= BLOCK_FLAG_MAPPED_FOR_DEFRAGMENTATION; + } + } + } + } + + // Go over all moves. Do actual data transfer. + if(pDefragCtx->res == VK_SUCCESS) + { + const VkDeviceSize nonCoherentAtomSize = m_hAllocator->m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; + VkMappedMemoryRange memRange = {}; + memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE; + + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { + const VmaDefragmentationMove& move = moves[moveIndex]; + + const BlockInfo& srcBlockInfo = blockInfo[move.srcBlockIndex]; + const BlockInfo& dstBlockInfo = blockInfo[move.dstBlockIndex]; + + VMA_ASSERT(srcBlockInfo.pMappedData && dstBlockInfo.pMappedData); + + // Invalidate source. + if(isNonCoherent) + { + VmaDeviceMemoryBlock* const pSrcBlock = m_Blocks[move.srcBlockIndex]; + memRange.memory = pSrcBlock->GetDeviceMemory(); + memRange.offset = VmaAlignDown(move.srcOffset, nonCoherentAtomSize); + memRange.size = VMA_MIN( + VmaAlignUp(move.size + (move.srcOffset - memRange.offset), nonCoherentAtomSize), + pSrcBlock->m_pMetadata->GetSize() - memRange.offset); + (*m_hAllocator->GetVulkanFunctions().vkInvalidateMappedMemoryRanges)(m_hAllocator->m_hDevice, 1, &memRange); + } + + // THE PLACE WHERE ACTUAL DATA COPY HAPPENS. + memmove( + reinterpret_cast<char*>(dstBlockInfo.pMappedData) + move.dstOffset, + reinterpret_cast<char*>(srcBlockInfo.pMappedData) + move.srcOffset, + static_cast<size_t>(move.size)); + + if(IsCorruptionDetectionEnabled()) + { + VmaWriteMagicValue(dstBlockInfo.pMappedData, move.dstOffset - VMA_DEBUG_MARGIN); + VmaWriteMagicValue(dstBlockInfo.pMappedData, move.dstOffset + move.size); + } + + // Flush destination. + if(isNonCoherent) + { + VmaDeviceMemoryBlock* const pDstBlock = m_Blocks[move.dstBlockIndex]; + memRange.memory = pDstBlock->GetDeviceMemory(); + memRange.offset = VmaAlignDown(move.dstOffset, nonCoherentAtomSize); + memRange.size = VMA_MIN( + VmaAlignUp(move.size + (move.dstOffset - memRange.offset), nonCoherentAtomSize), + pDstBlock->m_pMetadata->GetSize() - memRange.offset); + (*m_hAllocator->GetVulkanFunctions().vkFlushMappedMemoryRanges)(m_hAllocator->m_hDevice, 1, &memRange); + } + } + } + + // Go over all blocks in reverse order. Unmap those that were mapped just for defragmentation. + // Regardless of pCtx->res == VK_SUCCESS. + for(size_t blockIndex = blockCount; blockIndex--; ) + { + const BlockInfo& currBlockInfo = blockInfo[blockIndex]; + if((currBlockInfo.flags & BLOCK_FLAG_MAPPED_FOR_DEFRAGMENTATION) != 0) + { + VmaDeviceMemoryBlock* pBlock = m_Blocks[blockIndex]; + pBlock->Unmap(m_hAllocator, 1); + } + } +} + +void VmaBlockVector::ApplyDefragmentationMovesGpu( + class VmaBlockVectorDefragmentationContext* pDefragCtx, + const VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkCommandBuffer commandBuffer) +{ + const size_t blockCount = m_Blocks.size(); + + pDefragCtx->blockContexts.resize(blockCount); + for (size_t i = 0; i < blockCount; ++i) + pDefragCtx->blockContexts[i] = VmaBlockDefragmentationContext(); + + // Go over all moves. Mark blocks that are used with BLOCK_FLAG_USED. + const size_t moveCount = moves.size(); + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { + const VmaDefragmentationMove& move = moves[moveIndex]; + pDefragCtx->blockContexts[move.srcBlockIndex].flags |= VmaBlockDefragmentationContext::BLOCK_FLAG_USED; + pDefragCtx->blockContexts[move.dstBlockIndex].flags |= VmaBlockDefragmentationContext::BLOCK_FLAG_USED; + } + + VMA_ASSERT(pDefragCtx->res == VK_SUCCESS); + + // Go over all blocks. Create and bind buffer for whole block if necessary. + { + VkBufferCreateInfo bufCreateInfo = {}; + bufCreateInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; + bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | + VK_BUFFER_USAGE_TRANSFER_DST_BIT; + + for(size_t blockIndex = 0; pDefragCtx->res == VK_SUCCESS && blockIndex < blockCount; ++blockIndex) + { + VmaBlockDefragmentationContext& currBlockCtx = pDefragCtx->blockContexts[blockIndex]; + VmaDeviceMemoryBlock* pBlock = m_Blocks[blockIndex]; + if((currBlockCtx.flags & VmaBlockDefragmentationContext::BLOCK_FLAG_USED) != 0) + { + bufCreateInfo.size = pBlock->m_pMetadata->GetSize(); + pDefragCtx->res = (*m_hAllocator->GetVulkanFunctions().vkCreateBuffer)( + m_hAllocator->m_hDevice, &bufCreateInfo, m_hAllocator->GetAllocationCallbacks(), &currBlockCtx.hBuffer); + if(pDefragCtx->res == VK_SUCCESS) + { + pDefragCtx->res = (*m_hAllocator->GetVulkanFunctions().vkBindBufferMemory)( + m_hAllocator->m_hDevice, currBlockCtx.hBuffer, pBlock->GetDeviceMemory(), 0); + } + } + } + } + + // Go over all moves. Post data transfer commands to command buffer. + if(pDefragCtx->res == VK_SUCCESS) + { + /*const VkDeviceSize nonCoherentAtomSize = m_hAllocator->m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; + VkMappedMemoryRange memRange = {}; + memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;*/ + + for(size_t moveIndex = 0; moveIndex < moveCount; ++moveIndex) + { + const VmaDefragmentationMove& move = moves[moveIndex]; + + const VmaBlockDefragmentationContext& srcBlockCtx = pDefragCtx->blockContexts[move.srcBlockIndex]; + const VmaBlockDefragmentationContext& dstBlockCtx = pDefragCtx->blockContexts[move.dstBlockIndex]; + + VMA_ASSERT(srcBlockCtx.hBuffer && dstBlockCtx.hBuffer); + + VkBufferCopy region = { + move.srcOffset, + move.dstOffset, + move.size }; + (*m_hAllocator->GetVulkanFunctions().vkCmdCopyBuffer)( + commandBuffer, srcBlockCtx.hBuffer, dstBlockCtx.hBuffer, 1, ®ion); + } + } + + // Save buffers to defrag context for later destruction. + if(pDefragCtx->res == VK_SUCCESS && moveCount > 0) + { + pDefragCtx->res = VK_NOT_READY; + } +} + +void VmaBlockVector::FreeEmptyBlocks(VmaDefragmentationStats* pDefragmentationStats) +{ + m_HasEmptyBlock = false; + for(size_t blockIndex = m_Blocks.size(); blockIndex--; ) + { + VmaDeviceMemoryBlock* pBlock = m_Blocks[blockIndex]; + if(pBlock->m_pMetadata->IsEmpty()) + { + if(m_Blocks.size() > m_MinBlockCount) + { + if(pDefragmentationStats != VMA_NULL) + { + ++pDefragmentationStats->deviceMemoryBlocksFreed; + pDefragmentationStats->bytesFreed += pBlock->m_pMetadata->GetSize(); + } + + VmaVectorRemove(m_Blocks, blockIndex); + pBlock->Destroy(m_hAllocator); + vma_delete(m_hAllocator, pBlock); + } + else + { + m_HasEmptyBlock = true; + } + } + } +} + +#if VMA_STATS_STRING_ENABLED + +void VmaBlockVector::PrintDetailedMap(class VmaJsonWriter& json) +{ + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + json.BeginObject(); + + if(m_IsCustomPool) + { + json.WriteString("MemoryTypeIndex"); + json.WriteNumber(m_MemoryTypeIndex); + + json.WriteString("BlockSize"); + json.WriteNumber(m_PreferredBlockSize); + + json.WriteString("BlockCount"); + json.BeginObject(true); + if(m_MinBlockCount > 0) + { + json.WriteString("Min"); + json.WriteNumber((uint64_t)m_MinBlockCount); + } + if(m_MaxBlockCount < SIZE_MAX) + { + json.WriteString("Max"); + json.WriteNumber((uint64_t)m_MaxBlockCount); + } + json.WriteString("Cur"); + json.WriteNumber((uint64_t)m_Blocks.size()); + json.EndObject(); + + if(m_FrameInUseCount > 0) + { + json.WriteString("FrameInUseCount"); + json.WriteNumber(m_FrameInUseCount); + } + + if(m_Algorithm != 0) + { + json.WriteString("Algorithm"); + json.WriteString(VmaAlgorithmToStr(m_Algorithm)); + } + } + else + { + json.WriteString("PreferredBlockSize"); + json.WriteNumber(m_PreferredBlockSize); + } + + json.WriteString("Blocks"); + json.BeginObject(); + for(size_t i = 0; i < m_Blocks.size(); ++i) + { + json.BeginString(); + json.ContinueString(m_Blocks[i]->GetId()); + json.EndString(); + + m_Blocks[i]->m_pMetadata->PrintDetailedMap(json); + } + json.EndObject(); + + json.EndObject(); +} + +#endif // #if VMA_STATS_STRING_ENABLED + +void VmaBlockVector::Defragment( + class VmaBlockVectorDefragmentationContext* pCtx, + VmaDefragmentationStats* pStats, + VkDeviceSize& maxCpuBytesToMove, uint32_t& maxCpuAllocationsToMove, + VkDeviceSize& maxGpuBytesToMove, uint32_t& maxGpuAllocationsToMove, + VkCommandBuffer commandBuffer) +{ + pCtx->res = VK_SUCCESS; + + const VkMemoryPropertyFlags memPropFlags = + m_hAllocator->m_MemProps.memoryTypes[m_MemoryTypeIndex].propertyFlags; + const bool isHostVisible = (memPropFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0; + const bool isHostCoherent = (memPropFlags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0; + + const bool canDefragmentOnCpu = maxCpuBytesToMove > 0 && maxCpuAllocationsToMove > 0 && + isHostVisible; + const bool canDefragmentOnGpu = maxGpuBytesToMove > 0 && maxGpuAllocationsToMove > 0 && + (VMA_DEBUG_DETECT_CORRUPTION == 0 || !(isHostVisible && isHostCoherent)); + + // There are options to defragment this memory type. + if(canDefragmentOnCpu || canDefragmentOnGpu) + { + bool defragmentOnGpu; + // There is only one option to defragment this memory type. + if(canDefragmentOnGpu != canDefragmentOnCpu) + { + defragmentOnGpu = canDefragmentOnGpu; + } + // Both options are available: Heuristics to choose the best one. + else + { + defragmentOnGpu = (memPropFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0 || + m_hAllocator->IsIntegratedGpu(); + } + + bool overlappingMoveSupported = !defragmentOnGpu; + + if(m_hAllocator->m_UseMutex) + { + m_Mutex.LockWrite(); + pCtx->mutexLocked = true; + } + + pCtx->Begin(overlappingMoveSupported); + + // Defragment. + + const VkDeviceSize maxBytesToMove = defragmentOnGpu ? maxGpuBytesToMove : maxCpuBytesToMove; + const uint32_t maxAllocationsToMove = defragmentOnGpu ? maxGpuAllocationsToMove : maxCpuAllocationsToMove; + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> > moves = + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >(VmaStlAllocator<VmaDefragmentationMove>(m_hAllocator->GetAllocationCallbacks())); + pCtx->res = pCtx->GetAlgorithm()->Defragment(moves, maxBytesToMove, maxAllocationsToMove); + + // Accumulate statistics. + if(pStats != VMA_NULL) + { + const VkDeviceSize bytesMoved = pCtx->GetAlgorithm()->GetBytesMoved(); + const uint32_t allocationsMoved = pCtx->GetAlgorithm()->GetAllocationsMoved(); + pStats->bytesMoved += bytesMoved; + pStats->allocationsMoved += allocationsMoved; + VMA_ASSERT(bytesMoved <= maxBytesToMove); + VMA_ASSERT(allocationsMoved <= maxAllocationsToMove); + if(defragmentOnGpu) + { + maxGpuBytesToMove -= bytesMoved; + maxGpuAllocationsToMove -= allocationsMoved; + } + else + { + maxCpuBytesToMove -= bytesMoved; + maxCpuAllocationsToMove -= allocationsMoved; + } + } + + if(pCtx->res >= VK_SUCCESS) + { + if(defragmentOnGpu) + { + ApplyDefragmentationMovesGpu(pCtx, moves, commandBuffer); + } + else + { + ApplyDefragmentationMovesCpu(pCtx, moves); + } + } + } +} + +void VmaBlockVector::DefragmentationEnd( + class VmaBlockVectorDefragmentationContext* pCtx, + VmaDefragmentationStats* pStats) +{ + // Destroy buffers. + for(size_t blockIndex = pCtx->blockContexts.size(); blockIndex--; ) + { + VmaBlockDefragmentationContext& blockCtx = pCtx->blockContexts[blockIndex]; + if(blockCtx.hBuffer) + { + (*m_hAllocator->GetVulkanFunctions().vkDestroyBuffer)( + m_hAllocator->m_hDevice, blockCtx.hBuffer, m_hAllocator->GetAllocationCallbacks()); + } + } + + if(pCtx->res >= VK_SUCCESS) + { + FreeEmptyBlocks(pStats); + } + + if(pCtx->mutexLocked) + { + VMA_ASSERT(m_hAllocator->m_UseMutex); + m_Mutex.UnlockWrite(); + } +} + +size_t VmaBlockVector::CalcAllocationCount() const +{ + size_t result = 0; + for(size_t i = 0; i < m_Blocks.size(); ++i) + { + result += m_Blocks[i]->m_pMetadata->GetAllocationCount(); + } + return result; +} + +bool VmaBlockVector::IsBufferImageGranularityConflictPossible() const +{ + if(m_BufferImageGranularity == 1) + { + return false; + } + VmaSuballocationType lastSuballocType = VMA_SUBALLOCATION_TYPE_FREE; + for(size_t i = 0, count = m_Blocks.size(); i < count; ++i) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[i]; + VMA_ASSERT(m_Algorithm == 0); + VmaBlockMetadata_Generic* const pMetadata = (VmaBlockMetadata_Generic*)pBlock->m_pMetadata; + if(pMetadata->IsBufferImageGranularityConflictPossible(m_BufferImageGranularity, lastSuballocType)) + { + return true; + } + } + return false; +} + +void VmaBlockVector::MakePoolAllocationsLost( + uint32_t currentFrameIndex, + size_t* pLostAllocationCount) +{ + VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); + size_t lostAllocationCount = 0; + for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pBlock); + lostAllocationCount += pBlock->m_pMetadata->MakeAllocationsLost(currentFrameIndex, m_FrameInUseCount); + } + if(pLostAllocationCount != VMA_NULL) + { + *pLostAllocationCount = lostAllocationCount; + } +} + +VkResult VmaBlockVector::CheckCorruption() +{ + if(!IsCorruptionDetectionEnabled()) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pBlock); + VkResult res = pBlock->CheckCorruption(m_hAllocator); + if(res != VK_SUCCESS) + { + return res; + } + } + return VK_SUCCESS; +} + +void VmaBlockVector::AddStats(VmaStats* pStats) +{ + const uint32_t memTypeIndex = m_MemoryTypeIndex; + const uint32_t memHeapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(memTypeIndex); + + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + const VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pBlock); + VMA_HEAVY_ASSERT(pBlock->Validate()); + VmaStatInfo allocationStatInfo; + pBlock->m_pMetadata->CalcAllocationStatInfo(allocationStatInfo); + VmaAddStatInfo(pStats->total, allocationStatInfo); + VmaAddStatInfo(pStats->memoryType[memTypeIndex], allocationStatInfo); + VmaAddStatInfo(pStats->memoryHeap[memHeapIndex], allocationStatInfo); + } +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaDefragmentationAlgorithm_Generic members definition + +VmaDefragmentationAlgorithm_Generic::VmaDefragmentationAlgorithm_Generic( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex, + bool /*overlappingMoveSupported*/) : + VmaDefragmentationAlgorithm(hAllocator, pBlockVector, currentFrameIndex), + m_AllocationCount(0), + m_AllAllocations(false), + m_BytesMoved(0), + m_AllocationsMoved(0), + m_Blocks(VmaStlAllocator<BlockInfo*>(hAllocator->GetAllocationCallbacks())) +{ + // Create block info for each block. + const size_t blockCount = m_pBlockVector->m_Blocks.size(); + for(size_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + { + BlockInfo* pBlockInfo = vma_new(m_hAllocator, BlockInfo)(m_hAllocator->GetAllocationCallbacks()); + pBlockInfo->m_OriginalBlockIndex = blockIndex; + pBlockInfo->m_pBlock = m_pBlockVector->m_Blocks[blockIndex]; + m_Blocks.push_back(pBlockInfo); + } + + // Sort them by m_pBlock pointer value. + VMA_SORT(m_Blocks.begin(), m_Blocks.end(), BlockPointerLess()); +} + +VmaDefragmentationAlgorithm_Generic::~VmaDefragmentationAlgorithm_Generic() +{ + for(size_t i = m_Blocks.size(); i--; ) + { + vma_delete(m_hAllocator, m_Blocks[i]); + } +} + +void VmaDefragmentationAlgorithm_Generic::AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged) +{ + // Now as we are inside VmaBlockVector::m_Mutex, we can make final check if this allocation was not lost. + if(hAlloc->GetLastUseFrameIndex() != VMA_FRAME_INDEX_LOST) + { + VmaDeviceMemoryBlock* pBlock = hAlloc->GetBlock(); + BlockInfoVector::iterator it = VmaBinaryFindFirstNotLess(m_Blocks.begin(), m_Blocks.end(), pBlock, BlockPointerLess()); + if(it != m_Blocks.end() && (*it)->m_pBlock == pBlock) + { + AllocationInfo allocInfo = AllocationInfo(hAlloc, pChanged); + (*it)->m_Allocations.push_back(allocInfo); + } + else + { + VMA_ASSERT(0); + } + + ++m_AllocationCount; + } +} + +VkResult VmaDefragmentationAlgorithm_Generic::DefragmentRound( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove) +{ + if(m_Blocks.empty()) + { + return VK_SUCCESS; + } + + // This is a choice based on research. + // Option 1: + uint32_t strategy = VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT; + // Option 2: + //uint32_t strategy = VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT; + // Option 3: + //uint32_t strategy = VMA_ALLOCATION_CREATE_STRATEGY_MIN_FRAGMENTATION_BIT; + + size_t srcBlockMinIndex = 0; + // When FAST_ALGORITHM, move allocations from only last out of blocks that contain non-movable allocations. + /* + if(m_AlgorithmFlags & VMA_DEFRAGMENTATION_FAST_ALGORITHM_BIT) + { + const size_t blocksWithNonMovableCount = CalcBlocksWithNonMovableCount(); + if(blocksWithNonMovableCount > 0) + { + srcBlockMinIndex = blocksWithNonMovableCount - 1; + } + } + */ + + size_t srcBlockIndex = m_Blocks.size() - 1; + size_t srcAllocIndex = SIZE_MAX; + for(;;) + { + // 1. Find next allocation to move. + // 1.1. Start from last to first m_Blocks - they are sorted from most "destination" to most "source". + // 1.2. Then start from last to first m_Allocations. + while(srcAllocIndex >= m_Blocks[srcBlockIndex]->m_Allocations.size()) + { + if(m_Blocks[srcBlockIndex]->m_Allocations.empty()) + { + // Finished: no more allocations to process. + if(srcBlockIndex == srcBlockMinIndex) + { + return VK_SUCCESS; + } + else + { + --srcBlockIndex; + srcAllocIndex = SIZE_MAX; + } + } + else + { + srcAllocIndex = m_Blocks[srcBlockIndex]->m_Allocations.size() - 1; + } + } + + BlockInfo* pSrcBlockInfo = m_Blocks[srcBlockIndex]; + AllocationInfo& allocInfo = pSrcBlockInfo->m_Allocations[srcAllocIndex]; + + const VkDeviceSize size = allocInfo.m_hAllocation->GetSize(); + const VkDeviceSize srcOffset = allocInfo.m_hAllocation->GetOffset(); + const VkDeviceSize alignment = allocInfo.m_hAllocation->GetAlignment(); + const VmaSuballocationType suballocType = allocInfo.m_hAllocation->GetSuballocationType(); + + // 2. Try to find new place for this allocation in preceding or current block. + for(size_t dstBlockIndex = 0; dstBlockIndex <= srcBlockIndex; ++dstBlockIndex) + { + BlockInfo* pDstBlockInfo = m_Blocks[dstBlockIndex]; + VmaAllocationRequest dstAllocRequest; + if(pDstBlockInfo->m_pBlock->m_pMetadata->CreateAllocationRequest( + m_CurrentFrameIndex, + m_pBlockVector->GetFrameInUseCount(), + m_pBlockVector->GetBufferImageGranularity(), + size, + alignment, + false, // upperAddress + suballocType, + false, // canMakeOtherLost + strategy, + &dstAllocRequest) && + MoveMakesSense( + dstBlockIndex, dstAllocRequest.offset, srcBlockIndex, srcOffset)) + { + VMA_ASSERT(dstAllocRequest.itemsToMakeLostCount == 0); + + // Reached limit on number of allocations or bytes to move. + if((m_AllocationsMoved + 1 > maxAllocationsToMove) || + (m_BytesMoved + size > maxBytesToMove)) + { + return VK_SUCCESS; + } + + VmaDefragmentationMove move; + move.srcBlockIndex = pSrcBlockInfo->m_OriginalBlockIndex; + move.dstBlockIndex = pDstBlockInfo->m_OriginalBlockIndex; + move.srcOffset = srcOffset; + move.dstOffset = dstAllocRequest.offset; + move.size = size; + moves.push_back(move); + + pDstBlockInfo->m_pBlock->m_pMetadata->Alloc( + dstAllocRequest, + suballocType, + size, + false, // upperAddress + allocInfo.m_hAllocation); + pSrcBlockInfo->m_pBlock->m_pMetadata->FreeAtOffset(srcOffset); + + allocInfo.m_hAllocation->ChangeBlockAllocation(m_hAllocator, pDstBlockInfo->m_pBlock, dstAllocRequest.offset); + + if(allocInfo.m_pChanged != VMA_NULL) + { + *allocInfo.m_pChanged = VK_TRUE; + } + + ++m_AllocationsMoved; + m_BytesMoved += size; + + VmaVectorRemove(pSrcBlockInfo->m_Allocations, srcAllocIndex); + + break; + } + } + + // If not processed, this allocInfo remains in pBlockInfo->m_Allocations for next round. + + if(srcAllocIndex > 0) + { + --srcAllocIndex; + } + else + { + if(srcBlockIndex > 0) + { + --srcBlockIndex; + srcAllocIndex = SIZE_MAX; + } + else + { + return VK_SUCCESS; + } + } + } +} + +size_t VmaDefragmentationAlgorithm_Generic::CalcBlocksWithNonMovableCount() const +{ + size_t result = 0; + for(size_t i = 0; i < m_Blocks.size(); ++i) + { + if(m_Blocks[i]->m_HasNonMovableAllocations) + { + ++result; + } + } + return result; +} + +VkResult VmaDefragmentationAlgorithm_Generic::Defragment( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove) +{ + if(!m_AllAllocations && m_AllocationCount == 0) + { + return VK_SUCCESS; + } + + const size_t blockCount = m_Blocks.size(); + for(size_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + { + BlockInfo* pBlockInfo = m_Blocks[blockIndex]; + + if(m_AllAllocations) + { + VmaBlockMetadata_Generic* pMetadata = (VmaBlockMetadata_Generic*)pBlockInfo->m_pBlock->m_pMetadata; + for(VmaSuballocationList::const_iterator it = pMetadata->m_Suballocations.begin(); + it != pMetadata->m_Suballocations.end(); + ++it) + { + if(it->type != VMA_SUBALLOCATION_TYPE_FREE) + { + AllocationInfo allocInfo = AllocationInfo(it->hAllocation, VMA_NULL); + pBlockInfo->m_Allocations.push_back(allocInfo); + } + } + } + + pBlockInfo->CalcHasNonMovableAllocations(); + + // This is a choice based on research. + // Option 1: + pBlockInfo->SortAllocationsByOffsetDescending(); + // Option 2: + //pBlockInfo->SortAllocationsBySizeDescending(); + } + + // Sort m_Blocks this time by the main criterium, from most "destination" to most "source" blocks. + VMA_SORT(m_Blocks.begin(), m_Blocks.end(), BlockInfoCompareMoveDestination()); + + // This is a choice based on research. + const uint32_t roundCount = 2; + + // Execute defragmentation rounds (the main part). + VkResult result = VK_SUCCESS; + for(uint32_t round = 0; (round < roundCount) && (result == VK_SUCCESS); ++round) + { + result = DefragmentRound(moves, maxBytesToMove, maxAllocationsToMove); + } + + return result; +} + +bool VmaDefragmentationAlgorithm_Generic::MoveMakesSense( + size_t dstBlockIndex, VkDeviceSize dstOffset, + size_t srcBlockIndex, VkDeviceSize srcOffset) +{ + if(dstBlockIndex < srcBlockIndex) + { + return true; + } + if(dstBlockIndex > srcBlockIndex) + { + return false; + } + if(dstOffset < srcOffset) + { + return true; + } + return false; +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaDefragmentationAlgorithm_Fast + +VmaDefragmentationAlgorithm_Fast::VmaDefragmentationAlgorithm_Fast( + VmaAllocator hAllocator, + VmaBlockVector* pBlockVector, + uint32_t currentFrameIndex, + bool overlappingMoveSupported) : + VmaDefragmentationAlgorithm(hAllocator, pBlockVector, currentFrameIndex), + m_OverlappingMoveSupported(overlappingMoveSupported), + m_AllocationCount(0), + m_AllAllocations(false), + m_BytesMoved(0), + m_AllocationsMoved(0), + m_BlockInfos(VmaStlAllocator<BlockInfo>(hAllocator->GetAllocationCallbacks())) +{ + VMA_ASSERT(VMA_DEBUG_MARGIN == 0); + +} + +VmaDefragmentationAlgorithm_Fast::~VmaDefragmentationAlgorithm_Fast() +{ +} + +VkResult VmaDefragmentationAlgorithm_Fast::Defragment( + VmaVector< VmaDefragmentationMove, VmaStlAllocator<VmaDefragmentationMove> >& moves, + VkDeviceSize maxBytesToMove, + uint32_t maxAllocationsToMove) +{ + VMA_ASSERT(m_AllAllocations || m_pBlockVector->CalcAllocationCount() == m_AllocationCount); + + const size_t blockCount = m_pBlockVector->GetBlockCount(); + if(blockCount == 0 || maxBytesToMove == 0 || maxAllocationsToMove == 0) + { + return VK_SUCCESS; + } + + PreprocessMetadata(); + + // Sort blocks in order from most destination. + + m_BlockInfos.resize(blockCount); + for(size_t i = 0; i < blockCount; ++i) + { + m_BlockInfos[i].origBlockIndex = i; + } + + VMA_SORT(m_BlockInfos.begin(), m_BlockInfos.end(), [this](const BlockInfo& lhs, const BlockInfo& rhs) -> bool { + return m_pBlockVector->GetBlock(lhs.origBlockIndex)->m_pMetadata->GetSumFreeSize() < + m_pBlockVector->GetBlock(rhs.origBlockIndex)->m_pMetadata->GetSumFreeSize(); + }); + + // THE MAIN ALGORITHM + + FreeSpaceDatabase freeSpaceDb; + + size_t dstBlockInfoIndex = 0; + size_t dstOrigBlockIndex = m_BlockInfos[dstBlockInfoIndex].origBlockIndex; + VmaDeviceMemoryBlock* pDstBlock = m_pBlockVector->GetBlock(dstOrigBlockIndex); + VmaBlockMetadata_Generic* pDstMetadata = (VmaBlockMetadata_Generic*)pDstBlock->m_pMetadata; + VkDeviceSize dstBlockSize = pDstMetadata->GetSize(); + VkDeviceSize dstOffset = 0; + + bool end = false; + for(size_t srcBlockInfoIndex = 0; !end && srcBlockInfoIndex < blockCount; ++srcBlockInfoIndex) + { + const size_t srcOrigBlockIndex = m_BlockInfos[srcBlockInfoIndex].origBlockIndex; + VmaDeviceMemoryBlock* const pSrcBlock = m_pBlockVector->GetBlock(srcOrigBlockIndex); + VmaBlockMetadata_Generic* const pSrcMetadata = (VmaBlockMetadata_Generic*)pSrcBlock->m_pMetadata; + for(VmaSuballocationList::iterator srcSuballocIt = pSrcMetadata->m_Suballocations.begin(); + !end && srcSuballocIt != pSrcMetadata->m_Suballocations.end(); ) + { + VmaAllocation_T* const pAlloc = srcSuballocIt->hAllocation; + const VkDeviceSize srcAllocAlignment = pAlloc->GetAlignment(); + const VkDeviceSize srcAllocSize = srcSuballocIt->size; + if(m_AllocationsMoved == maxAllocationsToMove || + m_BytesMoved + srcAllocSize > maxBytesToMove) + { + end = true; + break; + } + const VkDeviceSize srcAllocOffset = srcSuballocIt->offset; + + // Try to place it in one of free spaces from the database. + size_t freeSpaceInfoIndex; + VkDeviceSize dstAllocOffset; + if(freeSpaceDb.Fetch(srcAllocAlignment, srcAllocSize, + freeSpaceInfoIndex, dstAllocOffset)) + { + size_t freeSpaceOrigBlockIndex = m_BlockInfos[freeSpaceInfoIndex].origBlockIndex; + VmaDeviceMemoryBlock* pFreeSpaceBlock = m_pBlockVector->GetBlock(freeSpaceOrigBlockIndex); + VmaBlockMetadata_Generic* pFreeSpaceMetadata = (VmaBlockMetadata_Generic*)pFreeSpaceBlock->m_pMetadata; + /*VkDeviceSize freeSpaceBlockSize = pFreeSpaceMetadata->GetSize();*/ + + // Same block + if(freeSpaceInfoIndex == srcBlockInfoIndex) + { + VMA_ASSERT(dstAllocOffset <= srcAllocOffset); + + // MOVE OPTION 1: Move the allocation inside the same block by decreasing offset. + + VmaSuballocation suballoc = *srcSuballocIt; + suballoc.offset = dstAllocOffset; + suballoc.hAllocation->ChangeOffset(dstAllocOffset); + m_BytesMoved += srcAllocSize; + ++m_AllocationsMoved; + + VmaSuballocationList::iterator nextSuballocIt = srcSuballocIt; + ++nextSuballocIt; + pSrcMetadata->m_Suballocations.erase(srcSuballocIt); + srcSuballocIt = nextSuballocIt; + + InsertSuballoc(pFreeSpaceMetadata, suballoc); + + VmaDefragmentationMove move = { + srcOrigBlockIndex, freeSpaceOrigBlockIndex, + srcAllocOffset, dstAllocOffset, + srcAllocSize }; + moves.push_back(move); + } + // Different block + else + { + // MOVE OPTION 2: Move the allocation to a different block. + + VMA_ASSERT(freeSpaceInfoIndex < srcBlockInfoIndex); + + VmaSuballocation suballoc = *srcSuballocIt; + suballoc.offset = dstAllocOffset; + suballoc.hAllocation->ChangeBlockAllocation(m_hAllocator, pFreeSpaceBlock, dstAllocOffset); + m_BytesMoved += srcAllocSize; + ++m_AllocationsMoved; + + VmaSuballocationList::iterator nextSuballocIt = srcSuballocIt; + ++nextSuballocIt; + pSrcMetadata->m_Suballocations.erase(srcSuballocIt); + srcSuballocIt = nextSuballocIt; + + InsertSuballoc(pFreeSpaceMetadata, suballoc); + + VmaDefragmentationMove move = { + srcOrigBlockIndex, freeSpaceOrigBlockIndex, + srcAllocOffset, dstAllocOffset, + srcAllocSize }; + moves.push_back(move); + } + } + else + { + dstAllocOffset = VmaAlignUp(dstOffset, srcAllocAlignment); + + // If the allocation doesn't fit before the end of dstBlock, forward to next block. + while(dstBlockInfoIndex < srcBlockInfoIndex && + dstAllocOffset + srcAllocSize > dstBlockSize) + { + // But before that, register remaining free space at the end of dst block. + freeSpaceDb.Register(dstBlockInfoIndex, dstOffset, dstBlockSize - dstOffset); + + ++dstBlockInfoIndex; + dstOrigBlockIndex = m_BlockInfos[dstBlockInfoIndex].origBlockIndex; + pDstBlock = m_pBlockVector->GetBlock(dstOrigBlockIndex); + pDstMetadata = (VmaBlockMetadata_Generic*)pDstBlock->m_pMetadata; + dstBlockSize = pDstMetadata->GetSize(); + dstOffset = 0; + dstAllocOffset = 0; + } + + // Same block + if(dstBlockInfoIndex == srcBlockInfoIndex) + { + VMA_ASSERT(dstAllocOffset <= srcAllocOffset); + + const bool overlap = dstAllocOffset + srcAllocSize > srcAllocOffset; + + bool skipOver = overlap; + if(overlap && m_OverlappingMoveSupported && dstAllocOffset < srcAllocOffset) + { + // If destination and source place overlap, skip if it would move it + // by only < 1/64 of its size. + skipOver = (srcAllocOffset - dstAllocOffset) * 64 < srcAllocSize; + } + + if(skipOver) + { + freeSpaceDb.Register(dstBlockInfoIndex, dstOffset, srcAllocOffset - dstOffset); + + dstOffset = srcAllocOffset + srcAllocSize; + ++srcSuballocIt; + } + // MOVE OPTION 1: Move the allocation inside the same block by decreasing offset. + else + { + srcSuballocIt->offset = dstAllocOffset; + srcSuballocIt->hAllocation->ChangeOffset(dstAllocOffset); + dstOffset = dstAllocOffset + srcAllocSize; + m_BytesMoved += srcAllocSize; + ++m_AllocationsMoved; + ++srcSuballocIt; + VmaDefragmentationMove move = { + srcOrigBlockIndex, dstOrigBlockIndex, + srcAllocOffset, dstAllocOffset, + srcAllocSize }; + moves.push_back(move); + } + } + // Different block + else + { + // MOVE OPTION 2: Move the allocation to a different block. + + VMA_ASSERT(dstBlockInfoIndex < srcBlockInfoIndex); + VMA_ASSERT(dstAllocOffset + srcAllocSize <= dstBlockSize); + + VmaSuballocation suballoc = *srcSuballocIt; + suballoc.offset = dstAllocOffset; + suballoc.hAllocation->ChangeBlockAllocation(m_hAllocator, pDstBlock, dstAllocOffset); + dstOffset = dstAllocOffset + srcAllocSize; + m_BytesMoved += srcAllocSize; + ++m_AllocationsMoved; + + VmaSuballocationList::iterator nextSuballocIt = srcSuballocIt; + ++nextSuballocIt; + pSrcMetadata->m_Suballocations.erase(srcSuballocIt); + srcSuballocIt = nextSuballocIt; + + pDstMetadata->m_Suballocations.push_back(suballoc); + + VmaDefragmentationMove move = { + srcOrigBlockIndex, dstOrigBlockIndex, + srcAllocOffset, dstAllocOffset, + srcAllocSize }; + moves.push_back(move); + } + } + } + } + + m_BlockInfos.clear(); + + PostprocessMetadata(); + + return VK_SUCCESS; +} + +void VmaDefragmentationAlgorithm_Fast::PreprocessMetadata() +{ + const size_t blockCount = m_pBlockVector->GetBlockCount(); + for(size_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + { + VmaBlockMetadata_Generic* const pMetadata = + (VmaBlockMetadata_Generic*)m_pBlockVector->GetBlock(blockIndex)->m_pMetadata; + pMetadata->m_FreeCount = 0; + pMetadata->m_SumFreeSize = pMetadata->GetSize(); + pMetadata->m_FreeSuballocationsBySize.clear(); + for(VmaSuballocationList::iterator it = pMetadata->m_Suballocations.begin(); + it != pMetadata->m_Suballocations.end(); ) + { + if(it->type == VMA_SUBALLOCATION_TYPE_FREE) + { + VmaSuballocationList::iterator nextIt = it; + ++nextIt; + pMetadata->m_Suballocations.erase(it); + it = nextIt; + } + else + { + ++it; + } + } + } +} + +void VmaDefragmentationAlgorithm_Fast::PostprocessMetadata() +{ + const size_t blockCount = m_pBlockVector->GetBlockCount(); + for(size_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + { + VmaBlockMetadata_Generic* const pMetadata = + (VmaBlockMetadata_Generic*)m_pBlockVector->GetBlock(blockIndex)->m_pMetadata; + const VkDeviceSize blockSize = pMetadata->GetSize(); + + // No allocations in this block - entire area is free. + if(pMetadata->m_Suballocations.empty()) + { + pMetadata->m_FreeCount = 1; + //pMetadata->m_SumFreeSize is already set to blockSize. + VmaSuballocation suballoc = { + 0, // offset + blockSize, // size + VMA_NULL, // hAllocation + VMA_SUBALLOCATION_TYPE_FREE }; + pMetadata->m_Suballocations.push_back(suballoc); + pMetadata->RegisterFreeSuballocation(pMetadata->m_Suballocations.begin()); + } + // There are some allocations in this block. + else + { + VkDeviceSize offset = 0; + VmaSuballocationList::iterator it; + for(it = pMetadata->m_Suballocations.begin(); + it != pMetadata->m_Suballocations.end(); + ++it) + { + VMA_ASSERT(it->type != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(it->offset >= offset); + + // Need to insert preceding free space. + if(it->offset > offset) + { + ++pMetadata->m_FreeCount; + const VkDeviceSize freeSize = it->offset - offset; + VmaSuballocation suballoc = { + offset, // offset + freeSize, // size + VMA_NULL, // hAllocation + VMA_SUBALLOCATION_TYPE_FREE }; + VmaSuballocationList::iterator precedingFreeIt = pMetadata->m_Suballocations.insert(it, suballoc); + if(freeSize >= VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + pMetadata->m_FreeSuballocationsBySize.push_back(precedingFreeIt); + } + } + + pMetadata->m_SumFreeSize -= it->size; + offset = it->offset + it->size; + } + + // Need to insert trailing free space. + if(offset < blockSize) + { + ++pMetadata->m_FreeCount; + const VkDeviceSize freeSize = blockSize - offset; + VmaSuballocation suballoc = { + offset, // offset + freeSize, // size + VMA_NULL, // hAllocation + VMA_SUBALLOCATION_TYPE_FREE }; + VMA_ASSERT(it == pMetadata->m_Suballocations.end()); + VmaSuballocationList::iterator trailingFreeIt = pMetadata->m_Suballocations.insert(it, suballoc); + if(freeSize > VMA_MIN_FREE_SUBALLOCATION_SIZE_TO_REGISTER) + { + pMetadata->m_FreeSuballocationsBySize.push_back(trailingFreeIt); + } + } + + VMA_SORT( + pMetadata->m_FreeSuballocationsBySize.begin(), + pMetadata->m_FreeSuballocationsBySize.end(), + VmaSuballocationItemSizeLess()); + } + + VMA_HEAVY_ASSERT(pMetadata->Validate()); + } +} + +void VmaDefragmentationAlgorithm_Fast::InsertSuballoc(VmaBlockMetadata_Generic* pMetadata, const VmaSuballocation& suballoc) +{ + // TODO: Optimize somehow. Remember iterator instead of searching for it linearly. + VmaSuballocationList::iterator it = pMetadata->m_Suballocations.begin(); + while(it != pMetadata->m_Suballocations.end()) + { + if(it->offset < suballoc.offset) + { + ++it; + } + } + pMetadata->m_Suballocations.insert(it, suballoc); +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaBlockVectorDefragmentationContext + +VmaBlockVectorDefragmentationContext::VmaBlockVectorDefragmentationContext( + VmaAllocator hAllocator, + VmaPool hCustomPool, + VmaBlockVector* pBlockVector, + uint32_t currFrameIndex, + uint32_t /*algorithmFlags*/) : + res(VK_SUCCESS), + mutexLocked(false), + blockContexts(VmaStlAllocator<VmaBlockDefragmentationContext>(hAllocator->GetAllocationCallbacks())), + m_hAllocator(hAllocator), + m_hCustomPool(hCustomPool), + m_pBlockVector(pBlockVector), + m_CurrFrameIndex(currFrameIndex), + /*m_AlgorithmFlags(algorithmFlags),*/ + m_pAlgorithm(VMA_NULL), + m_Allocations(VmaStlAllocator<AllocInfo>(hAllocator->GetAllocationCallbacks())), + m_AllAllocations(false) +{ +} + +VmaBlockVectorDefragmentationContext::~VmaBlockVectorDefragmentationContext() +{ + vma_delete(m_hAllocator, m_pAlgorithm); +} + +void VmaBlockVectorDefragmentationContext::AddAllocation(VmaAllocation hAlloc, VkBool32* pChanged) +{ + AllocInfo info = { hAlloc, pChanged }; + m_Allocations.push_back(info); +} + +void VmaBlockVectorDefragmentationContext::Begin(bool overlappingMoveSupported) +{ + const bool allAllocations = m_AllAllocations || + m_Allocations.size() == m_pBlockVector->CalcAllocationCount(); + + /******************************** + HERE IS THE CHOICE OF DEFRAGMENTATION ALGORITHM. + ********************************/ + + /* + Fast algorithm is supported only when certain criteria are met: + - VMA_DEBUG_MARGIN is 0. + - All allocations in this block vector are moveable. + - There is no possibility of image/buffer granularity conflict. + */ + if(VMA_DEBUG_MARGIN == 0 && + allAllocations && + !m_pBlockVector->IsBufferImageGranularityConflictPossible()) + { + m_pAlgorithm = vma_new(m_hAllocator, VmaDefragmentationAlgorithm_Fast)( + m_hAllocator, m_pBlockVector, m_CurrFrameIndex, overlappingMoveSupported); + } + else + { + m_pAlgorithm = vma_new(m_hAllocator, VmaDefragmentationAlgorithm_Generic)( + m_hAllocator, m_pBlockVector, m_CurrFrameIndex, overlappingMoveSupported); + } + + if(allAllocations) + { + m_pAlgorithm->AddAll(); + } + else + { + for(size_t i = 0, count = m_Allocations.size(); i < count; ++i) + { + m_pAlgorithm->AddAllocation(m_Allocations[i].hAlloc, m_Allocations[i].pChanged); + } + } +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaDefragmentationContext + +VmaDefragmentationContext_T::VmaDefragmentationContext_T( + VmaAllocator hAllocator, + uint32_t currFrameIndex, + uint32_t flags, + VmaDefragmentationStats* pStats) : + m_hAllocator(hAllocator), + m_CurrFrameIndex(currFrameIndex), + m_Flags(flags), + m_pStats(pStats), + m_CustomPoolContexts(VmaStlAllocator<VmaBlockVectorDefragmentationContext*>(hAllocator->GetAllocationCallbacks())) +{ + memset(m_DefaultPoolContexts, 0, sizeof(m_DefaultPoolContexts)); +} + +VmaDefragmentationContext_T::~VmaDefragmentationContext_T() +{ + for(size_t i = m_CustomPoolContexts.size(); i--; ) + { + VmaBlockVectorDefragmentationContext* pBlockVectorCtx = m_CustomPoolContexts[i]; + pBlockVectorCtx->GetBlockVector()->DefragmentationEnd(pBlockVectorCtx, m_pStats); + vma_delete(m_hAllocator, pBlockVectorCtx); + } + for(size_t i = m_hAllocator->m_MemProps.memoryTypeCount; i--; ) + { + VmaBlockVectorDefragmentationContext* pBlockVectorCtx = m_DefaultPoolContexts[i]; + if(pBlockVectorCtx) + { + pBlockVectorCtx->GetBlockVector()->DefragmentationEnd(pBlockVectorCtx, m_pStats); + vma_delete(m_hAllocator, pBlockVectorCtx); + } + } +} + +void VmaDefragmentationContext_T::AddPools(uint32_t poolCount, VmaPool* pPools) +{ + for(uint32_t poolIndex = 0; poolIndex < poolCount; ++poolIndex) + { + VmaPool pool = pPools[poolIndex]; + VMA_ASSERT(pool); + // Pools with algorithm other than default are not defragmented. + if(pool->m_BlockVector.GetAlgorithm() == 0) + { + VmaBlockVectorDefragmentationContext* pBlockVectorDefragCtx = VMA_NULL; + + for(size_t i = m_CustomPoolContexts.size(); i--; ) + { + if(m_CustomPoolContexts[i]->GetCustomPool() == pool) + { + pBlockVectorDefragCtx = m_CustomPoolContexts[i]; + break; + } + } + + if(!pBlockVectorDefragCtx) + { + pBlockVectorDefragCtx = vma_new(m_hAllocator, VmaBlockVectorDefragmentationContext)( + m_hAllocator, + pool, + &pool->m_BlockVector, + m_CurrFrameIndex, + m_Flags); + m_CustomPoolContexts.push_back(pBlockVectorDefragCtx); + } + + pBlockVectorDefragCtx->AddAll(); + } + } +} + +void VmaDefragmentationContext_T::AddAllocations( + uint32_t allocationCount, + VmaAllocation* pAllocations, + VkBool32* pAllocationsChanged) +{ + // Dispatch pAllocations among defragmentators. Create them when necessary. + for(uint32_t allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + const VmaAllocation hAlloc = pAllocations[allocIndex]; + VMA_ASSERT(hAlloc); + // DedicatedAlloc cannot be defragmented. + if((hAlloc->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK) && + // Lost allocation cannot be defragmented. + (hAlloc->GetLastUseFrameIndex() != VMA_FRAME_INDEX_LOST)) + { + VmaBlockVectorDefragmentationContext* pBlockVectorDefragCtx = VMA_NULL; + + const VmaPool hAllocPool = hAlloc->GetPool(); + // This allocation belongs to custom pool. + if(hAllocPool != VK_NULL_HANDLE) + { + // Pools with algorithm other than default are not defragmented. + if(hAllocPool->m_BlockVector.GetAlgorithm() == 0) + { + for(size_t i = m_CustomPoolContexts.size(); i--; ) + { + if(m_CustomPoolContexts[i]->GetCustomPool() == hAllocPool) + { + pBlockVectorDefragCtx = m_CustomPoolContexts[i]; + break; + } + } + if(!pBlockVectorDefragCtx) + { + pBlockVectorDefragCtx = vma_new(m_hAllocator, VmaBlockVectorDefragmentationContext)( + m_hAllocator, + hAllocPool, + &hAllocPool->m_BlockVector, + m_CurrFrameIndex, + m_Flags); + m_CustomPoolContexts.push_back(pBlockVectorDefragCtx); + } + } + } + // This allocation belongs to default pool. + else + { + const uint32_t memTypeIndex = hAlloc->GetMemoryTypeIndex(); + pBlockVectorDefragCtx = m_DefaultPoolContexts[memTypeIndex]; + if(!pBlockVectorDefragCtx) + { + pBlockVectorDefragCtx = vma_new(m_hAllocator, VmaBlockVectorDefragmentationContext)( + m_hAllocator, + VMA_NULL, // hCustomPool + m_hAllocator->m_pBlockVectors[memTypeIndex], + m_CurrFrameIndex, + m_Flags); + m_DefaultPoolContexts[memTypeIndex] = pBlockVectorDefragCtx; + } + } + + if(pBlockVectorDefragCtx) + { + VkBool32* const pChanged = (pAllocationsChanged != VMA_NULL) ? + &pAllocationsChanged[allocIndex] : VMA_NULL; + pBlockVectorDefragCtx->AddAllocation(hAlloc, pChanged); + } + } + } +} + +VkResult VmaDefragmentationContext_T::Defragment( + VkDeviceSize maxCpuBytesToMove, uint32_t maxCpuAllocationsToMove, + VkDeviceSize maxGpuBytesToMove, uint32_t maxGpuAllocationsToMove, + VkCommandBuffer commandBuffer, VmaDefragmentationStats* pStats) +{ + if(pStats) + { + memset(pStats, 0, sizeof(VmaDefragmentationStats)); + } + + if(commandBuffer == VK_NULL_HANDLE) + { + maxGpuBytesToMove = 0; + maxGpuAllocationsToMove = 0; + } + + VkResult res = VK_SUCCESS; + + // Process default pools. + for(uint32_t memTypeIndex = 0; + memTypeIndex < m_hAllocator->GetMemoryTypeCount() && res >= VK_SUCCESS; + ++memTypeIndex) + { + VmaBlockVectorDefragmentationContext* pBlockVectorCtx = m_DefaultPoolContexts[memTypeIndex]; + if(pBlockVectorCtx) + { + VMA_ASSERT(pBlockVectorCtx->GetBlockVector()); + pBlockVectorCtx->GetBlockVector()->Defragment( + pBlockVectorCtx, + pStats, + maxCpuBytesToMove, maxCpuAllocationsToMove, + maxGpuBytesToMove, maxGpuAllocationsToMove, + commandBuffer); + if(pBlockVectorCtx->res != VK_SUCCESS) + { + res = pBlockVectorCtx->res; + } + } + } + + // Process custom pools. + for(size_t customCtxIndex = 0, customCtxCount = m_CustomPoolContexts.size(); + customCtxIndex < customCtxCount && res >= VK_SUCCESS; + ++customCtxIndex) + { + VmaBlockVectorDefragmentationContext* pBlockVectorCtx = m_CustomPoolContexts[customCtxIndex]; + VMA_ASSERT(pBlockVectorCtx && pBlockVectorCtx->GetBlockVector()); + pBlockVectorCtx->GetBlockVector()->Defragment( + pBlockVectorCtx, + pStats, + maxCpuBytesToMove, maxCpuAllocationsToMove, + maxGpuBytesToMove, maxGpuAllocationsToMove, + commandBuffer); + if(pBlockVectorCtx->res != VK_SUCCESS) + { + res = pBlockVectorCtx->res; + } + } + + return res; +} + +//////////////////////////////////////////////////////////////////////////////// +// VmaRecorder + +#if VMA_RECORDING_ENABLED + +VmaRecorder::VmaRecorder() : + m_UseMutex(true), + m_Flags(0), + m_File(VMA_NULL), + m_Freq(INT64_MAX), + m_StartCounter(INT64_MAX) +{ +} + +VkResult VmaRecorder::Init(const VmaRecordSettings& settings, bool useMutex) +{ + m_UseMutex = useMutex; + m_Flags = settings.flags; + + QueryPerformanceFrequency((LARGE_INTEGER*)&m_Freq); + QueryPerformanceCounter((LARGE_INTEGER*)&m_StartCounter); + + // Open file for writing. + errno_t err = fopen_s(&m_File, settings.pFilePath, "wb"); + if(err != 0) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + + // Write header. + fprintf(m_File, "%s\n", "Vulkan Memory Allocator,Calls recording"); + fprintf(m_File, "%s\n", "1,5"); + + return VK_SUCCESS; +} + +VmaRecorder::~VmaRecorder() +{ + if(m_File != VMA_NULL) + { + fclose(m_File); + } +} + +void VmaRecorder::RecordCreateAllocator(uint32_t frameIndex) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaCreateAllocator\n", callParams.threadId, callParams.time, frameIndex); + Flush(); +} + +void VmaRecorder::RecordDestroyAllocator(uint32_t frameIndex) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDestroyAllocator\n", callParams.threadId, callParams.time, frameIndex); + Flush(); +} + +void VmaRecorder::RecordCreatePool(uint32_t frameIndex, const VmaPoolCreateInfo& createInfo, VmaPool pool) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaCreatePool,%u,%u,%llu,%llu,%llu,%u,%p\n", callParams.threadId, callParams.time, frameIndex, + createInfo.memoryTypeIndex, + createInfo.flags, + createInfo.blockSize, + (uint64_t)createInfo.minBlockCount, + (uint64_t)createInfo.maxBlockCount, + createInfo.frameInUseCount, + pool); + Flush(); +} + +void VmaRecorder::RecordDestroyPool(uint32_t frameIndex, VmaPool pool) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDestroyPool,%p\n", callParams.threadId, callParams.time, frameIndex, + pool); + Flush(); +} + +void VmaRecorder::RecordAllocateMemory(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(createInfo.flags, createInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaAllocateMemory,%llu,%llu,%u,%u,%u,%u,%u,%u,%p,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + vkMemReq.size, + vkMemReq.alignment, + vkMemReq.memoryTypeBits, + createInfo.flags, + createInfo.usage, + createInfo.requiredFlags, + createInfo.preferredFlags, + createInfo.memoryTypeBits, + createInfo.pool, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordAllocateMemoryPages(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + const VmaAllocationCreateInfo& createInfo, + uint64_t allocationCount, + const VmaAllocation* pAllocations) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(createInfo.flags, createInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaAllocateMemoryPages,%llu,%llu,%u,%u,%u,%u,%u,%u,%p,", callParams.threadId, callParams.time, frameIndex, + vkMemReq.size, + vkMemReq.alignment, + vkMemReq.memoryTypeBits, + createInfo.flags, + createInfo.usage, + createInfo.requiredFlags, + createInfo.preferredFlags, + createInfo.memoryTypeBits, + createInfo.pool); + PrintPointerList(allocationCount, pAllocations); + fprintf(m_File, ",%s\n", userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordAllocateMemoryForBuffer(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(createInfo.flags, createInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaAllocateMemoryForBuffer,%llu,%llu,%u,%u,%u,%u,%u,%u,%u,%u,%p,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + vkMemReq.size, + vkMemReq.alignment, + vkMemReq.memoryTypeBits, + requiresDedicatedAllocation ? 1 : 0, + prefersDedicatedAllocation ? 1 : 0, + createInfo.flags, + createInfo.usage, + createInfo.requiredFlags, + createInfo.preferredFlags, + createInfo.memoryTypeBits, + createInfo.pool, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordAllocateMemoryForImage(uint32_t frameIndex, + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + const VmaAllocationCreateInfo& createInfo, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(createInfo.flags, createInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaAllocateMemoryForImage,%llu,%llu,%u,%u,%u,%u,%u,%u,%u,%u,%p,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + vkMemReq.size, + vkMemReq.alignment, + vkMemReq.memoryTypeBits, + requiresDedicatedAllocation ? 1 : 0, + prefersDedicatedAllocation ? 1 : 0, + createInfo.flags, + createInfo.usage, + createInfo.requiredFlags, + createInfo.preferredFlags, + createInfo.memoryTypeBits, + createInfo.pool, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordFreeMemory(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaFreeMemory,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordFreeMemoryPages(uint32_t frameIndex, + uint64_t allocationCount, + const VmaAllocation* pAllocations) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaFreeMemoryPages,", callParams.threadId, callParams.time, frameIndex); + PrintPointerList(allocationCount, pAllocations); + fprintf(m_File, "\n"); + Flush(); +} + +void VmaRecorder::RecordResizeAllocation( + uint32_t frameIndex, + VmaAllocation allocation, + VkDeviceSize newSize) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaResizeAllocation,%p,%llu\n", callParams.threadId, callParams.time, frameIndex, + allocation, newSize); + Flush(); +} + +void VmaRecorder::RecordSetAllocationUserData(uint32_t frameIndex, + VmaAllocation allocation, + const void* pUserData) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr( + allocation->IsUserDataString() ? VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT : 0, + pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaSetAllocationUserData,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordCreateLostAllocation(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaCreateLostAllocation,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordMapMemory(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaMapMemory,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordUnmapMemory(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaUnmapMemory,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordFlushAllocation(uint32_t frameIndex, + VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaFlushAllocation,%p,%llu,%llu\n", callParams.threadId, callParams.time, frameIndex, + allocation, + offset, + size); + Flush(); +} + +void VmaRecorder::RecordInvalidateAllocation(uint32_t frameIndex, + VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaInvalidateAllocation,%p,%llu,%llu\n", callParams.threadId, callParams.time, frameIndex, + allocation, + offset, + size); + Flush(); +} + +void VmaRecorder::RecordCreateBuffer(uint32_t frameIndex, + const VkBufferCreateInfo& bufCreateInfo, + const VmaAllocationCreateInfo& allocCreateInfo, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(allocCreateInfo.flags, allocCreateInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaCreateBuffer,%u,%llu,%u,%u,%u,%u,%u,%u,%u,%p,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + bufCreateInfo.flags, + bufCreateInfo.size, + bufCreateInfo.usage, + bufCreateInfo.sharingMode, + allocCreateInfo.flags, + allocCreateInfo.usage, + allocCreateInfo.requiredFlags, + allocCreateInfo.preferredFlags, + allocCreateInfo.memoryTypeBits, + allocCreateInfo.pool, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordCreateImage(uint32_t frameIndex, + const VkImageCreateInfo& imageCreateInfo, + const VmaAllocationCreateInfo& allocCreateInfo, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + UserDataString userDataStr(allocCreateInfo.flags, allocCreateInfo.pUserData); + fprintf(m_File, "%u,%.3f,%u,vmaCreateImage,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%p,%p,%s\n", callParams.threadId, callParams.time, frameIndex, + imageCreateInfo.flags, + imageCreateInfo.imageType, + imageCreateInfo.format, + imageCreateInfo.extent.width, + imageCreateInfo.extent.height, + imageCreateInfo.extent.depth, + imageCreateInfo.mipLevels, + imageCreateInfo.arrayLayers, + imageCreateInfo.samples, + imageCreateInfo.tiling, + imageCreateInfo.usage, + imageCreateInfo.sharingMode, + imageCreateInfo.initialLayout, + allocCreateInfo.flags, + allocCreateInfo.usage, + allocCreateInfo.requiredFlags, + allocCreateInfo.preferredFlags, + allocCreateInfo.memoryTypeBits, + allocCreateInfo.pool, + allocation, + userDataStr.GetString()); + Flush(); +} + +void VmaRecorder::RecordDestroyBuffer(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDestroyBuffer,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordDestroyImage(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDestroyImage,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordTouchAllocation(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaTouchAllocation,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordGetAllocationInfo(uint32_t frameIndex, + VmaAllocation allocation) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaGetAllocationInfo,%p\n", callParams.threadId, callParams.time, frameIndex, + allocation); + Flush(); +} + +void VmaRecorder::RecordMakePoolAllocationsLost(uint32_t frameIndex, + VmaPool pool) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaMakePoolAllocationsLost,%p\n", callParams.threadId, callParams.time, frameIndex, + pool); + Flush(); +} + +void VmaRecorder::RecordDefragmentationBegin(uint32_t frameIndex, + const VmaDefragmentationInfo2& info, + VmaDefragmentationContext ctx) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDefragmentationBegin,%u,", callParams.threadId, callParams.time, frameIndex, + info.flags); + PrintPointerList(info.allocationCount, info.pAllocations); + fprintf(m_File, ","); + PrintPointerList(info.poolCount, info.pPools); + fprintf(m_File, ",%llu,%u,%llu,%u,%p,%p\n", + info.maxCpuBytesToMove, + info.maxCpuAllocationsToMove, + info.maxGpuBytesToMove, + info.maxGpuAllocationsToMove, + info.commandBuffer, + ctx); + Flush(); +} + +void VmaRecorder::RecordDefragmentationEnd(uint32_t frameIndex, + VmaDefragmentationContext ctx) +{ + CallParams callParams; + GetBasicParams(callParams); + + VmaMutexLock lock(m_FileMutex, m_UseMutex); + fprintf(m_File, "%u,%.3f,%u,vmaDefragmentationEnd,%p\n", callParams.threadId, callParams.time, frameIndex, + ctx); + Flush(); +} + +VmaRecorder::UserDataString::UserDataString(VmaAllocationCreateFlags allocFlags, const void* pUserData) +{ + if(pUserData != VMA_NULL) + { + if((allocFlags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0) + { + m_Str = (const char*)pUserData; + } + else + { + sprintf_s(m_PtrStr, "%p", pUserData); + m_Str = m_PtrStr; + } + } + else + { + m_Str = ""; + } +} + +void VmaRecorder::WriteConfiguration( + const VkPhysicalDeviceProperties& devProps, + const VkPhysicalDeviceMemoryProperties& memProps, + bool dedicatedAllocationExtensionEnabled) +{ + fprintf(m_File, "Config,Begin\n"); + + fprintf(m_File, "PhysicalDevice,apiVersion,%u\n", devProps.apiVersion); + fprintf(m_File, "PhysicalDevice,driverVersion,%u\n", devProps.driverVersion); + fprintf(m_File, "PhysicalDevice,vendorID,%u\n", devProps.vendorID); + fprintf(m_File, "PhysicalDevice,deviceID,%u\n", devProps.deviceID); + fprintf(m_File, "PhysicalDevice,deviceType,%u\n", devProps.deviceType); + fprintf(m_File, "PhysicalDevice,deviceName,%s\n", devProps.deviceName); + + fprintf(m_File, "PhysicalDeviceLimits,maxMemoryAllocationCount,%u\n", devProps.limits.maxMemoryAllocationCount); + fprintf(m_File, "PhysicalDeviceLimits,bufferImageGranularity,%llu\n", devProps.limits.bufferImageGranularity); + fprintf(m_File, "PhysicalDeviceLimits,nonCoherentAtomSize,%llu\n", devProps.limits.nonCoherentAtomSize); + + fprintf(m_File, "PhysicalDeviceMemory,HeapCount,%u\n", memProps.memoryHeapCount); + for(uint32_t i = 0; i < memProps.memoryHeapCount; ++i) + { + fprintf(m_File, "PhysicalDeviceMemory,Heap,%u,size,%llu\n", i, memProps.memoryHeaps[i].size); + fprintf(m_File, "PhysicalDeviceMemory,Heap,%u,flags,%u\n", i, memProps.memoryHeaps[i].flags); + } + fprintf(m_File, "PhysicalDeviceMemory,TypeCount,%u\n", memProps.memoryTypeCount); + for(uint32_t i = 0; i < memProps.memoryTypeCount; ++i) + { + fprintf(m_File, "PhysicalDeviceMemory,Type,%u,heapIndex,%u\n", i, memProps.memoryTypes[i].heapIndex); + fprintf(m_File, "PhysicalDeviceMemory,Type,%u,propertyFlags,%u\n", i, memProps.memoryTypes[i].propertyFlags); + } + + fprintf(m_File, "Extension,VK_KHR_dedicated_allocation,%u\n", dedicatedAllocationExtensionEnabled ? 1 : 0); + + fprintf(m_File, "Macro,VMA_DEBUG_ALWAYS_DEDICATED_MEMORY,%u\n", VMA_DEBUG_ALWAYS_DEDICATED_MEMORY ? 1 : 0); + fprintf(m_File, "Macro,VMA_DEBUG_ALIGNMENT,%llu\n", (VkDeviceSize)VMA_DEBUG_ALIGNMENT); + fprintf(m_File, "Macro,VMA_DEBUG_MARGIN,%llu\n", (VkDeviceSize)VMA_DEBUG_MARGIN); + fprintf(m_File, "Macro,VMA_DEBUG_INITIALIZE_ALLOCATIONS,%u\n", VMA_DEBUG_INITIALIZE_ALLOCATIONS ? 1 : 0); + fprintf(m_File, "Macro,VMA_DEBUG_DETECT_CORRUPTION,%u\n", VMA_DEBUG_DETECT_CORRUPTION ? 1 : 0); + fprintf(m_File, "Macro,VMA_DEBUG_GLOBAL_MUTEX,%u\n", VMA_DEBUG_GLOBAL_MUTEX ? 1 : 0); + fprintf(m_File, "Macro,VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY,%llu\n", (VkDeviceSize)VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY); + fprintf(m_File, "Macro,VMA_SMALL_HEAP_MAX_SIZE,%llu\n", (VkDeviceSize)VMA_SMALL_HEAP_MAX_SIZE); + fprintf(m_File, "Macro,VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE,%llu\n", (VkDeviceSize)VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE); + + fprintf(m_File, "Config,End\n"); +} + +void VmaRecorder::GetBasicParams(CallParams& outParams) +{ + outParams.threadId = GetCurrentThreadId(); + + LARGE_INTEGER counter; + QueryPerformanceCounter(&counter); + outParams.time = (double)(counter.QuadPart - m_StartCounter) / (double)m_Freq; +} + +void VmaRecorder::PrintPointerList(uint64_t count, const VmaAllocation* pItems) +{ + if(count) + { + fprintf(m_File, "%p", pItems[0]); + for(uint64_t i = 1; i < count; ++i) + { + fprintf(m_File, " %p", pItems[i]); + } + } +} + +void VmaRecorder::Flush() +{ + if((m_Flags & VMA_RECORD_FLUSH_AFTER_CALL_BIT) != 0) + { + fflush(m_File); + } +} + +#endif // #if VMA_RECORDING_ENABLED + +//////////////////////////////////////////////////////////////////////////////// +// VmaAllocator_T + +VmaAllocator_T::VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo) : + m_UseMutex((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT) == 0), + m_UseKhrDedicatedAllocation((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT) != 0), + m_hDevice(pCreateInfo->device), + m_AllocationCallbacksSpecified(pCreateInfo->pAllocationCallbacks != VMA_NULL), + m_AllocationCallbacks(pCreateInfo->pAllocationCallbacks ? + *pCreateInfo->pAllocationCallbacks : VmaEmptyAllocationCallbacks), + m_PreferredLargeHeapBlockSize(0), + m_PhysicalDevice(pCreateInfo->physicalDevice), + m_CurrentFrameIndex(0), + m_Pools(VmaStlAllocator<VmaPool>(GetAllocationCallbacks())), + m_NextPoolId(0) +#if VMA_RECORDING_ENABLED + ,m_pRecorder(VMA_NULL) +#endif +{ + if(VMA_DEBUG_DETECT_CORRUPTION) + { + // Needs to be multiply of uint32_t size because we are going to write VMA_CORRUPTION_DETECTION_MAGIC_VALUE to it. + VMA_ASSERT(VMA_DEBUG_MARGIN % sizeof(uint32_t) == 0); + } + + VMA_ASSERT(pCreateInfo->physicalDevice && pCreateInfo->device); + +#if !(VMA_DEDICATED_ALLOCATION) + if((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT) != 0) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT set but required extensions are disabled by preprocessor macros."); + } +#endif + + memset(&m_DeviceMemoryCallbacks, 0 ,sizeof(m_DeviceMemoryCallbacks)); + memset(&m_PhysicalDeviceProperties, 0, sizeof(m_PhysicalDeviceProperties)); + memset(&m_MemProps, 0, sizeof(m_MemProps)); + + memset(&m_pBlockVectors, 0, sizeof(m_pBlockVectors)); + memset(&m_pDedicatedAllocations, 0, sizeof(m_pDedicatedAllocations)); + + for(uint32_t i = 0; i < VK_MAX_MEMORY_HEAPS; ++i) + { + m_HeapSizeLimit[i] = VK_WHOLE_SIZE; + } + + if(pCreateInfo->pDeviceMemoryCallbacks != VMA_NULL) + { + m_DeviceMemoryCallbacks.pfnAllocate = pCreateInfo->pDeviceMemoryCallbacks->pfnAllocate; + m_DeviceMemoryCallbacks.pfnFree = pCreateInfo->pDeviceMemoryCallbacks->pfnFree; + } + + ImportVulkanFunctions(pCreateInfo->pVulkanFunctions); + + (*m_VulkanFunctions.vkGetPhysicalDeviceProperties)(m_PhysicalDevice, &m_PhysicalDeviceProperties); + (*m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties)(m_PhysicalDevice, &m_MemProps); + + VMA_ASSERT(VmaIsPow2(VMA_DEBUG_ALIGNMENT)); + VMA_ASSERT(VmaIsPow2(VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY)); + VMA_ASSERT(VmaIsPow2(m_PhysicalDeviceProperties.limits.bufferImageGranularity)); + VMA_ASSERT(VmaIsPow2(m_PhysicalDeviceProperties.limits.nonCoherentAtomSize)); + + m_PreferredLargeHeapBlockSize = (pCreateInfo->preferredLargeHeapBlockSize != 0) ? + pCreateInfo->preferredLargeHeapBlockSize : static_cast<VkDeviceSize>(VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE); + + if(pCreateInfo->pHeapSizeLimit != VMA_NULL) + { + for(uint32_t heapIndex = 0; heapIndex < GetMemoryHeapCount(); ++heapIndex) + { + const VkDeviceSize limit = pCreateInfo->pHeapSizeLimit[heapIndex]; + if(limit != VK_WHOLE_SIZE) + { + m_HeapSizeLimit[heapIndex] = limit; + if(limit < m_MemProps.memoryHeaps[heapIndex].size) + { + m_MemProps.memoryHeaps[heapIndex].size = limit; + } + } + } + } + + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + const VkDeviceSize preferredBlockSize = CalcPreferredBlockSize(memTypeIndex); + + m_pBlockVectors[memTypeIndex] = vma_new(this, VmaBlockVector)( + this, + memTypeIndex, + preferredBlockSize, + 0, + SIZE_MAX, + GetBufferImageGranularity(), + pCreateInfo->frameInUseCount, + false, // isCustomPool + false, // explicitBlockSize + false); // linearAlgorithm + // No need to call m_pBlockVectors[memTypeIndex][blockVectorTypeIndex]->CreateMinBlocks here, + // becase minBlockCount is 0. + m_pDedicatedAllocations[memTypeIndex] = vma_new(this, AllocationVectorType)(VmaStlAllocator<VmaAllocation>(GetAllocationCallbacks())); + + } +} + +VkResult VmaAllocator_T::Init(const VmaAllocatorCreateInfo* pCreateInfo) +{ + VkResult res = VK_SUCCESS; + + if(pCreateInfo->pRecordSettings != VMA_NULL && + !VmaStrIsEmpty(pCreateInfo->pRecordSettings->pFilePath)) + { +#if VMA_RECORDING_ENABLED + m_pRecorder = vma_new(this, VmaRecorder)(); + res = m_pRecorder->Init(*pCreateInfo->pRecordSettings, m_UseMutex); + if(res != VK_SUCCESS) + { + return res; + } + m_pRecorder->WriteConfiguration( + m_PhysicalDeviceProperties, + m_MemProps, + m_UseKhrDedicatedAllocation); + m_pRecorder->RecordCreateAllocator(GetCurrentFrameIndex()); +#else + VMA_ASSERT(0 && "VmaAllocatorCreateInfo::pRecordSettings used, but not supported due to VMA_RECORDING_ENABLED not defined to 1."); + return VK_ERROR_FEATURE_NOT_PRESENT; +#endif + } + + return res; +} + +VmaAllocator_T::~VmaAllocator_T() +{ +#if VMA_RECORDING_ENABLED + if(m_pRecorder != VMA_NULL) + { + m_pRecorder->RecordDestroyAllocator(GetCurrentFrameIndex()); + vma_delete(this, m_pRecorder); + } +#endif + + VMA_ASSERT(m_Pools.empty()); + + for(size_t i = GetMemoryTypeCount(); i--; ) + { + vma_delete(this, m_pDedicatedAllocations[i]); + vma_delete(this, m_pBlockVectors[i]); + } +} + +void VmaAllocator_T::ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunctions) +{ +#if VMA_STATIC_VULKAN_FUNCTIONS == 1 + m_VulkanFunctions.vkGetPhysicalDeviceProperties = &vkGetPhysicalDeviceProperties; + m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties = &vkGetPhysicalDeviceMemoryProperties; + m_VulkanFunctions.vkAllocateMemory = &vkAllocateMemory; + m_VulkanFunctions.vkFreeMemory = &vkFreeMemory; + m_VulkanFunctions.vkMapMemory = &vkMapMemory; + m_VulkanFunctions.vkUnmapMemory = &vkUnmapMemory; + m_VulkanFunctions.vkFlushMappedMemoryRanges = &vkFlushMappedMemoryRanges; + m_VulkanFunctions.vkInvalidateMappedMemoryRanges = &vkInvalidateMappedMemoryRanges; + m_VulkanFunctions.vkBindBufferMemory = &vkBindBufferMemory; + m_VulkanFunctions.vkBindImageMemory = &vkBindImageMemory; + m_VulkanFunctions.vkGetBufferMemoryRequirements = &vkGetBufferMemoryRequirements; + m_VulkanFunctions.vkGetImageMemoryRequirements = &vkGetImageMemoryRequirements; + m_VulkanFunctions.vkCreateBuffer = &vkCreateBuffer; + m_VulkanFunctions.vkDestroyBuffer = &vkDestroyBuffer; + m_VulkanFunctions.vkCreateImage = &vkCreateImage; + m_VulkanFunctions.vkDestroyImage = &vkDestroyImage; + m_VulkanFunctions.vkCmdCopyBuffer = &vkCmdCopyBuffer; +#if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { + m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR = + (PFN_vkGetBufferMemoryRequirements2KHR)vkGetDeviceProcAddr(m_hDevice, "vkGetBufferMemoryRequirements2KHR"); + m_VulkanFunctions.vkGetImageMemoryRequirements2KHR = + (PFN_vkGetImageMemoryRequirements2KHR)vkGetDeviceProcAddr(m_hDevice, "vkGetImageMemoryRequirements2KHR"); + } +#endif // #if VMA_DEDICATED_ALLOCATION +#endif // #if VMA_STATIC_VULKAN_FUNCTIONS == 1 + +#define VMA_COPY_IF_NOT_NULL(funcName) \ + if(pVulkanFunctions->funcName != VMA_NULL) m_VulkanFunctions.funcName = pVulkanFunctions->funcName; + + if(pVulkanFunctions != VMA_NULL) + { + VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceProperties); + VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceMemoryProperties); + VMA_COPY_IF_NOT_NULL(vkAllocateMemory); + VMA_COPY_IF_NOT_NULL(vkFreeMemory); + VMA_COPY_IF_NOT_NULL(vkMapMemory); + VMA_COPY_IF_NOT_NULL(vkUnmapMemory); + VMA_COPY_IF_NOT_NULL(vkFlushMappedMemoryRanges); + VMA_COPY_IF_NOT_NULL(vkInvalidateMappedMemoryRanges); + VMA_COPY_IF_NOT_NULL(vkBindBufferMemory); + VMA_COPY_IF_NOT_NULL(vkBindImageMemory); + VMA_COPY_IF_NOT_NULL(vkGetBufferMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkGetImageMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkCreateBuffer); + VMA_COPY_IF_NOT_NULL(vkDestroyBuffer); + VMA_COPY_IF_NOT_NULL(vkCreateImage); + VMA_COPY_IF_NOT_NULL(vkDestroyImage); + VMA_COPY_IF_NOT_NULL(vkCmdCopyBuffer); +#if VMA_DEDICATED_ALLOCATION + VMA_COPY_IF_NOT_NULL(vkGetBufferMemoryRequirements2KHR); + VMA_COPY_IF_NOT_NULL(vkGetImageMemoryRequirements2KHR); +#endif + } + +#undef VMA_COPY_IF_NOT_NULL + + // If these asserts are hit, you must either #define VMA_STATIC_VULKAN_FUNCTIONS 1 + // or pass valid pointers as VmaAllocatorCreateInfo::pVulkanFunctions. + VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceProperties != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkAllocateMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkFreeMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkMapMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkUnmapMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkFlushMappedMemoryRanges != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkInvalidateMappedMemoryRanges != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkBindBufferMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkBindImageMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetBufferMemoryRequirements != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetImageMemoryRequirements != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCreateBuffer != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkDestroyBuffer != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCreateImage != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkDestroyImage != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCmdCopyBuffer != VMA_NULL); +#if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { + VMA_ASSERT(m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetImageMemoryRequirements2KHR != VMA_NULL); + } +#endif +} + +VkDeviceSize VmaAllocator_T::CalcPreferredBlockSize(uint32_t memTypeIndex) +{ + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); + const VkDeviceSize heapSize = m_MemProps.memoryHeaps[heapIndex].size; + const bool isSmallHeap = heapSize <= VMA_SMALL_HEAP_MAX_SIZE; + return isSmallHeap ? (heapSize / 8) : m_PreferredLargeHeapBlockSize; +} + +VkResult VmaAllocator_T::AllocateMemoryOfType( + VkDeviceSize size, + VkDeviceSize alignment, + bool dedicatedAllocation, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + const VmaAllocationCreateInfo& createInfo, + uint32_t memTypeIndex, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + VMA_ASSERT(pAllocations != VMA_NULL); + VMA_DEBUG_LOG(" AllocateMemory: MemoryTypeIndex=%u, AllocationCount=%zu, Size=%llu", memTypeIndex, allocationCount, vkMemReq.size); + + VmaAllocationCreateInfo finalCreateInfo = createInfo; + + // If memory type is not HOST_VISIBLE, disable MAPPED. + if((finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0 && + (m_MemProps.memoryTypes[memTypeIndex].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + finalCreateInfo.flags &= ~VMA_ALLOCATION_CREATE_MAPPED_BIT; + } + + VmaBlockVector* const blockVector = m_pBlockVectors[memTypeIndex]; + VMA_ASSERT(blockVector); + + const VkDeviceSize preferredBlockSize = blockVector->GetPreferredBlockSize(); + bool preferDedicatedMemory = + VMA_DEBUG_ALWAYS_DEDICATED_MEMORY || + dedicatedAllocation || + // Heuristics: Allocate dedicated memory if requested size if greater than half of preferred block size. + size > preferredBlockSize / 2; + + if(preferDedicatedMemory && + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) == 0 && + finalCreateInfo.pool == VK_NULL_HANDLE) + { + finalCreateInfo.flags |= VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT; + } + + if((finalCreateInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0) + { + if((finalCreateInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) != 0) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + else + { + return AllocateDedicatedMemory( + size, + suballocType, + memTypeIndex, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, + finalCreateInfo.pUserData, + dedicatedBuffer, + dedicatedImage, + allocationCount, + pAllocations); + } + } + else + { + VkResult res = blockVector->Allocate( + VK_NULL_HANDLE, // hCurrentPool + m_CurrentFrameIndex.load(), + size, + alignment, + finalCreateInfo, + suballocType, + allocationCount, + pAllocations); + if(res == VK_SUCCESS) + { + return res; + } + + // 5. Try dedicated memory. + if((finalCreateInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) != 0) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + else + { + res = AllocateDedicatedMemory( + size, + suballocType, + memTypeIndex, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, + finalCreateInfo.pUserData, + dedicatedBuffer, + dedicatedImage, + allocationCount, + pAllocations); + if(res == VK_SUCCESS) + { + // Succeeded: AllocateDedicatedMemory function already filld pMemory, nothing more to do here. + VMA_DEBUG_LOG(" Allocated as DedicatedMemory"); + return VK_SUCCESS; + } + else + { + // Everything failed: Return error code. + VMA_DEBUG_LOG(" vkAllocateMemory FAILED"); + return res; + } + } + } +} + +VkResult VmaAllocator_T::AllocateDedicatedMemory( + VkDeviceSize size, + VmaSuballocationType suballocType, + uint32_t memTypeIndex, + bool map, + bool isUserDataString, + void* pUserData, + VkBuffer /*dedicatedBuffer*/, + VkImage /*dedicatedImage*/, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + VMA_ASSERT(allocationCount > 0 && pAllocations); + + VkMemoryAllocateInfo allocInfo = {}; + allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + allocInfo.memoryTypeIndex = memTypeIndex; + allocInfo.allocationSize = size; + +#if VMA_DEDICATED_ALLOCATION + VkMemoryDedicatedAllocateInfoKHR dedicatedAllocInfo = {}; + dedicatedAllocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO_KHR; + if(m_UseKhrDedicatedAllocation) + { + if(dedicatedBuffer != VK_NULL_HANDLE) + { + VMA_ASSERT(dedicatedImage == VK_NULL_HANDLE); + dedicatedAllocInfo.buffer = dedicatedBuffer; + allocInfo.pNext = &dedicatedAllocInfo; + } + else if(dedicatedImage != VK_NULL_HANDLE) + { + dedicatedAllocInfo.image = dedicatedImage; + allocInfo.pNext = &dedicatedAllocInfo; + } + } +#endif // #if VMA_DEDICATED_ALLOCATION + + size_t allocIndex; + VkResult res = VK_SUCCESS; + for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + res = AllocateDedicatedMemoryPage( + size, + suballocType, + memTypeIndex, + allocInfo, + map, + isUserDataString, + pUserData, + pAllocations + allocIndex); + if(res != VK_SUCCESS) + { + break; + } + } + + if(res == VK_SUCCESS) + { + // Register them in m_pDedicatedAllocations. + { + VmaMutexLockWrite lock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); + AllocationVectorType* pDedicatedAllocations = m_pDedicatedAllocations[memTypeIndex]; + VMA_ASSERT(pDedicatedAllocations); + for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + VmaVectorInsertSorted<VmaPointerLess>(*pDedicatedAllocations, pAllocations[allocIndex]); + } + } + + VMA_DEBUG_LOG(" Allocated DedicatedMemory Count=%zu, MemoryTypeIndex=#%u", allocationCount, memTypeIndex); + } + else + { + // Free all already created allocations. + while(allocIndex--) + { + VmaAllocation currAlloc = pAllocations[allocIndex]; + VkDeviceMemory hMemory = currAlloc->GetMemory(); + + /* + There is no need to call this, because Vulkan spec allows to skip vkUnmapMemory + before vkFreeMemory. + + if(currAlloc->GetMappedData() != VMA_NULL) + { + (*m_VulkanFunctions.vkUnmapMemory)(m_hDevice, hMemory); + } + */ + + FreeVulkanMemory(memTypeIndex, currAlloc->GetSize(), hMemory); + + currAlloc->SetUserData(this, VMA_NULL); + vma_delete(this, currAlloc); + } + + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + } + + return res; +} + +VkResult VmaAllocator_T::AllocateDedicatedMemoryPage( + VkDeviceSize size, + VmaSuballocationType suballocType, + uint32_t memTypeIndex, + const VkMemoryAllocateInfo& allocInfo, + bool map, + bool isUserDataString, + void* pUserData, + VmaAllocation* pAllocation) +{ + VkDeviceMemory hMemory = VK_NULL_HANDLE; + VkResult res = AllocateVulkanMemory(&allocInfo, &hMemory); + if(res < 0) + { + VMA_DEBUG_LOG(" vkAllocateMemory FAILED"); + return res; + } + + void* pMappedData = VMA_NULL; + if(map) + { + res = (*m_VulkanFunctions.vkMapMemory)( + m_hDevice, + hMemory, + 0, + VK_WHOLE_SIZE, + 0, + &pMappedData); + if(res < 0) + { + VMA_DEBUG_LOG(" vkMapMemory FAILED"); + FreeVulkanMemory(memTypeIndex, size, hMemory); + return res; + } + } + + *pAllocation = vma_new(this, VmaAllocation_T)(m_CurrentFrameIndex.load(), isUserDataString); + (*pAllocation)->InitDedicatedAllocation(memTypeIndex, hMemory, suballocType, pMappedData, size); + (*pAllocation)->SetUserData(this, pUserData); + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) + { + FillAllocation(*pAllocation, VMA_ALLOCATION_FILL_PATTERN_CREATED); + } + + return VK_SUCCESS; +} + +void VmaAllocator_T::GetBufferMemoryRequirements( + VkBuffer hBuffer, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const +{ +#if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { + VkBufferMemoryRequirementsInfo2KHR memReqInfo = {}; + memReqInfo.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2_KHR; + memReqInfo.buffer = hBuffer; + + VkMemoryDedicatedRequirementsKHR memDedicatedReq = {}; + memDedicatedReq.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR; + + VkMemoryRequirements2KHR memReq2 = {}; + memReq2.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR; + memReq2.pNext = &memDedicatedReq; + + (*m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); + + memReq = memReq2.memoryRequirements; + requiresDedicatedAllocation = (memDedicatedReq.requiresDedicatedAllocation != VK_FALSE); + prefersDedicatedAllocation = (memDedicatedReq.prefersDedicatedAllocation != VK_FALSE); + } + else +#endif // #if VMA_DEDICATED_ALLOCATION + { + (*m_VulkanFunctions.vkGetBufferMemoryRequirements)(m_hDevice, hBuffer, &memReq); + requiresDedicatedAllocation = false; + prefersDedicatedAllocation = false; + } +} + +void VmaAllocator_T::GetImageMemoryRequirements( + VkImage hImage, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const +{ +#if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { + VkImageMemoryRequirementsInfo2KHR memReqInfo = {}; + memReqInfo.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2_KHR; + memReqInfo.image = hImage; + + VkMemoryDedicatedRequirementsKHR memDedicatedReq = {}; + memDedicatedReq.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR; + + VkMemoryRequirements2KHR memReq2 = {}; + memReq2.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR; + memReq2.pNext = &memDedicatedReq; + + (*m_VulkanFunctions.vkGetImageMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); + + memReq = memReq2.memoryRequirements; + requiresDedicatedAllocation = (memDedicatedReq.requiresDedicatedAllocation != VK_FALSE); + prefersDedicatedAllocation = (memDedicatedReq.prefersDedicatedAllocation != VK_FALSE); + } + else +#endif // #if VMA_DEDICATED_ALLOCATION + { + (*m_VulkanFunctions.vkGetImageMemoryRequirements)(m_hDevice, hImage, &memReq); + requiresDedicatedAllocation = false; + prefersDedicatedAllocation = false; + } +} + +VkResult VmaAllocator_T::AllocateMemory( + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + + VMA_ASSERT(VmaIsPow2(vkMemReq.alignment)); + + if(vkMemReq.size == 0) + { + return VK_ERROR_VALIDATION_FAILED_EXT; + } + if((createInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0 && + (createInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) != 0) + { + VMA_ASSERT(0 && "Specifying VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT together with VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT makes no sense."); + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + if((createInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0 && + (createInfo.flags & VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT) != 0) + { + VMA_ASSERT(0 && "Specifying VMA_ALLOCATION_CREATE_MAPPED_BIT together with VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT is invalid."); + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + if(requiresDedicatedAllocation) + { + if((createInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) != 0) + { + VMA_ASSERT(0 && "VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT specified while dedicated allocation is required."); + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + if(createInfo.pool != VK_NULL_HANDLE) + { + VMA_ASSERT(0 && "Pool specified while dedicated allocation is required."); + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + } + if((createInfo.pool != VK_NULL_HANDLE) && + ((createInfo.flags & (VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT)) != 0)) + { + VMA_ASSERT(0 && "Specifying VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT when pool != null is invalid."); + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + + if(createInfo.pool != VK_NULL_HANDLE) + { + const VkDeviceSize alignmentForPool = VMA_MAX( + vkMemReq.alignment, + GetMemoryTypeMinAlignment(createInfo.pool->m_BlockVector.GetMemoryTypeIndex())); + return createInfo.pool->m_BlockVector.Allocate( + createInfo.pool, + m_CurrentFrameIndex.load(), + vkMemReq.size, + alignmentForPool, + createInfo, + suballocType, + allocationCount, + pAllocations); + } + else + { + // Bit mask of memory Vulkan types acceptable for this allocation. + uint32_t memoryTypeBits = vkMemReq.memoryTypeBits; + uint32_t memTypeIndex = UINT32_MAX; + VkResult res = vmaFindMemoryTypeIndex(this, memoryTypeBits, &createInfo, &memTypeIndex); + if(res == VK_SUCCESS) + { + VkDeviceSize alignmentForMemType = VMA_MAX( + vkMemReq.alignment, + GetMemoryTypeMinAlignment(memTypeIndex)); + + res = AllocateMemoryOfType( + vkMemReq.size, + alignmentForMemType, + requiresDedicatedAllocation || prefersDedicatedAllocation, + dedicatedBuffer, + dedicatedImage, + createInfo, + memTypeIndex, + suballocType, + allocationCount, + pAllocations); + // Succeeded on first try. + if(res == VK_SUCCESS) + { + return res; + } + // Allocation from this memory type failed. Try other compatible memory types. + else + { + for(;;) + { + // Remove old memTypeIndex from list of possibilities. + memoryTypeBits &= ~(1u << memTypeIndex); + // Find alternative memTypeIndex. + res = vmaFindMemoryTypeIndex(this, memoryTypeBits, &createInfo, &memTypeIndex); + if(res == VK_SUCCESS) + { + alignmentForMemType = VMA_MAX( + vkMemReq.alignment, + GetMemoryTypeMinAlignment(memTypeIndex)); + + res = AllocateMemoryOfType( + vkMemReq.size, + alignmentForMemType, + requiresDedicatedAllocation || prefersDedicatedAllocation, + dedicatedBuffer, + dedicatedImage, + createInfo, + memTypeIndex, + suballocType, + allocationCount, + pAllocations); + // Allocation from this alternative memory type succeeded. + if(res == VK_SUCCESS) + { + return res; + } + // else: Allocation from this memory type failed. Try next one - next loop iteration. + } + // No other matching memory type index could be found. + else + { + // Not returning res, which is VK_ERROR_FEATURE_NOT_PRESENT, because we already failed to allocate once. + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + } + } + } + // Can't find any single memory type maching requirements. res is VK_ERROR_FEATURE_NOT_PRESENT. + else + return res; + } +} + +void VmaAllocator_T::FreeMemory( + size_t allocationCount, + const VmaAllocation* pAllocations) +{ + VMA_ASSERT(pAllocations); + + for(size_t allocIndex = allocationCount; allocIndex--; ) + { + VmaAllocation allocation = pAllocations[allocIndex]; + + if(allocation != VK_NULL_HANDLE) + { + if(TouchAllocation(allocation)) + { + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) + { + FillAllocation(allocation, VMA_ALLOCATION_FILL_PATTERN_DESTROYED); + } + + switch(allocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaBlockVector* pBlockVector = VMA_NULL; + VmaPool hPool = allocation->GetPool(); + if(hPool != VK_NULL_HANDLE) + { + pBlockVector = &hPool->m_BlockVector; + } + else + { + const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + pBlockVector = m_pBlockVectors[memTypeIndex]; + } + pBlockVector->Free(allocation); + } + break; + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + FreeDedicatedMemory(allocation); + break; + default: + VMA_ASSERT(0); + } + } + + allocation->SetUserData(this, VMA_NULL); + vma_delete(this, allocation); + } + } +} + +VkResult VmaAllocator_T::ResizeAllocation( + const VmaAllocation alloc, + VkDeviceSize newSize) +{ + if(newSize == 0 || alloc->GetLastUseFrameIndex() == VMA_FRAME_INDEX_LOST) + { + return VK_ERROR_VALIDATION_FAILED_EXT; + } + if(newSize == alloc->GetSize()) + { + return VK_SUCCESS; + } + + switch(alloc->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + return VK_ERROR_FEATURE_NOT_PRESENT; + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + if(alloc->GetBlock()->m_pMetadata->ResizeAllocation(alloc, newSize)) + { + alloc->ChangeSize(newSize); + VMA_HEAVY_ASSERT(alloc->GetBlock()->m_pMetadata->Validate()); + return VK_SUCCESS; + } + else + { + return VkResult(-1000069000); // VK_ERROR_OUT_OF_POOL_MEMORY + } + default: + VMA_ASSERT(0); + return VK_ERROR_VALIDATION_FAILED_EXT; + } +} + +void VmaAllocator_T::CalculateStats(VmaStats* pStats) +{ + // Initialize. + InitStatInfo(pStats->total); + for(size_t i = 0; i < VK_MAX_MEMORY_TYPES; ++i) + InitStatInfo(pStats->memoryType[i]); + for(size_t i = 0; i < VK_MAX_MEMORY_HEAPS; ++i) + InitStatInfo(pStats->memoryHeap[i]); + + // Process default pools. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + VmaBlockVector* const pBlockVector = m_pBlockVectors[memTypeIndex]; + VMA_ASSERT(pBlockVector); + pBlockVector->AddStats(pStats); + } + + // Process custom pools. + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + for(size_t poolIndex = 0, poolCount = m_Pools.size(); poolIndex < poolCount; ++poolIndex) + { + m_Pools[poolIndex]->m_BlockVector.AddStats(pStats); + } + } + + // Process dedicated allocations. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + const uint32_t memHeapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); + VmaMutexLockRead dedicatedAllocationsLock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); + AllocationVectorType* const pDedicatedAllocVector = m_pDedicatedAllocations[memTypeIndex]; + VMA_ASSERT(pDedicatedAllocVector); + for(size_t allocIndex = 0, allocCount = pDedicatedAllocVector->size(); allocIndex < allocCount; ++allocIndex) + { + VmaStatInfo allocationStatInfo; + (*pDedicatedAllocVector)[allocIndex]->DedicatedAllocCalcStatsInfo(allocationStatInfo); + VmaAddStatInfo(pStats->total, allocationStatInfo); + VmaAddStatInfo(pStats->memoryType[memTypeIndex], allocationStatInfo); + VmaAddStatInfo(pStats->memoryHeap[memHeapIndex], allocationStatInfo); + } + } + + // Postprocess. + VmaPostprocessCalcStatInfo(pStats->total); + for(size_t i = 0; i < GetMemoryTypeCount(); ++i) + VmaPostprocessCalcStatInfo(pStats->memoryType[i]); + for(size_t i = 0; i < GetMemoryHeapCount(); ++i) + VmaPostprocessCalcStatInfo(pStats->memoryHeap[i]); +} + +static const uint32_t VMA_VENDOR_ID_AMD = 4098; + +VkResult VmaAllocator_T::DefragmentationBegin( + const VmaDefragmentationInfo2& info, + VmaDefragmentationStats* pStats, + VmaDefragmentationContext* pContext) +{ + if(info.pAllocationsChanged != VMA_NULL) + { + memset(info.pAllocationsChanged, 0, info.allocationCount * sizeof(VkBool32)); + } + + *pContext = vma_new(this, VmaDefragmentationContext_T)( + this, m_CurrentFrameIndex.load(), info.flags, pStats); + + (*pContext)->AddPools(info.poolCount, info.pPools); + (*pContext)->AddAllocations( + info.allocationCount, info.pAllocations, info.pAllocationsChanged); + + VkResult res = (*pContext)->Defragment( + info.maxCpuBytesToMove, info.maxCpuAllocationsToMove, + info.maxGpuBytesToMove, info.maxGpuAllocationsToMove, + info.commandBuffer, pStats); + + if(res != VK_NOT_READY) + { + vma_delete(this, *pContext); + *pContext = VMA_NULL; + } + + return res; +} + +VkResult VmaAllocator_T::DefragmentationEnd( + VmaDefragmentationContext context) +{ + vma_delete(this, context); + return VK_SUCCESS; +} + +void VmaAllocator_T::GetAllocationInfo(VmaAllocation hAllocation, VmaAllocationInfo* pAllocationInfo) +{ + if(hAllocation->CanBecomeLost()) + { + /* + Warning: This is a carefully designed algorithm. + Do not modify unless you really know what you're doing :) + */ + const uint32_t localCurrFrameIndex = m_CurrentFrameIndex.load(); + uint32_t localLastUseFrameIndex = hAllocation->GetLastUseFrameIndex(); + for(;;) + { + if(localLastUseFrameIndex == VMA_FRAME_INDEX_LOST) + { + pAllocationInfo->memoryType = UINT32_MAX; + pAllocationInfo->deviceMemory = VK_NULL_HANDLE; + pAllocationInfo->offset = 0; + pAllocationInfo->size = hAllocation->GetSize(); + pAllocationInfo->pMappedData = VMA_NULL; + pAllocationInfo->pUserData = hAllocation->GetUserData(); + return; + } + else if(localLastUseFrameIndex == localCurrFrameIndex) + { + pAllocationInfo->memoryType = hAllocation->GetMemoryTypeIndex(); + pAllocationInfo->deviceMemory = hAllocation->GetMemory(); + pAllocationInfo->offset = hAllocation->GetOffset(); + pAllocationInfo->size = hAllocation->GetSize(); + pAllocationInfo->pMappedData = VMA_NULL; + pAllocationInfo->pUserData = hAllocation->GetUserData(); + return; + } + else // Last use time earlier than current time. + { + if(hAllocation->CompareExchangeLastUseFrameIndex(localLastUseFrameIndex, localCurrFrameIndex)) + { + localLastUseFrameIndex = localCurrFrameIndex; + } + } + } + } + else + { +#if VMA_STATS_STRING_ENABLED + uint32_t localCurrFrameIndex = m_CurrentFrameIndex.load(); + uint32_t localLastUseFrameIndex = hAllocation->GetLastUseFrameIndex(); + for(;;) + { + VMA_ASSERT(localLastUseFrameIndex != VMA_FRAME_INDEX_LOST); + if(localLastUseFrameIndex == localCurrFrameIndex) + { + break; + } + else // Last use time earlier than current time. + { + if(hAllocation->CompareExchangeLastUseFrameIndex(localLastUseFrameIndex, localCurrFrameIndex)) + { + localLastUseFrameIndex = localCurrFrameIndex; + } + } + } +#endif + + pAllocationInfo->memoryType = hAllocation->GetMemoryTypeIndex(); + pAllocationInfo->deviceMemory = hAllocation->GetMemory(); + pAllocationInfo->offset = hAllocation->GetOffset(); + pAllocationInfo->size = hAllocation->GetSize(); + pAllocationInfo->pMappedData = hAllocation->GetMappedData(); + pAllocationInfo->pUserData = hAllocation->GetUserData(); + } +} + +bool VmaAllocator_T::TouchAllocation(VmaAllocation hAllocation) +{ + // This is a stripped-down version of VmaAllocator_T::GetAllocationInfo. + if(hAllocation->CanBecomeLost()) + { + uint32_t localCurrFrameIndex = m_CurrentFrameIndex.load(); + uint32_t localLastUseFrameIndex = hAllocation->GetLastUseFrameIndex(); + for(;;) + { + if(localLastUseFrameIndex == VMA_FRAME_INDEX_LOST) + { + return false; + } + else if(localLastUseFrameIndex == localCurrFrameIndex) + { + return true; + } + else // Last use time earlier than current time. + { + if(hAllocation->CompareExchangeLastUseFrameIndex(localLastUseFrameIndex, localCurrFrameIndex)) + { + localLastUseFrameIndex = localCurrFrameIndex; + } + } + } + } + else + { +#if VMA_STATS_STRING_ENABLED + uint32_t localCurrFrameIndex = m_CurrentFrameIndex.load(); + uint32_t localLastUseFrameIndex = hAllocation->GetLastUseFrameIndex(); + for(;;) + { + VMA_ASSERT(localLastUseFrameIndex != VMA_FRAME_INDEX_LOST); + if(localLastUseFrameIndex == localCurrFrameIndex) + { + break; + } + else // Last use time earlier than current time. + { + if(hAllocation->CompareExchangeLastUseFrameIndex(localLastUseFrameIndex, localCurrFrameIndex)) + { + localLastUseFrameIndex = localCurrFrameIndex; + } + } + } +#endif + + return true; + } +} + +VkResult VmaAllocator_T::CreatePool(const VmaPoolCreateInfo* pCreateInfo, VmaPool* pPool) +{ + VMA_DEBUG_LOG(" CreatePool: MemoryTypeIndex=%u, flags=%u", pCreateInfo->memoryTypeIndex, pCreateInfo->flags); + + VmaPoolCreateInfo newCreateInfo = *pCreateInfo; + + if(newCreateInfo.maxBlockCount == 0) + { + newCreateInfo.maxBlockCount = SIZE_MAX; + } + if(newCreateInfo.minBlockCount > newCreateInfo.maxBlockCount) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + + const VkDeviceSize preferredBlockSize = CalcPreferredBlockSize(newCreateInfo.memoryTypeIndex); + + *pPool = vma_new(this, VmaPool_T)(this, newCreateInfo, preferredBlockSize); + + VkResult res = (*pPool)->m_BlockVector.CreateMinBlocks(); + if(res != VK_SUCCESS) + { + vma_delete(this, *pPool); + *pPool = VMA_NULL; + return res; + } + + // Add to m_Pools. + { + VmaMutexLockWrite lock(m_PoolsMutex, m_UseMutex); + (*pPool)->SetId(m_NextPoolId++); + VmaVectorInsertSorted<VmaPointerLess>(m_Pools, *pPool); + } + + return VK_SUCCESS; +} + +void VmaAllocator_T::DestroyPool(VmaPool pool) +{ + // Remove from m_Pools. + { + VmaMutexLockWrite lock(m_PoolsMutex, m_UseMutex); + bool success = VmaVectorRemoveSorted<VmaPointerLess>(m_Pools, pool); + (void) success; + VMA_ASSERT(success && "Pool not found in Allocator."); + } + + vma_delete(this, pool); +} + +void VmaAllocator_T::GetPoolStats(VmaPool pool, VmaPoolStats* pPoolStats) +{ + pool->m_BlockVector.GetPoolStats(pPoolStats); +} + +void VmaAllocator_T::SetCurrentFrameIndex(uint32_t frameIndex) +{ + m_CurrentFrameIndex.store(frameIndex); +} + +void VmaAllocator_T::MakePoolAllocationsLost( + VmaPool hPool, + size_t* pLostAllocationCount) +{ + hPool->m_BlockVector.MakePoolAllocationsLost( + m_CurrentFrameIndex.load(), + pLostAllocationCount); +} + +VkResult VmaAllocator_T::CheckPoolCorruption(VmaPool hPool) +{ + return hPool->m_BlockVector.CheckCorruption(); +} + +VkResult VmaAllocator_T::CheckCorruption(uint32_t memoryTypeBits) +{ + VkResult finalRes = VK_ERROR_FEATURE_NOT_PRESENT; + + // Process default pools. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + if(((1u << memTypeIndex) & memoryTypeBits) != 0) + { + VmaBlockVector* const pBlockVector = m_pBlockVectors[memTypeIndex]; + VMA_ASSERT(pBlockVector); + VkResult localRes = pBlockVector->CheckCorruption(); + switch(localRes) + { + case VK_ERROR_FEATURE_NOT_PRESENT: + break; + case VK_SUCCESS: + finalRes = VK_SUCCESS; + break; + default: + return localRes; + } + } + } + + // Process custom pools. + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + for(size_t poolIndex = 0, poolCount = m_Pools.size(); poolIndex < poolCount; ++poolIndex) + { + if(((1u << m_Pools[poolIndex]->m_BlockVector.GetMemoryTypeIndex()) & memoryTypeBits) != 0) + { + VkResult localRes = m_Pools[poolIndex]->m_BlockVector.CheckCorruption(); + switch(localRes) + { + case VK_ERROR_FEATURE_NOT_PRESENT: + break; + case VK_SUCCESS: + finalRes = VK_SUCCESS; + break; + default: + return localRes; + } + } + } + } + + return finalRes; +} + +void VmaAllocator_T::CreateLostAllocation(VmaAllocation* pAllocation) +{ + *pAllocation = vma_new(this, VmaAllocation_T)(VMA_FRAME_INDEX_LOST, false); + (*pAllocation)->InitLost(); +} + +VkResult VmaAllocator_T::AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory) +{ + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(pAllocateInfo->memoryTypeIndex); + + VkResult res; + if(m_HeapSizeLimit[heapIndex] != VK_WHOLE_SIZE) + { + VmaMutexLock lock(m_HeapSizeLimitMutex, m_UseMutex); + if(m_HeapSizeLimit[heapIndex] >= pAllocateInfo->allocationSize) + { + res = (*m_VulkanFunctions.vkAllocateMemory)(m_hDevice, pAllocateInfo, GetAllocationCallbacks(), pMemory); + if(res == VK_SUCCESS) + { + m_HeapSizeLimit[heapIndex] -= pAllocateInfo->allocationSize; + } + } + else + { + res = VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + } + else + { + res = (*m_VulkanFunctions.vkAllocateMemory)(m_hDevice, pAllocateInfo, GetAllocationCallbacks(), pMemory); + } + + if(res == VK_SUCCESS && m_DeviceMemoryCallbacks.pfnAllocate != VMA_NULL) + { + (*m_DeviceMemoryCallbacks.pfnAllocate)(this, pAllocateInfo->memoryTypeIndex, *pMemory, pAllocateInfo->allocationSize); + } + + return res; +} + +void VmaAllocator_T::FreeVulkanMemory(uint32_t memoryType, VkDeviceSize size, VkDeviceMemory hMemory) +{ + if(m_DeviceMemoryCallbacks.pfnFree != VMA_NULL) + { + (*m_DeviceMemoryCallbacks.pfnFree)(this, memoryType, hMemory, size); + } + + (*m_VulkanFunctions.vkFreeMemory)(m_hDevice, hMemory, GetAllocationCallbacks()); + + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memoryType); + if(m_HeapSizeLimit[heapIndex] != VK_WHOLE_SIZE) + { + VmaMutexLock lock(m_HeapSizeLimitMutex, m_UseMutex); + m_HeapSizeLimit[heapIndex] += size; + } +} + +VkResult VmaAllocator_T::Map(VmaAllocation hAllocation, void** ppData) +{ + if(hAllocation->CanBecomeLost()) + { + return VK_ERROR_MEMORY_MAP_FAILED; + } + + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* const pBlock = hAllocation->GetBlock(); + char *pBytes = VMA_NULL; + VkResult res = pBlock->Map(this, 1, (void**)&pBytes); + if(res == VK_SUCCESS) + { + *ppData = pBytes + (ptrdiff_t)hAllocation->GetOffset(); + hAllocation->BlockAllocMap(); + } + return res; + } + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + return hAllocation->DedicatedAllocMap(this, ppData); + default: + VMA_ASSERT(0); + return VK_ERROR_MEMORY_MAP_FAILED; + } +} + +void VmaAllocator_T::Unmap(VmaAllocation hAllocation) +{ + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* const pBlock = hAllocation->GetBlock(); + hAllocation->BlockAllocUnmap(); + pBlock->Unmap(this, 1); + } + break; + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + hAllocation->DedicatedAllocUnmap(this); + break; + default: + VMA_ASSERT(0); + } +} + +VkResult VmaAllocator_T::BindBufferMemory(VmaAllocation hAllocation, VkBuffer hBuffer) +{ + VkResult res = VK_SUCCESS; + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + res = GetVulkanFunctions().vkBindBufferMemory( + m_hDevice, + hBuffer, + hAllocation->GetMemory(), + 0); //memoryOffset + break; + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); + VMA_ASSERT(pBlock && "Binding buffer to allocation that doesn't belong to any block. Is the allocation lost?"); + res = pBlock->BindBufferMemory(this, hAllocation, hBuffer); + break; + } + default: + VMA_ASSERT(0); + } + return res; +} + +VkResult VmaAllocator_T::BindImageMemory(VmaAllocation hAllocation, VkImage hImage) +{ + VkResult res = VK_SUCCESS; + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + res = GetVulkanFunctions().vkBindImageMemory( + m_hDevice, + hImage, + hAllocation->GetMemory(), + 0); //memoryOffset + break; + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); + VMA_ASSERT(pBlock && "Binding image to allocation that doesn't belong to any block. Is the allocation lost?"); + res = pBlock->BindImageMemory(this, hAllocation, hImage); + break; + } + default: + VMA_ASSERT(0); + } + return res; +} + +void VmaAllocator_T::FlushOrInvalidateAllocation( + VmaAllocation hAllocation, + VkDeviceSize offset, VkDeviceSize size, + VMA_CACHE_OPERATION op) +{ + const uint32_t memTypeIndex = hAllocation->GetMemoryTypeIndex(); + if(size > 0 && IsMemoryTypeNonCoherent(memTypeIndex)) + { + const VkDeviceSize allocationSize = hAllocation->GetSize(); + VMA_ASSERT(offset <= allocationSize); + + const VkDeviceSize nonCoherentAtomSize = m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; + + VkMappedMemoryRange memRange = {}; + memRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE; + memRange.memory = hAllocation->GetMemory(); + + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + memRange.offset = VmaAlignDown(offset, nonCoherentAtomSize); + if(size == VK_WHOLE_SIZE) + { + memRange.size = allocationSize - memRange.offset; + } + else + { + VMA_ASSERT(offset + size <= allocationSize); + memRange.size = VMA_MIN( + VmaAlignUp(size + (offset - memRange.offset), nonCoherentAtomSize), + allocationSize - memRange.offset); + } + break; + + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + // 1. Still within this allocation. + memRange.offset = VmaAlignDown(offset, nonCoherentAtomSize); + if(size == VK_WHOLE_SIZE) + { + size = allocationSize - offset; + } + else + { + VMA_ASSERT(offset + size <= allocationSize); + } + memRange.size = VmaAlignUp(size + (offset - memRange.offset), nonCoherentAtomSize); + + // 2. Adjust to whole block. + const VkDeviceSize allocationOffset = hAllocation->GetOffset(); + VMA_ASSERT(allocationOffset % nonCoherentAtomSize == 0); + const VkDeviceSize blockSize = hAllocation->GetBlock()->m_pMetadata->GetSize(); + memRange.offset += allocationOffset; + memRange.size = VMA_MIN(memRange.size, blockSize - memRange.offset); + + break; + } + + default: + VMA_ASSERT(0); + } + + switch(op) + { + case VMA_CACHE_FLUSH: + (*GetVulkanFunctions().vkFlushMappedMemoryRanges)(m_hDevice, 1, &memRange); + break; + case VMA_CACHE_INVALIDATE: + (*GetVulkanFunctions().vkInvalidateMappedMemoryRanges)(m_hDevice, 1, &memRange); + break; + default: + VMA_ASSERT(0); + } + } + // else: Just ignore this call. +} + +void VmaAllocator_T::FreeDedicatedMemory(VmaAllocation allocation) +{ + VMA_ASSERT(allocation && allocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); + + const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + { + VmaMutexLockWrite lock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); + AllocationVectorType* const pDedicatedAllocations = m_pDedicatedAllocations[memTypeIndex]; + VMA_ASSERT(pDedicatedAllocations); + bool success = VmaVectorRemoveSorted<VmaPointerLess>(*pDedicatedAllocations, allocation); + (void) success; + VMA_ASSERT(success); + } + + VkDeviceMemory hMemory = allocation->GetMemory(); + + /* + There is no need to call this, because Vulkan spec allows to skip vkUnmapMemory + before vkFreeMemory. + + if(allocation->GetMappedData() != VMA_NULL) + { + (*m_VulkanFunctions.vkUnmapMemory)(m_hDevice, hMemory); + } + */ + + FreeVulkanMemory(memTypeIndex, allocation->GetSize(), hMemory); + + VMA_DEBUG_LOG(" Freed DedicatedMemory MemoryTypeIndex=%u", memTypeIndex); +} + +void VmaAllocator_T::FillAllocation(const VmaAllocation hAllocation, uint8_t pattern) +{ + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS && + !hAllocation->CanBecomeLost() && + (m_MemProps.memoryTypes[hAllocation->GetMemoryTypeIndex()].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0) + { + void* pData = VMA_NULL; + VkResult res = Map(hAllocation, &pData); + if(res == VK_SUCCESS) + { + memset(pData, (int)pattern, (size_t)hAllocation->GetSize()); + FlushOrInvalidateAllocation(hAllocation, 0, VK_WHOLE_SIZE, VMA_CACHE_FLUSH); + Unmap(hAllocation); + } + else + { + VMA_ASSERT(0 && "VMA_DEBUG_INITIALIZE_ALLOCATIONS is enabled, but couldn't map memory to fill allocation."); + } + } +} + +#if VMA_STATS_STRING_ENABLED + +void VmaAllocator_T::PrintDetailedMap(VmaJsonWriter& json) +{ + bool dedicatedAllocationsStarted = false; + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + VmaMutexLockRead dedicatedAllocationsLock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); + AllocationVectorType* const pDedicatedAllocVector = m_pDedicatedAllocations[memTypeIndex]; + VMA_ASSERT(pDedicatedAllocVector); + if(pDedicatedAllocVector->empty() == false) + { + if(dedicatedAllocationsStarted == false) + { + dedicatedAllocationsStarted = true; + json.WriteString("DedicatedAllocations"); + json.BeginObject(); + } + + json.BeginString("Type "); + json.ContinueString(memTypeIndex); + json.EndString(); + + json.BeginArray(); + + for(size_t i = 0; i < pDedicatedAllocVector->size(); ++i) + { + json.BeginObject(true); + const VmaAllocation hAlloc = (*pDedicatedAllocVector)[i]; + hAlloc->PrintParameters(json); + json.EndObject(); + } + + json.EndArray(); + } + } + if(dedicatedAllocationsStarted) + { + json.EndObject(); + } + + { + bool allocationsStarted = false; + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + if(m_pBlockVectors[memTypeIndex]->IsEmpty() == false) + { + if(allocationsStarted == false) + { + allocationsStarted = true; + json.WriteString("DefaultPools"); + json.BeginObject(); + } + + json.BeginString("Type "); + json.ContinueString(memTypeIndex); + json.EndString(); + + m_pBlockVectors[memTypeIndex]->PrintDetailedMap(json); + } + } + if(allocationsStarted) + { + json.EndObject(); + } + } + + // Custom pools + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + const size_t poolCount = m_Pools.size(); + if(poolCount > 0) + { + json.WriteString("Pools"); + json.BeginObject(); + for(size_t poolIndex = 0; poolIndex < poolCount; ++poolIndex) + { + json.BeginString(); + json.ContinueString(m_Pools[poolIndex]->GetId()); + json.EndString(); + + m_Pools[poolIndex]->m_BlockVector.PrintDetailedMap(json); + } + json.EndObject(); + } + } +} + +#endif // #if VMA_STATS_STRING_ENABLED + +//////////////////////////////////////////////////////////////////////////////// +// Public interface + +VkResult vmaCreateAllocator( + const VmaAllocatorCreateInfo* pCreateInfo, + VmaAllocator* pAllocator) +{ + VMA_ASSERT(pCreateInfo && pAllocator); + VMA_DEBUG_LOG("vmaCreateAllocator"); + *pAllocator = vma_new(pCreateInfo->pAllocationCallbacks, VmaAllocator_T)(pCreateInfo); + return (*pAllocator)->Init(pCreateInfo); +} + +void vmaDestroyAllocator( + VmaAllocator allocator) +{ + if(allocator != VK_NULL_HANDLE) + { + VMA_DEBUG_LOG("vmaDestroyAllocator"); + VkAllocationCallbacks allocationCallbacks = allocator->m_AllocationCallbacks; + vma_delete(&allocationCallbacks, allocator); + } +} + +void vmaGetPhysicalDeviceProperties( + VmaAllocator allocator, + const VkPhysicalDeviceProperties **ppPhysicalDeviceProperties) +{ + VMA_ASSERT(allocator && ppPhysicalDeviceProperties); + *ppPhysicalDeviceProperties = &allocator->m_PhysicalDeviceProperties; +} + +void vmaGetMemoryProperties( + VmaAllocator allocator, + const VkPhysicalDeviceMemoryProperties** ppPhysicalDeviceMemoryProperties) +{ + VMA_ASSERT(allocator && ppPhysicalDeviceMemoryProperties); + *ppPhysicalDeviceMemoryProperties = &allocator->m_MemProps; +} + +void vmaGetMemoryTypeProperties( + VmaAllocator allocator, + uint32_t memoryTypeIndex, + VkMemoryPropertyFlags* pFlags) +{ + VMA_ASSERT(allocator && pFlags); + VMA_ASSERT(memoryTypeIndex < allocator->GetMemoryTypeCount()); + *pFlags = allocator->m_MemProps.memoryTypes[memoryTypeIndex].propertyFlags; +} + +void vmaSetCurrentFrameIndex( + VmaAllocator allocator, + uint32_t frameIndex) +{ + VMA_ASSERT(allocator); + VMA_ASSERT(frameIndex != VMA_FRAME_INDEX_LOST); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->SetCurrentFrameIndex(frameIndex); +} + +void vmaCalculateStats( + VmaAllocator allocator, + VmaStats* pStats) +{ + VMA_ASSERT(allocator && pStats); + VMA_DEBUG_GLOBAL_MUTEX_LOCK + allocator->CalculateStats(pStats); +} + +#if VMA_STATS_STRING_ENABLED + +void vmaBuildStatsString( + VmaAllocator allocator, + char** ppStatsString, + VkBool32 detailedMap) +{ + VMA_ASSERT(allocator && ppStatsString); + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VmaStringBuilder sb(allocator); + { + VmaJsonWriter json(allocator->GetAllocationCallbacks(), sb); + json.BeginObject(); + + VmaStats stats; + allocator->CalculateStats(&stats); + + json.WriteString("Total"); + VmaPrintStatInfo(json, stats.total); + + for(uint32_t heapIndex = 0; heapIndex < allocator->GetMemoryHeapCount(); ++heapIndex) + { + json.BeginString("Heap "); + json.ContinueString(heapIndex); + json.EndString(); + json.BeginObject(); + + json.WriteString("Size"); + json.WriteNumber(allocator->m_MemProps.memoryHeaps[heapIndex].size); + + json.WriteString("Flags"); + json.BeginArray(true); + if((allocator->m_MemProps.memoryHeaps[heapIndex].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) != 0) + { + json.WriteString("DEVICE_LOCAL"); + } + json.EndArray(); + + if(stats.memoryHeap[heapIndex].blockCount > 0) + { + json.WriteString("Stats"); + VmaPrintStatInfo(json, stats.memoryHeap[heapIndex]); + } + + for(uint32_t typeIndex = 0; typeIndex < allocator->GetMemoryTypeCount(); ++typeIndex) + { + if(allocator->MemoryTypeIndexToHeapIndex(typeIndex) == heapIndex) + { + json.BeginString("Type "); + json.ContinueString(typeIndex); + json.EndString(); + + json.BeginObject(); + + json.WriteString("Flags"); + json.BeginArray(true); + VkMemoryPropertyFlags flags = allocator->m_MemProps.memoryTypes[typeIndex].propertyFlags; + if((flags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0) + { + json.WriteString("DEVICE_LOCAL"); + } + if((flags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0) + { + json.WriteString("HOST_VISIBLE"); + } + if((flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0) + { + json.WriteString("HOST_COHERENT"); + } + if((flags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) != 0) + { + json.WriteString("HOST_CACHED"); + } + if((flags & VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT) != 0) + { + json.WriteString("LAZILY_ALLOCATED"); + } + json.EndArray(); + + if(stats.memoryType[typeIndex].blockCount > 0) + { + json.WriteString("Stats"); + VmaPrintStatInfo(json, stats.memoryType[typeIndex]); + } + + json.EndObject(); + } + } + + json.EndObject(); + } + if(detailedMap == VK_TRUE) + { + allocator->PrintDetailedMap(json); + } + + json.EndObject(); + } + + const size_t len = sb.GetLength(); + char* const pChars = vma_new_array(allocator, char, len + 1); + if(len > 0) + { + memcpy(pChars, sb.GetData(), len); + } + pChars[len] = '\0'; + *ppStatsString = pChars; +} + +void vmaFreeStatsString( + VmaAllocator allocator, + char* pStatsString) +{ + if(pStatsString != VMA_NULL) + { + VMA_ASSERT(allocator); + size_t len = strlen(pStatsString); + vma_delete_array(allocator, pStatsString, len + 1); + } +} + +#endif // #if VMA_STATS_STRING_ENABLED + +/* +This function is not protected by any mutex because it just reads immutable data. +*/ +VkResult vmaFindMemoryTypeIndex( + VmaAllocator allocator, + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + if(pAllocationCreateInfo->memoryTypeBits != 0) + { + memoryTypeBits &= pAllocationCreateInfo->memoryTypeBits; + } + + uint32_t requiredFlags = pAllocationCreateInfo->requiredFlags; + uint32_t preferredFlags = pAllocationCreateInfo->preferredFlags; + + const bool mapped = (pAllocationCreateInfo->flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0; + if(mapped) + { + preferredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + } + + // Convert usage to requiredFlags and preferredFlags. + switch(pAllocationCreateInfo->usage) + { + case VMA_MEMORY_USAGE_UNKNOWN: + break; + case VMA_MEMORY_USAGE_GPU_ONLY: + if(!allocator->IsIntegratedGpu() || (preferredFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + preferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + break; + case VMA_MEMORY_USAGE_CPU_ONLY: + requiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; + break; + case VMA_MEMORY_USAGE_CPU_TO_GPU: + requiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + if(!allocator->IsIntegratedGpu() || (preferredFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + preferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + break; + case VMA_MEMORY_USAGE_GPU_TO_CPU: + requiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + preferredFlags |= VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + break; + default: + break; + } + + *pMemoryTypeIndex = UINT32_MAX; + uint32_t minCost = UINT32_MAX; + for(uint32_t memTypeIndex = 0, memTypeBit = 1; + memTypeIndex < allocator->GetMemoryTypeCount(); + ++memTypeIndex, memTypeBit <<= 1) + { + // This memory type is acceptable according to memoryTypeBits bitmask. + if((memTypeBit & memoryTypeBits) != 0) + { + const VkMemoryPropertyFlags currFlags = + allocator->m_MemProps.memoryTypes[memTypeIndex].propertyFlags; + // This memory type contains requiredFlags. + if((requiredFlags & ~currFlags) == 0) + { + // Calculate cost as number of bits from preferredFlags not present in this memory type. + uint32_t currCost = VmaCountBitsSet(preferredFlags & ~currFlags); + // Remember memory type with lowest cost. + if(currCost < minCost) + { + *pMemoryTypeIndex = memTypeIndex; + if(currCost == 0) + { + return VK_SUCCESS; + } + minCost = currCost; + } + } + } + } + return (*pMemoryTypeIndex != UINT32_MAX) ? VK_SUCCESS : VK_ERROR_FEATURE_NOT_PRESENT; +} + +VkResult vmaFindMemoryTypeIndexForBufferInfo( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pBufferCreateInfo != VMA_NULL); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + const VkDevice hDev = allocator->m_hDevice; + VkBuffer hBuffer = VK_NULL_HANDLE; + VkResult res = allocator->GetVulkanFunctions().vkCreateBuffer( + hDev, pBufferCreateInfo, allocator->GetAllocationCallbacks(), &hBuffer); + if(res == VK_SUCCESS) + { + VkMemoryRequirements memReq = {}; + allocator->GetVulkanFunctions().vkGetBufferMemoryRequirements( + hDev, hBuffer, &memReq); + + res = vmaFindMemoryTypeIndex( + allocator, + memReq.memoryTypeBits, + pAllocationCreateInfo, + pMemoryTypeIndex); + + allocator->GetVulkanFunctions().vkDestroyBuffer( + hDev, hBuffer, allocator->GetAllocationCallbacks()); + } + return res; +} + +VkResult vmaFindMemoryTypeIndexForImageInfo( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pImageCreateInfo != VMA_NULL); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + const VkDevice hDev = allocator->m_hDevice; + VkImage hImage = VK_NULL_HANDLE; + VkResult res = allocator->GetVulkanFunctions().vkCreateImage( + hDev, pImageCreateInfo, allocator->GetAllocationCallbacks(), &hImage); + if(res == VK_SUCCESS) + { + VkMemoryRequirements memReq = {}; + allocator->GetVulkanFunctions().vkGetImageMemoryRequirements( + hDev, hImage, &memReq); + + res = vmaFindMemoryTypeIndex( + allocator, + memReq.memoryTypeBits, + pAllocationCreateInfo, + pMemoryTypeIndex); + + allocator->GetVulkanFunctions().vkDestroyImage( + hDev, hImage, allocator->GetAllocationCallbacks()); + } + return res; +} + +VkResult vmaCreatePool( + VmaAllocator allocator, + const VmaPoolCreateInfo* pCreateInfo, + VmaPool* pPool) +{ + VMA_ASSERT(allocator && pCreateInfo && pPool); + + VMA_DEBUG_LOG("vmaCreatePool"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult res = allocator->CreatePool(pCreateInfo, pPool); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordCreatePool(allocator->GetCurrentFrameIndex(), *pCreateInfo, *pPool); + } +#endif + + return res; +} + +void vmaDestroyPool( + VmaAllocator allocator, + VmaPool pool) +{ + VMA_ASSERT(allocator); + + if(pool == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyPool"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordDestroyPool(allocator->GetCurrentFrameIndex(), pool); + } +#endif + + allocator->DestroyPool(pool); +} + +void vmaGetPoolStats( + VmaAllocator allocator, + VmaPool pool, + VmaPoolStats* pPoolStats) +{ + VMA_ASSERT(allocator && pool && pPoolStats); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->GetPoolStats(pool, pPoolStats); +} + +void vmaMakePoolAllocationsLost( + VmaAllocator allocator, + VmaPool pool, + size_t* pLostAllocationCount) +{ + VMA_ASSERT(allocator && pool); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordMakePoolAllocationsLost(allocator->GetCurrentFrameIndex(), pool); + } +#endif + + allocator->MakePoolAllocationsLost(pool, pLostAllocationCount); +} + +VkResult vmaCheckPoolCorruption(VmaAllocator allocator, VmaPool pool) +{ + VMA_ASSERT(allocator && pool); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VMA_DEBUG_LOG("vmaCheckPoolCorruption"); + + return allocator->CheckPoolCorruption(pool); +} + +VkResult vmaAllocateMemory( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pVkMemoryRequirements && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult result = allocator->AllocateMemory( + *pVkMemoryRequirements, + false, // requiresDedicatedAllocation + false, // prefersDedicatedAllocation + VK_NULL_HANDLE, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_UNKNOWN, + 1, // allocationCount + pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordAllocateMemory( + allocator->GetCurrentFrameIndex(), + *pVkMemoryRequirements, + *pCreateInfo, + *pAllocation); + } +#endif + + if(pAllocationInfo != VMA_NULL && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +VkResult vmaAllocateMemoryPages( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + size_t allocationCount, + VmaAllocation* pAllocations, + VmaAllocationInfo* pAllocationInfo) +{ + if(allocationCount == 0) + { + return VK_SUCCESS; + } + + VMA_ASSERT(allocator && pVkMemoryRequirements && pCreateInfo && pAllocations); + + VMA_DEBUG_LOG("vmaAllocateMemoryPages"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult result = allocator->AllocateMemory( + *pVkMemoryRequirements, + false, // requiresDedicatedAllocation + false, // prefersDedicatedAllocation + VK_NULL_HANDLE, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_UNKNOWN, + allocationCount, + pAllocations); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordAllocateMemoryPages( + allocator->GetCurrentFrameIndex(), + *pVkMemoryRequirements, + *pCreateInfo, + (uint64_t)allocationCount, + pAllocations); + } +#endif + + if(pAllocationInfo != VMA_NULL && result == VK_SUCCESS) + { + for(size_t i = 0; i < allocationCount; ++i) + { + allocator->GetAllocationInfo(pAllocations[i], pAllocationInfo + i); + } + } + + return result; +} + +VkResult vmaAllocateMemoryForBuffer( + VmaAllocator allocator, + VkBuffer buffer, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && buffer != VK_NULL_HANDLE && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemoryForBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetBufferMemoryRequirements(buffer, vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation); + + VkResult result = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + buffer, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_BUFFER, + 1, // allocationCount + pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordAllocateMemoryForBuffer( + allocator->GetCurrentFrameIndex(), + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + *pCreateInfo, + *pAllocation); + } +#endif + + if(pAllocationInfo && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +VkResult vmaAllocateMemoryForImage( + VmaAllocator allocator, + VkImage image, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && image != VK_NULL_HANDLE && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemoryForImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetImageMemoryRequirements(image, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + VkResult result = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + VK_NULL_HANDLE, // dedicatedBuffer + image, // dedicatedImage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN, + 1, // allocationCount + pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordAllocateMemoryForImage( + allocator->GetCurrentFrameIndex(), + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + *pCreateInfo, + *pAllocation); + } +#endif + + if(pAllocationInfo && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +void vmaFreeMemory( + VmaAllocator allocator, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator); + + if(allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaFreeMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordFreeMemory( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + allocator->FreeMemory( + 1, // allocationCount + &allocation); +} + +void vmaFreeMemoryPages( + VmaAllocator allocator, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + if(allocationCount == 0) + { + return; + } + + VMA_ASSERT(allocator); + + VMA_DEBUG_LOG("vmaFreeMemoryPages"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordFreeMemoryPages( + allocator->GetCurrentFrameIndex(), + (uint64_t)allocationCount, + pAllocations); + } +#endif + + allocator->FreeMemory(allocationCount, pAllocations); +} + +VkResult vmaResizeAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize newSize) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_LOG("vmaResizeAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordResizeAllocation( + allocator->GetCurrentFrameIndex(), + allocation, + newSize); + } +#endif + + return allocator->ResizeAllocation(allocation, newSize); +} + +void vmaGetAllocationInfo( + VmaAllocator allocator, + VmaAllocation allocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && allocation && pAllocationInfo); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordGetAllocationInfo( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + allocator->GetAllocationInfo(allocation, pAllocationInfo); +} + +VkBool32 vmaTouchAllocation( + VmaAllocator allocator, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordTouchAllocation( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + return allocator->TouchAllocation(allocation); +} + +void vmaSetAllocationUserData( + VmaAllocator allocator, + VmaAllocation allocation, + void* pUserData) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocation->SetUserData(allocator, pUserData); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordSetAllocationUserData( + allocator->GetCurrentFrameIndex(), + allocation, + pUserData); + } +#endif +} + +void vmaCreateLostAllocation( + VmaAllocator allocator, + VmaAllocation* pAllocation) +{ + VMA_ASSERT(allocator && pAllocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + + allocator->CreateLostAllocation(pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordCreateLostAllocation( + allocator->GetCurrentFrameIndex(), + *pAllocation); + } +#endif +} + +VkResult vmaMapMemory( + VmaAllocator allocator, + VmaAllocation allocation, + void** ppData) +{ + VMA_ASSERT(allocator && allocation && ppData); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult res = allocator->Map(allocation, ppData); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordMapMemory( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + return res; +} + +void vmaUnmapMemory( + VmaAllocator allocator, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordUnmapMemory( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + allocator->Unmap(allocation); +} + +void vmaFlushAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_LOG("vmaFlushAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->FlushOrInvalidateAllocation(allocation, offset, size, VMA_CACHE_FLUSH); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordFlushAllocation( + allocator->GetCurrentFrameIndex(), + allocation, offset, size); + } +#endif +} + +void vmaInvalidateAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_LOG("vmaInvalidateAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->FlushOrInvalidateAllocation(allocation, offset, size, VMA_CACHE_INVALIDATE); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordInvalidateAllocation( + allocator->GetCurrentFrameIndex(), + allocation, offset, size); + } +#endif +} + +VkResult vmaCheckCorruption(VmaAllocator allocator, uint32_t memoryTypeBits) +{ + VMA_ASSERT(allocator); + + VMA_DEBUG_LOG("vmaCheckCorruption"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->CheckCorruption(memoryTypeBits); +} + +VkResult vmaDefragment( + VmaAllocator allocator, + VmaAllocation* pAllocations, + size_t allocationCount, + VkBool32* pAllocationsChanged, + const VmaDefragmentationInfo *pDefragmentationInfo, + VmaDefragmentationStats* pDefragmentationStats) +{ + // Deprecated interface, reimplemented using new one. + + VmaDefragmentationInfo2 info2 = {}; + info2.allocationCount = (uint32_t)allocationCount; + info2.pAllocations = pAllocations; + info2.pAllocationsChanged = pAllocationsChanged; + if(pDefragmentationInfo != VMA_NULL) + { + info2.maxCpuAllocationsToMove = pDefragmentationInfo->maxAllocationsToMove; + info2.maxCpuBytesToMove = pDefragmentationInfo->maxBytesToMove; + } + else + { + info2.maxCpuAllocationsToMove = UINT32_MAX; + info2.maxCpuBytesToMove = VK_WHOLE_SIZE; + } + // info2.flags, maxGpuAllocationsToMove, maxGpuBytesToMove, commandBuffer deliberately left zero. + + VmaDefragmentationContext ctx; + VkResult res = vmaDefragmentationBegin(allocator, &info2, pDefragmentationStats, &ctx); + if(res == VK_NOT_READY) + { + res = vmaDefragmentationEnd( allocator, ctx); + } + return res; +} + +VkResult vmaDefragmentationBegin( + VmaAllocator allocator, + const VmaDefragmentationInfo2* pInfo, + VmaDefragmentationStats* pStats, + VmaDefragmentationContext *pContext) +{ + VMA_ASSERT(allocator && pInfo && pContext); + + // Degenerate case: Nothing to defragment. + if(pInfo->allocationCount == 0 && pInfo->poolCount == 0) + { + return VK_SUCCESS; + } + + VMA_ASSERT(pInfo->allocationCount == 0 || pInfo->pAllocations != VMA_NULL); + VMA_ASSERT(pInfo->poolCount == 0 || pInfo->pPools != VMA_NULL); + VMA_HEAVY_ASSERT(VmaValidatePointerArray(pInfo->allocationCount, pInfo->pAllocations)); + VMA_HEAVY_ASSERT(VmaValidatePointerArray(pInfo->poolCount, pInfo->pPools)); + + VMA_DEBUG_LOG("vmaDefragmentationBegin"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult res = allocator->DefragmentationBegin(*pInfo, pStats, pContext); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordDefragmentationBegin( + allocator->GetCurrentFrameIndex(), *pInfo, *pContext); + } +#endif + + return res; +} + +VkResult vmaDefragmentationEnd( + VmaAllocator allocator, + VmaDefragmentationContext context) +{ + VMA_ASSERT(allocator); + + VMA_DEBUG_LOG("vmaDefragmentationEnd"); + + if(context != VK_NULL_HANDLE) + { + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordDefragmentationEnd( + allocator->GetCurrentFrameIndex(), context); + } +#endif + + return allocator->DefragmentationEnd(context); + } + else + { + return VK_SUCCESS; + } +} + +VkResult vmaBindBufferMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkBuffer buffer) +{ + VMA_ASSERT(allocator && allocation && buffer); + + VMA_DEBUG_LOG("vmaBindBufferMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindBufferMemory(allocation, buffer); +} + +VkResult vmaBindImageMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkImage image) +{ + VMA_ASSERT(allocator && allocation && image); + + VMA_DEBUG_LOG("vmaBindImageMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindImageMemory(allocation, image); +} + +VkResult vmaCreateBuffer( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkBuffer* pBuffer, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pBufferCreateInfo && pAllocationCreateInfo && pBuffer && pAllocation); + + if(pBufferCreateInfo->size == 0) + { + return VK_ERROR_VALIDATION_FAILED_EXT; + } + + VMA_DEBUG_LOG("vmaCreateBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *pBuffer = VK_NULL_HANDLE; + *pAllocation = VK_NULL_HANDLE; + + // 1. Create VkBuffer. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateBuffer)( + allocator->m_hDevice, + pBufferCreateInfo, + allocator->GetAllocationCallbacks(), + pBuffer); + if(res >= 0) + { + // 2. vkGetBufferMemoryRequirements. + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetBufferMemoryRequirements(*pBuffer, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + // Make sure alignment requirements for specific buffer usages reported + // in Physical Device Properties are included in alignment reported by memory requirements. + if((pBufferCreateInfo->usage & VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT) != 0) + { + VMA_ASSERT(vkMemReq.alignment % + allocator->m_PhysicalDeviceProperties.limits.minTexelBufferOffsetAlignment == 0); + } + if((pBufferCreateInfo->usage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT) != 0) + { + VMA_ASSERT(vkMemReq.alignment % + allocator->m_PhysicalDeviceProperties.limits.minUniformBufferOffsetAlignment == 0); + } + if((pBufferCreateInfo->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) != 0) + { + VMA_ASSERT(vkMemReq.alignment % + allocator->m_PhysicalDeviceProperties.limits.minStorageBufferOffsetAlignment == 0); + } + + // 3. Allocate memory using allocator. + res = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + *pBuffer, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + *pAllocationCreateInfo, + VMA_SUBALLOCATION_TYPE_BUFFER, + 1, // allocationCount + pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordCreateBuffer( + allocator->GetCurrentFrameIndex(), + *pBufferCreateInfo, + *pAllocationCreateInfo, + *pAllocation); + } +#endif + + if(res >= 0) + { + // 3. Bind buffer with memory. + res = allocator->BindBufferMemory(*pAllocation, *pBuffer); + if(res >= 0) + { + // All steps succeeded. + #if VMA_STATS_STRING_ENABLED + (*pAllocation)->InitBufferImageUsage(pBufferCreateInfo->usage); + #endif + if(pAllocationInfo != VMA_NULL) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return VK_SUCCESS; + } + allocator->FreeMemory( + 1, // allocationCount + pAllocation); + *pAllocation = VK_NULL_HANDLE; + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + return res; +} + +void vmaDestroyBuffer( + VmaAllocator allocator, + VkBuffer buffer, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator); + + if(buffer == VK_NULL_HANDLE && allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordDestroyBuffer( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + if(buffer != VK_NULL_HANDLE) + { + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, buffer, allocator->GetAllocationCallbacks()); + } + + if(allocation != VK_NULL_HANDLE) + { + allocator->FreeMemory( + 1, // allocationCount + &allocation); + } +} + +VkResult vmaCreateImage( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkImage* pImage, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pImageCreateInfo && pAllocationCreateInfo && pImage && pAllocation); + + if(pImageCreateInfo->extent.width == 0 || + pImageCreateInfo->extent.height == 0 || + pImageCreateInfo->extent.depth == 0 || + pImageCreateInfo->mipLevels == 0 || + pImageCreateInfo->arrayLayers == 0) + { + return VK_ERROR_VALIDATION_FAILED_EXT; + } + + VMA_DEBUG_LOG("vmaCreateImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *pImage = VK_NULL_HANDLE; + *pAllocation = VK_NULL_HANDLE; + + // 1. Create VkImage. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateImage)( + allocator->m_hDevice, + pImageCreateInfo, + allocator->GetAllocationCallbacks(), + pImage); + if(res >= 0) + { + VmaSuballocationType suballocType = pImageCreateInfo->tiling == VK_IMAGE_TILING_OPTIMAL ? + VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL : + VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR; + + // 2. Allocate memory using allocator. + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetImageMemoryRequirements(*pImage, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + res = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + VK_NULL_HANDLE, // dedicatedBuffer + *pImage, // dedicatedImage + *pAllocationCreateInfo, + suballocType, + 1, // allocationCount + pAllocation); + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordCreateImage( + allocator->GetCurrentFrameIndex(), + *pImageCreateInfo, + *pAllocationCreateInfo, + *pAllocation); + } +#endif + + if(res >= 0) + { + // 3. Bind image with memory. + res = allocator->BindImageMemory(*pAllocation, *pImage); + if(res >= 0) + { + // All steps succeeded. + #if VMA_STATS_STRING_ENABLED + (*pAllocation)->InitBufferImageUsage(pImageCreateInfo->usage); + #endif + if(pAllocationInfo != VMA_NULL) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return VK_SUCCESS; + } + allocator->FreeMemory( + 1, // allocationCount + pAllocation); + *pAllocation = VK_NULL_HANDLE; + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, *pImage, allocator->GetAllocationCallbacks()); + *pImage = VK_NULL_HANDLE; + return res; + } + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, *pImage, allocator->GetAllocationCallbacks()); + *pImage = VK_NULL_HANDLE; + return res; + } + return res; +} + +void vmaDestroyImage( + VmaAllocator allocator, + VkImage image, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator); + + if(image == VK_NULL_HANDLE && allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + +#if VMA_RECORDING_ENABLED + if(allocator->GetRecorder() != VMA_NULL) + { + allocator->GetRecorder()->RecordDestroyImage( + allocator->GetCurrentFrameIndex(), + allocation); + } +#endif + + if(image != VK_NULL_HANDLE) + { + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, image, allocator->GetAllocationCallbacks()); + } + if(allocation != VK_NULL_HANDLE) + { + allocator->FreeMemory( + 1, // allocationCount + &allocation); + } +} + +#endif // #ifdef VMA_IMPLEMENTATION diff --git a/src/3rdparty/double-conversion/double-conversion.cc b/src/3rdparty/double-conversion/double-conversion.cc index ecd1a5ef3f..881ca0adbc 100644 --- a/src/3rdparty/double-conversion/double-conversion.cc +++ b/src/3rdparty/double-conversion/double-conversion.cc @@ -38,6 +38,11 @@ #include <double-conversion/strtod.h> #include <double-conversion/utils.h> +// Fix warning C4244: 'argument': conversion from 'const uc16' to 'char', possible loss of data +#ifdef _MSC_VER + __pragma(warning(disable: 4244)) +#endif + namespace double_conversion { const DoubleToStringConverter& DoubleToStringConverter::EcmaScriptConverter() { diff --git a/src/3rdparty/forkfd/forkfd.c b/src/3rdparty/forkfd/forkfd.c index bef109e401..12537b6199 100644 --- a/src/3rdparty/forkfd/forkfd.c +++ b/src/3rdparty/forkfd/forkfd.c @@ -267,7 +267,7 @@ static int tryReaping(pid_t pid, struct pipe_payload *payload) static void freeInfo(Header *header, ProcessInfo *entry) { entry->deathPipe = -1; - entry->pid = 0; + ffd_atomic_store(&entry->pid, 0, FFD_ATOMIC_RELEASE); (void)ffd_atomic_add_fetch(&header->busyCount, -1, FFD_ATOMIC_RELEASE); assert(header->busyCount >= 0); @@ -519,9 +519,9 @@ static void cleanup() ffd_atomic_store(&forkfd_status, 0, FFD_ATOMIC_RELAXED); /* free any arrays we might have */ - array = children.header.nextArray; + array = ffd_atomic_load(&children.header.nextArray, FFD_ATOMIC_ACQUIRE); while (array != NULL) { - BigArray *next = array->header.nextArray; + BigArray *next = ffd_atomic_load(&array->header.nextArray, FFD_ATOMIC_ACQUIRE); free(array); array = next; } diff --git a/src/3rdparty/harfbuzz/src/harfbuzz-indic.cpp b/src/3rdparty/harfbuzz/src/harfbuzz-indic.cpp index 704ea9774a..de3bcb2bbf 100644 --- a/src/3rdparty/harfbuzz/src/harfbuzz-indic.cpp +++ b/src/3rdparty/harfbuzz/src/harfbuzz-indic.cpp @@ -1471,7 +1471,7 @@ static bool indic_shape_syllable(HB_Bool openType, HB_ShaperItem *item, bool inv while (finalOrder[toMove].form && fixed < len-1) { IDEBUG(" fixed = %d, toMove=%d, moving form %d with pos %d", fixed, toMove, finalOrder[toMove].form, finalOrder[toMove].position); for (i = fixed; i < len; i++) { -// IDEBUG() << " i=" << i << "uc=" << hex << uc[i] << "form=" << form(uc[i]) +// IDEBUG() << " i=" << i << "uc=" << Qt::hex << uc[i] << "form=" << form(uc[i]) // << "position=" << position[i]; if (form(uc[i]) == finalOrder[toMove].form && position[i] == finalOrder[toMove].position) { diff --git a/src/3rdparty/harfbuzz/tests/shaping/main.cpp b/src/3rdparty/harfbuzz/tests/shaping/main.cpp index 10818c565c..16f469029b 100644 --- a/src/3rdparty/harfbuzz/tests/shaping/main.cpp +++ b/src/3rdparty/harfbuzz/tests/shaping/main.cpp @@ -370,7 +370,7 @@ void tst_QScriptEngine::greek() QString str; str.append(uc); if (str.normalized(QString::NormalizationForm_D).normalized(QString::NormalizationForm_C) != str) { - //qDebug() << "skipping" << hex << uc; + //qDebug() << "skipping" << Qt::hex << uc; continue; } if (uc == 0x1fc1 || uc == 0x1fed) @@ -389,7 +389,7 @@ void tst_QScriptEngine::greek() QString str; str.append(uc); if (str.normalized(QString::NormalizationForm_D).normalized(QString::NormalizationForm_C) != str) { - //qDebug() << "skipping" << hex << uc; + //qDebug() << "skipping" << Qt::hex << uc; continue; } if (uc == 0x1fc1 || uc == 0x1fed) diff --git a/src/3rdparty/md4c.pri b/src/3rdparty/md4c.pri new file mode 100644 index 0000000000..e0150dc6ed --- /dev/null +++ b/src/3rdparty/md4c.pri @@ -0,0 +1,3 @@ +INCLUDEPATH += $$PWD/md4c +HEADERS += $$PWD/md4c/md4c.h +SOURCES += $$PWD/md4c/md4c.c diff --git a/src/3rdparty/md4c/LICENSE.md b/src/3rdparty/md4c/LICENSE.md new file mode 100644 index 0000000000..d58ef9341d --- /dev/null +++ b/src/3rdparty/md4c/LICENSE.md @@ -0,0 +1,22 @@ + +# The MIT License (MIT) + +Copyright © 2016-2019 Martin Mitáš + +Permission is hereby granted, free of charge, to any person obtaining a +copy of this software and associated documentation files (the “Software”), +to deal in the Software without restriction, including without limitation +the rights to use, copy, modify, merge, publish, distribute, sublicense, +and/or sell copies of the Software, and to permit persons to whom the +Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included +in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS +OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +IN THE SOFTWARE. diff --git a/src/3rdparty/md4c/md4c.c b/src/3rdparty/md4c/md4c.c new file mode 100644 index 0000000000..01e63a5fd2 --- /dev/null +++ b/src/3rdparty/md4c/md4c.c @@ -0,0 +1,6109 @@ +/* + * MD4C: Markdown parser for C + * (http://github.com/mity/md4c) + * + * Copyright (c) 2016-2019 Martin Mitas + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + */ + +#include "md4c.h" + +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + + +/***************************** + *** Miscellaneous Stuff *** + *****************************/ + +#ifdef _MSC_VER + /* MSVC does not understand "inline" when building as pure C (not C++). + * However it understands "__inline" */ + #ifndef __cplusplus + #define inline __inline + #endif +#endif + +#ifdef _T + #undef _T +#endif +#if defined MD4C_USE_UTF16 + #define _T(x) L##x +#else + #define _T(x) x +#endif + +/* Misc. macros. */ +#define SIZEOF_ARRAY(a) (sizeof(a) / sizeof(a[0])) + +#define STRINGIZE_(x) #x +#define STRINGIZE(x) STRINGIZE_(x) + +#ifndef TRUE + #define TRUE 1 + #define FALSE 0 +#endif + + +/************************ + *** Internal Types *** + ************************/ + +/* These are omnipresent so lets save some typing. */ +#define CHAR MD_CHAR +#define SZ MD_SIZE +#define OFF MD_OFFSET + +typedef struct MD_MARK_tag MD_MARK; +typedef struct MD_BLOCK_tag MD_BLOCK; +typedef struct MD_CONTAINER_tag MD_CONTAINER; +typedef struct MD_REF_DEF_tag MD_REF_DEF; + + +/* During analyzes of inline marks, we need to manage some "mark chains", + * of (yet unresolved) openers. This structure holds start/end of the chain. + * The chain internals are then realized through MD_MARK::prev and ::next. + */ +typedef struct MD_MARKCHAIN_tag MD_MARKCHAIN; +struct MD_MARKCHAIN_tag { + int head; /* Index of first mark in the chain, or -1 if empty. */ + int tail; /* Index of last mark in the chain, or -1 if empty. */ +}; + +/* Context propagated through all the parsing. */ +typedef struct MD_CTX_tag MD_CTX; +struct MD_CTX_tag { + /* Immutable stuff (parameters of md_parse()). */ + const CHAR* text; + SZ size; + MD_PARSER parser; + void* userdata; + + /* When this is true, it allows some optimizations. */ + int doc_ends_with_newline; + + /* Helper temporary growing buffer. */ + CHAR* buffer; + unsigned alloc_buffer; + + /* Reference definitions. */ + MD_REF_DEF* ref_defs; + int n_ref_defs; + int alloc_ref_defs; + void** ref_def_hashtable; + int ref_def_hashtable_size; + + /* Stack of inline/span markers. + * This is only used for parsing a single block contents but by storing it + * here we may reuse the stack for subsequent blocks; i.e. we have fewer + * (re)allocations. */ + MD_MARK* marks; + int n_marks; + int alloc_marks; + +#if defined MD4C_USE_UTF16 + char mark_char_map[128]; +#else + char mark_char_map[256]; +#endif + + /* For resolving of inline spans. */ + MD_MARKCHAIN mark_chains[11]; +#define PTR_CHAIN ctx->mark_chains[0] +#define TABLECELLBOUNDARIES ctx->mark_chains[1] +#define ASTERISK_OPENERS_extraword_mod3_0 ctx->mark_chains[2] +#define ASTERISK_OPENERS_extraword_mod3_1 ctx->mark_chains[3] +#define ASTERISK_OPENERS_extraword_mod3_2 ctx->mark_chains[4] +#define ASTERISK_OPENERS_intraword_mod3_0 ctx->mark_chains[5] +#define ASTERISK_OPENERS_intraword_mod3_1 ctx->mark_chains[6] +#define ASTERISK_OPENERS_intraword_mod3_2 ctx->mark_chains[7] +#define UNDERSCORE_OPENERS ctx->mark_chains[8] +#define TILDE_OPENERS ctx->mark_chains[9] +#define BRACKET_OPENERS ctx->mark_chains[10] +#define OPENERS_CHAIN_FIRST 2 +#define OPENERS_CHAIN_LAST 10 + + int n_table_cell_boundaries; + + /* For resolving links. */ + int unresolved_link_head; + int unresolved_link_tail; + + /* For resolving raw HTML. */ + OFF html_comment_horizon; + OFF html_proc_instr_horizon; + OFF html_decl_horizon; + OFF html_cdata_horizon; + + /* For block analysis. + * Notes: + * -- It holds MD_BLOCK as well as MD_LINE structures. After each + * MD_BLOCK, its (multiple) MD_LINE(s) follow. + * -- For MD_BLOCK_HTML and MD_BLOCK_CODE, MD_VERBATIMLINE(s) are used + * instead of MD_LINE(s). + */ + void* block_bytes; + MD_BLOCK* current_block; + int n_block_bytes; + int alloc_block_bytes; + + /* For container block analysis. */ + MD_CONTAINER* containers; + int n_containers; + int alloc_containers; + + /* Minimal indentation to call the block "indented code block". */ + unsigned code_indent_offset; + + /* Contextual info for line analysis. */ + SZ code_fence_length; /* For checking closing fence length. */ + int html_block_type; /* For checking closing raw HTML condition. */ + int last_line_has_list_loosening_effect; + int last_list_item_starts_with_two_blank_lines; +}; + +enum MD_LINETYPE_tag { + MD_LINE_BLANK, + MD_LINE_HR, + MD_LINE_ATXHEADER, + MD_LINE_SETEXTHEADER, + MD_LINE_SETEXTUNDERLINE, + MD_LINE_INDENTEDCODE, + MD_LINE_FENCEDCODE, + MD_LINE_HTML, + MD_LINE_TEXT, + MD_LINE_TABLE, + MD_LINE_TABLEUNDERLINE +}; +typedef enum MD_LINETYPE_tag MD_LINETYPE; + +typedef struct MD_LINE_ANALYSIS_tag MD_LINE_ANALYSIS; +struct MD_LINE_ANALYSIS_tag { + MD_LINETYPE type : 16; + unsigned data : 16; + OFF beg; + OFF end; + unsigned indent; /* Indentation level. */ +}; + +typedef struct MD_LINE_tag MD_LINE; +struct MD_LINE_tag { + OFF beg; + OFF end; +}; + +typedef struct MD_VERBATIMLINE_tag MD_VERBATIMLINE; +struct MD_VERBATIMLINE_tag { + OFF beg; + OFF end; + OFF indent; +}; + + +/******************* + *** Debugging *** + *******************/ + +#define MD_LOG(msg) \ + do { \ + if(ctx->parser.debug_log != NULL) \ + ctx->parser.debug_log((msg), ctx->userdata); \ + } while(0) + +#ifdef DEBUG + #define MD_ASSERT(cond) \ + do { \ + if(!(cond)) { \ + MD_LOG(__FILE__ ":" STRINGIZE(__LINE__) ": " \ + "Assertion '" STRINGIZE(cond) "' failed."); \ + exit(1); \ + } \ + } while(0) + + #define MD_UNREACHABLE() MD_ASSERT(1 == 0) +#else + #ifdef __GNUC__ + #define MD_ASSERT(cond) do { if(!(cond)) __builtin_unreachable(); } while(0) + #define MD_UNREACHABLE() do { __builtin_unreachable(); } while(0) + #elif defined _MSC_VER && _MSC_VER > 120 + #define MD_ASSERT(cond) do { __assume(cond); } while(0) + #define MD_UNREACHABLE() do { __assume(0); } while(0) + #else + #define MD_ASSERT(cond) do {} while(0) + #define MD_UNREACHABLE() do {} while(0) + #endif +#endif + + +/***************** + *** Helpers *** + *****************/ + +/* Character accessors. */ +#define CH(off) (ctx->text[(off)]) +#define STR(off) (ctx->text + (off)) + +/* Character classification. + * Note we assume ASCII compatibility of code points < 128 here. */ +#define ISIN_(ch, ch_min, ch_max) ((ch_min) <= (unsigned)(ch) && (unsigned)(ch) <= (ch_max)) +#define ISANYOF_(ch, palette) (md_strchr((palette), (ch)) != NULL) +#define ISANYOF2_(ch, ch1, ch2) ((ch) == (ch1) || (ch) == (ch2)) +#define ISANYOF3_(ch, ch1, ch2, ch3) ((ch) == (ch1) || (ch) == (ch2) || (ch) == (ch3)) +#define ISASCII_(ch) ((unsigned)(ch) <= 127) +#define ISBLANK_(ch) (ISANYOF2_((ch), _T(' '), _T('\t'))) +#define ISNEWLINE_(ch) (ISANYOF2_((ch), _T('\r'), _T('\n'))) +#define ISWHITESPACE_(ch) (ISBLANK_(ch) || ISANYOF2_((ch), _T('\v'), _T('\f'))) +#define ISCNTRL_(ch) ((unsigned)(ch) <= 31 || (unsigned)(ch) == 127) +#define ISPUNCT_(ch) (ISIN_(ch, 33, 47) || ISIN_(ch, 58, 64) || ISIN_(ch, 91, 96) || ISIN_(ch, 123, 126)) +#define ISUPPER_(ch) (ISIN_(ch, _T('A'), _T('Z'))) +#define ISLOWER_(ch) (ISIN_(ch, _T('a'), _T('z'))) +#define ISALPHA_(ch) (ISUPPER_(ch) || ISLOWER_(ch)) +#define ISDIGIT_(ch) (ISIN_(ch, _T('0'), _T('9'))) +#define ISXDIGIT_(ch) (ISDIGIT_(ch) || ISIN_(ch, _T('A'), _T('F')) || ISIN_(ch, _T('a'), _T('f'))) +#define ISALNUM_(ch) (ISALPHA_(ch) || ISDIGIT_(ch)) + +#define ISANYOF(off, palette) ISANYOF_(CH(off), (palette)) +#define ISANYOF2(off, ch1, ch2) ISANYOF2_(CH(off), (ch1), (ch2)) +#define ISANYOF3(off, ch1, ch2, ch3) ISANYOF3_(CH(off), (ch1), (ch2), (ch3)) +#define ISASCII(off) ISASCII_(CH(off)) +#define ISBLANK(off) ISBLANK_(CH(off)) +#define ISNEWLINE(off) ISNEWLINE_(CH(off)) +#define ISWHITESPACE(off) ISWHITESPACE_(CH(off)) +#define ISCNTRL(off) ISCNTRL_(CH(off)) +#define ISPUNCT(off) ISPUNCT_(CH(off)) +#define ISUPPER(off) ISUPPER_(CH(off)) +#define ISLOWER(off) ISLOWER_(CH(off)) +#define ISALPHA(off) ISALPHA_(CH(off)) +#define ISDIGIT(off) ISDIGIT_(CH(off)) +#define ISXDIGIT(off) ISXDIGIT_(CH(off)) +#define ISALNUM(off) ISALNUM_(CH(off)) +static inline const CHAR* +md_strchr(const CHAR* str, CHAR ch) +{ + OFF i; + for(i = 0; str[i] != _T('\0'); i++) { + if(ch == str[i]) + return (str + i); + } + return NULL; +} + +/* Case insensitive check of string equality. */ +static inline int +md_ascii_case_eq(const CHAR* s1, const CHAR* s2, SZ n) +{ + OFF i; + for(i = 0; i < n; i++) { + CHAR ch1 = s1[i]; + CHAR ch2 = s2[i]; + + if(ISLOWER_(ch1)) + ch1 += ('A'-'a'); + if(ISLOWER_(ch2)) + ch2 += ('A'-'a'); + if(ch1 != ch2) + return FALSE; + } + return TRUE; +} + +static inline int +md_ascii_eq(const CHAR* s1, const CHAR* s2, SZ n) +{ + return memcmp(s1, s2, n * sizeof(CHAR)) == 0; +} + +static int +md_text_with_null_replacement(MD_CTX* ctx, MD_TEXTTYPE type, const CHAR* str, SZ size) +{ + OFF off = 0; + int ret = 0; + + while(1) { + while(off < size && str[off] != _T('\0')) + off++; + + if(off > 0) { + ret = ctx->parser.text(type, str, off, ctx->userdata); + if(ret != 0) + return ret; + + str += off; + size -= off; + off = 0; + } + + if(off >= size) + return 0; + + ret = ctx->parser.text(MD_TEXT_NULLCHAR, _T(""), 1, ctx->userdata); + if(ret != 0) + return ret; + off++; + } +} + + +#define MD_CHECK(func) \ + do { \ + ret = (func); \ + if(ret < 0) \ + goto abort; \ + } while(0) + + +#define MD_TEMP_BUFFER(sz) \ + do { \ + if(sz > ctx->alloc_buffer) { \ + CHAR* new_buffer; \ + SZ new_size = ((sz) + (sz) / 2 + 128) & ~127; \ + \ + new_buffer = realloc(ctx->buffer, new_size); \ + if(new_buffer == NULL) { \ + MD_LOG("realloc() failed."); \ + ret = -1; \ + goto abort; \ + } \ + \ + ctx->buffer = new_buffer; \ + ctx->alloc_buffer = new_size; \ + } \ + } while(0) + + +#define MD_ENTER_BLOCK(type, arg) \ + do { \ + ret = ctx->parser.enter_block((type), (arg), ctx->userdata); \ + if(ret != 0) { \ + MD_LOG("Aborted from enter_block() callback."); \ + goto abort; \ + } \ + } while(0) + +#define MD_LEAVE_BLOCK(type, arg) \ + do { \ + ret = ctx->parser.leave_block((type), (arg), ctx->userdata); \ + if(ret != 0) { \ + MD_LOG("Aborted from leave_block() callback."); \ + goto abort; \ + } \ + } while(0) + +#define MD_ENTER_SPAN(type, arg) \ + do { \ + ret = ctx->parser.enter_span((type), (arg), ctx->userdata); \ + if(ret != 0) { \ + MD_LOG("Aborted from enter_span() callback."); \ + goto abort; \ + } \ + } while(0) + +#define MD_LEAVE_SPAN(type, arg) \ + do { \ + ret = ctx->parser.leave_span((type), (arg), ctx->userdata); \ + if(ret != 0) { \ + MD_LOG("Aborted from leave_span() callback."); \ + goto abort; \ + } \ + } while(0) + +#define MD_TEXT(type, str, size) \ + do { \ + if(size > 0) { \ + ret = ctx->parser.text((type), (str), (size), ctx->userdata); \ + if(ret != 0) { \ + MD_LOG("Aborted from text() callback."); \ + goto abort; \ + } \ + } \ + } while(0) + +#define MD_TEXT_INSECURE(type, str, size) \ + do { \ + if(size > 0) { \ + ret = md_text_with_null_replacement(ctx, type, str, size); \ + if(ret != 0) { \ + MD_LOG("Aborted from text() callback."); \ + goto abort; \ + } \ + } \ + } while(0) + + + +/************************* + *** Unicode Support *** + *************************/ + +typedef struct MD_UNICODE_FOLD_INFO_tag MD_UNICODE_FOLD_INFO; +struct MD_UNICODE_FOLD_INFO_tag { + unsigned codepoints[3]; + int n_codepoints; +}; + + +#if defined MD4C_USE_UTF16 || defined MD4C_USE_UTF8 + /* Binary search over sorted "map" of codepoints. Consecutive sequences + * of codepoints may be encoded in the map by just using the + * (MIN_CODEPOINT | 0x40000000) and (MAX_CODEPOINT | 0x80000000). + * + * Returns index of the found record in the map (in the case of ranges, + * the minimal value is used); or -1 on failure. */ + static int + md_unicode_bsearch__(unsigned codepoint, const unsigned* map, size_t map_size) + { + int beg, end; + int pivot_beg, pivot_end; + + beg = 0; + end = (int) map_size-1; + while(beg <= end) { + /* Pivot may be a range, not just a single value. */ + pivot_beg = pivot_end = (beg + end) / 2; + if(map[pivot_end] & 0x40000000) + pivot_end++; + if(map[pivot_beg] & 0x80000000) + pivot_beg--; + + if(codepoint < (map[pivot_beg] & 0x00ffffff)) + end = pivot_beg - 1; + else if(codepoint > (map[pivot_end] & 0x00ffffff)) + beg = pivot_end + 1; + else + return pivot_beg; + } + + return -1; + } + + static int + md_is_unicode_whitespace__(unsigned codepoint) + { +#define R(cp_min, cp_max) ((cp_min) | 0x40000000), ((cp_max) | 0x80000000) +#define S(cp) (cp) + /* Unicode "Zs" category. + * (generated by scripts/build_whitespace_map.py) */ + static const unsigned WHITESPACE_MAP[] = { + S(0x0020), S(0x00a0), S(0x1680), R(0x2000,0x200a), S(0x202f), S(0x205f), S(0x3000) + }; +#undef R +#undef S + + /* The ASCII ones are the most frequently used ones, also CommonMark + * specification requests few more in this range. */ + if(codepoint <= 0x7f) + return ISWHITESPACE_(codepoint); + + return (md_unicode_bsearch__(codepoint, WHITESPACE_MAP, SIZEOF_ARRAY(WHITESPACE_MAP)) >= 0); + } + + static int + md_is_unicode_punct__(unsigned codepoint) + { +#define R(cp_min, cp_max) ((cp_min) | 0x40000000), ((cp_max) | 0x80000000) +#define S(cp) (cp) + /* Unicode "Pc", "Pd", "Pe", "Pf", "Pi", "Po", "Ps" categories. + * (generated by scripts/build_punct_map.py) */ + static const unsigned PUNCT_MAP[] = { + R(0x0021,0x0023), R(0x0025,0x002a), R(0x002c,0x002f), R(0x003a,0x003b), R(0x003f,0x0040), + R(0x005b,0x005d), S(0x005f), S(0x007b), S(0x007d), S(0x00a1), S(0x00a7), S(0x00ab), R(0x00b6,0x00b7), + S(0x00bb), S(0x00bf), S(0x037e), S(0x0387), R(0x055a,0x055f), R(0x0589,0x058a), S(0x05be), S(0x05c0), + S(0x05c3), S(0x05c6), R(0x05f3,0x05f4), R(0x0609,0x060a), R(0x060c,0x060d), S(0x061b), R(0x061e,0x061f), + R(0x066a,0x066d), S(0x06d4), R(0x0700,0x070d), R(0x07f7,0x07f9), R(0x0830,0x083e), S(0x085e), + R(0x0964,0x0965), S(0x0970), S(0x09fd), S(0x0a76), S(0x0af0), S(0x0c77), S(0x0c84), S(0x0df4), S(0x0e4f), + R(0x0e5a,0x0e5b), R(0x0f04,0x0f12), S(0x0f14), R(0x0f3a,0x0f3d), S(0x0f85), R(0x0fd0,0x0fd4), + R(0x0fd9,0x0fda), R(0x104a,0x104f), S(0x10fb), R(0x1360,0x1368), S(0x1400), S(0x166e), R(0x169b,0x169c), + R(0x16eb,0x16ed), R(0x1735,0x1736), R(0x17d4,0x17d6), R(0x17d8,0x17da), R(0x1800,0x180a), + R(0x1944,0x1945), R(0x1a1e,0x1a1f), R(0x1aa0,0x1aa6), R(0x1aa8,0x1aad), R(0x1b5a,0x1b60), + R(0x1bfc,0x1bff), R(0x1c3b,0x1c3f), R(0x1c7e,0x1c7f), R(0x1cc0,0x1cc7), S(0x1cd3), R(0x2010,0x2027), + R(0x2030,0x2043), R(0x2045,0x2051), R(0x2053,0x205e), R(0x207d,0x207e), R(0x208d,0x208e), + R(0x2308,0x230b), R(0x2329,0x232a), R(0x2768,0x2775), R(0x27c5,0x27c6), R(0x27e6,0x27ef), + R(0x2983,0x2998), R(0x29d8,0x29db), R(0x29fc,0x29fd), R(0x2cf9,0x2cfc), R(0x2cfe,0x2cff), S(0x2d70), + R(0x2e00,0x2e2e), R(0x2e30,0x2e4f), R(0x3001,0x3003), R(0x3008,0x3011), R(0x3014,0x301f), S(0x3030), + S(0x303d), S(0x30a0), S(0x30fb), R(0xa4fe,0xa4ff), R(0xa60d,0xa60f), S(0xa673), S(0xa67e), + R(0xa6f2,0xa6f7), R(0xa874,0xa877), R(0xa8ce,0xa8cf), R(0xa8f8,0xa8fa), S(0xa8fc), R(0xa92e,0xa92f), + S(0xa95f), R(0xa9c1,0xa9cd), R(0xa9de,0xa9df), R(0xaa5c,0xaa5f), R(0xaade,0xaadf), R(0xaaf0,0xaaf1), + S(0xabeb), R(0xfd3e,0xfd3f), R(0xfe10,0xfe19), R(0xfe30,0xfe52), R(0xfe54,0xfe61), S(0xfe63), S(0xfe68), + R(0xfe6a,0xfe6b), R(0xff01,0xff03), R(0xff05,0xff0a), R(0xff0c,0xff0f), R(0xff1a,0xff1b), + R(0xff1f,0xff20), R(0xff3b,0xff3d), S(0xff3f), S(0xff5b), S(0xff5d), R(0xff5f,0xff65), R(0x10100,0x10102), + S(0x1039f), S(0x103d0), S(0x1056f), S(0x10857), S(0x1091f), S(0x1093f), R(0x10a50,0x10a58), S(0x10a7f), + R(0x10af0,0x10af6), R(0x10b39,0x10b3f), R(0x10b99,0x10b9c), R(0x10f55,0x10f59), R(0x11047,0x1104d), + R(0x110bb,0x110bc), R(0x110be,0x110c1), R(0x11140,0x11143), R(0x11174,0x11175), R(0x111c5,0x111c8), + S(0x111cd), S(0x111db), R(0x111dd,0x111df), R(0x11238,0x1123d), S(0x112a9), R(0x1144b,0x1144f), + S(0x1145b), S(0x1145d), S(0x114c6), R(0x115c1,0x115d7), R(0x11641,0x11643), R(0x11660,0x1166c), + R(0x1173c,0x1173e), S(0x1183b), S(0x119e2), R(0x11a3f,0x11a46), R(0x11a9a,0x11a9c), R(0x11a9e,0x11aa2), + R(0x11c41,0x11c45), R(0x11c70,0x11c71), R(0x11ef7,0x11ef8), S(0x11fff), R(0x12470,0x12474), + R(0x16a6e,0x16a6f), S(0x16af5), R(0x16b37,0x16b3b), S(0x16b44), R(0x16e97,0x16e9a), S(0x16fe2), + S(0x1bc9f), R(0x1da87,0x1da8b), R(0x1e95e,0x1e95f) + }; +#undef R +#undef S + + /* The ASCII ones are the most frequently used ones, also CommonMark + * specification requests few more in this range. */ + if(codepoint <= 0x7f) + return ISPUNCT_(codepoint); + + return (md_unicode_bsearch__(codepoint, PUNCT_MAP, SIZEOF_ARRAY(PUNCT_MAP)) >= 0); + } + + static void + md_get_unicode_fold_info(unsigned codepoint, MD_UNICODE_FOLD_INFO* info) + { +#define R(cp_min, cp_max) ((cp_min) | 0x40000000), ((cp_max) | 0x80000000) +#define S(cp) (cp) + /* Unicode "Pc", "Pd", "Pe", "Pf", "Pi", "Po", "Ps" categories. + * (generated by scripts/build_punct_map.py) */ + static const unsigned FOLD_MAP_1[] = { + R(0x0041,0x005a), S(0x00b5), R(0x00c0,0x00d6), R(0x00d8,0x00de), R(0x0100,0x012e), R(0x0132,0x0136), + R(0x0139,0x0147), R(0x014a,0x0176), S(0x0178), R(0x0179,0x017d), S(0x017f), S(0x0181), S(0x0182), + S(0x0186), S(0x0187), S(0x0189), S(0x018b), S(0x018e), S(0x018f), S(0x0190), S(0x0191), S(0x0193), + S(0x0194), S(0x0196), S(0x0197), S(0x0198), S(0x019c), S(0x019d), S(0x019f), R(0x01a0,0x01a4), S(0x01a6), + S(0x01a7), S(0x01a9), S(0x01ac), S(0x01ae), S(0x01af), S(0x01b1), S(0x01b3), S(0x01b7), S(0x01b8), + S(0x01bc), S(0x01c4), S(0x01c5), S(0x01c7), S(0x01c8), S(0x01ca), R(0x01cb,0x01db), R(0x01de,0x01ee), + S(0x01f1), S(0x01f2), S(0x01f6), S(0x01f7), R(0x01f8,0x021e), S(0x0220), R(0x0222,0x0232), S(0x023a), + S(0x023b), S(0x023d), S(0x023e), S(0x0241), S(0x0243), S(0x0244), S(0x0245), R(0x0246,0x024e), S(0x0345), + S(0x0370), S(0x0376), S(0x037f), S(0x0386), R(0x0388,0x038a), S(0x038c), S(0x038e), R(0x0391,0x03a1), + R(0x03a3,0x03ab), S(0x03c2), S(0x03cf), S(0x03d0), S(0x03d1), S(0x03d5), S(0x03d6), R(0x03d8,0x03ee), + S(0x03f0), S(0x03f1), S(0x03f4), S(0x03f5), S(0x03f7), S(0x03f9), S(0x03fa), R(0x03fd,0x03ff), + R(0x0400,0x040f), R(0x0410,0x042f), R(0x0460,0x0480), R(0x048a,0x04be), S(0x04c0), R(0x04c1,0x04cd), + R(0x04d0,0x052e), R(0x0531,0x0556), R(0x10a0,0x10c5), S(0x10c7), S(0x10cd), R(0x13f8,0x13fd), S(0x1c80), + S(0x1c81), S(0x1c82), S(0x1c83), S(0x1c85), S(0x1c86), S(0x1c87), S(0x1c88), R(0x1c90,0x1cba), + R(0x1cbd,0x1cbf), R(0x1e00,0x1e94), S(0x1e9b), R(0x1ea0,0x1efe), R(0x1f08,0x1f0f), R(0x1f18,0x1f1d), + R(0x1f28,0x1f2f), R(0x1f38,0x1f3f), R(0x1f48,0x1f4d), S(0x1f59), S(0x1f5b), S(0x1f5d), S(0x1f5f), + R(0x1f68,0x1f6f), S(0x1fb8), S(0x1fba), S(0x1fbe), R(0x1fc8,0x1fcb), S(0x1fd8), S(0x1fda), S(0x1fe8), + S(0x1fea), S(0x1fec), S(0x1ff8), S(0x1ffa), S(0x2126), S(0x212a), S(0x212b), S(0x2132), R(0x2160,0x216f), + S(0x2183), R(0x24b6,0x24cf), R(0x2c00,0x2c2e), S(0x2c60), S(0x2c62), S(0x2c63), S(0x2c64), + R(0x2c67,0x2c6b), S(0x2c6d), S(0x2c6e), S(0x2c6f), S(0x2c70), S(0x2c72), S(0x2c75), S(0x2c7e), + R(0x2c80,0x2ce2), S(0x2ceb), S(0x2cf2), R(0xa640,0xa66c), R(0xa680,0xa69a), R(0xa722,0xa72e), + R(0xa732,0xa76e), S(0xa779), S(0xa77d), R(0xa77e,0xa786), S(0xa78b), S(0xa78d), S(0xa790), + R(0xa796,0xa7a8), S(0xa7aa), S(0xa7ab), S(0xa7ac), S(0xa7ad), S(0xa7ae), S(0xa7b0), S(0xa7b1), S(0xa7b2), + S(0xa7b3), R(0xa7b4,0xa7be), S(0xa7c2), S(0xa7c4), S(0xa7c5), S(0xa7c6), R(0xab70,0xabbf), + R(0xff21,0xff3a), R(0x10400,0x10427), R(0x104b0,0x104d3), R(0x10c80,0x10cb2), R(0x118a0,0x118bf), + R(0x16e40,0x16e5f), R(0x1e900,0x1e921) + }; + static const unsigned FOLD_MAP_1_DATA[] = { + 0x0061, 0x007a, 0x03bc, 0x00e0, 0x00f6, 0x00f8, 0x00fe, 0x0101, 0x012f, 0x0133, 0x0137, 0x013a, 0x0148, + 0x014b, 0x0177, 0x00ff, 0x017a, 0x017e, 0x0073, 0x0253, 0x0183, 0x0254, 0x0188, 0x0256, 0x018c, 0x01dd, + 0x0259, 0x025b, 0x0192, 0x0260, 0x0263, 0x0269, 0x0268, 0x0199, 0x026f, 0x0272, 0x0275, 0x01a1, 0x01a5, + 0x0280, 0x01a8, 0x0283, 0x01ad, 0x0288, 0x01b0, 0x028a, 0x01b4, 0x0292, 0x01b9, 0x01bd, 0x01c6, 0x01c6, + 0x01c9, 0x01c9, 0x01cc, 0x01cc, 0x01dc, 0x01df, 0x01ef, 0x01f3, 0x01f3, 0x0195, 0x01bf, 0x01f9, 0x021f, + 0x019e, 0x0223, 0x0233, 0x2c65, 0x023c, 0x019a, 0x2c66, 0x0242, 0x0180, 0x0289, 0x028c, 0x0247, 0x024f, + 0x03b9, 0x0371, 0x0377, 0x03f3, 0x03ac, 0x03ad, 0x03af, 0x03cc, 0x03cd, 0x03b1, 0x03c1, 0x03c3, 0x03cb, + 0x03c3, 0x03d7, 0x03b2, 0x03b8, 0x03c6, 0x03c0, 0x03d9, 0x03ef, 0x03ba, 0x03c1, 0x03b8, 0x03b5, 0x03f8, + 0x03f2, 0x03fb, 0x037b, 0x037d, 0x0450, 0x045f, 0x0430, 0x044f, 0x0461, 0x0481, 0x048b, 0x04bf, 0x04cf, + 0x04c2, 0x04ce, 0x04d1, 0x052f, 0x0561, 0x0586, 0x2d00, 0x2d25, 0x2d27, 0x2d2d, 0x13f0, 0x13f5, 0x0432, + 0x0434, 0x043e, 0x0441, 0x0442, 0x044a, 0x0463, 0xa64b, 0x10d0, 0x10fa, 0x10fd, 0x10ff, 0x1e01, 0x1e95, + 0x1e61, 0x1ea1, 0x1eff, 0x1f00, 0x1f07, 0x1f10, 0x1f15, 0x1f20, 0x1f27, 0x1f30, 0x1f37, 0x1f40, 0x1f45, + 0x1f51, 0x1f53, 0x1f55, 0x1f57, 0x1f60, 0x1f67, 0x1fb0, 0x1f70, 0x03b9, 0x1f72, 0x1f75, 0x1fd0, 0x1f76, + 0x1fe0, 0x1f7a, 0x1fe5, 0x1f78, 0x1f7c, 0x03c9, 0x006b, 0x00e5, 0x214e, 0x2170, 0x217f, 0x2184, 0x24d0, + 0x24e9, 0x2c30, 0x2c5e, 0x2c61, 0x026b, 0x1d7d, 0x027d, 0x2c68, 0x2c6c, 0x0251, 0x0271, 0x0250, 0x0252, + 0x2c73, 0x2c76, 0x023f, 0x2c81, 0x2ce3, 0x2cec, 0x2cf3, 0xa641, 0xa66d, 0xa681, 0xa69b, 0xa723, 0xa72f, + 0xa733, 0xa76f, 0xa77a, 0x1d79, 0xa77f, 0xa787, 0xa78c, 0x0265, 0xa791, 0xa797, 0xa7a9, 0x0266, 0x025c, + 0x0261, 0x026c, 0x026a, 0x029e, 0x0287, 0x029d, 0xab53, 0xa7b5, 0xa7bf, 0xa7c3, 0xa794, 0x0282, 0x1d8e, + 0x13a0, 0x13ef, 0xff41, 0xff5a, 0x10428, 0x1044f, 0x104d8, 0x104fb, 0x10cc0, 0x10cf2, 0x118c0, 0x118df, + 0x16e60, 0x16e7f, 0x1e922, 0x1e943 + }; + static const unsigned FOLD_MAP_2[] = { + S(0x00df), S(0x0130), S(0x0149), S(0x01f0), S(0x0587), S(0x1e96), S(0x1e97), S(0x1e98), S(0x1e99), + S(0x1e9a), S(0x1e9e), S(0x1f50), R(0x1f80,0x1f87), R(0x1f88,0x1f8f), R(0x1f90,0x1f97), R(0x1f98,0x1f9f), + R(0x1fa0,0x1fa7), R(0x1fa8,0x1faf), S(0x1fb2), S(0x1fb3), S(0x1fb4), S(0x1fb6), S(0x1fbc), S(0x1fc2), + S(0x1fc3), S(0x1fc4), S(0x1fc6), S(0x1fcc), S(0x1fd6), S(0x1fe4), S(0x1fe6), S(0x1ff2), S(0x1ff3), + S(0x1ff4), S(0x1ff6), S(0x1ffc), S(0xfb00), S(0xfb01), S(0xfb02), S(0xfb05), S(0xfb06), S(0xfb13), + S(0xfb14), S(0xfb15), S(0xfb16), S(0xfb17) + }; + static const unsigned FOLD_MAP_2_DATA[] = { + 0x0073,0x0073, 0x0069,0x0307, 0x02bc,0x006e, 0x006a,0x030c, 0x0565,0x0582, 0x0068,0x0331, 0x0074,0x0308, + 0x0077,0x030a, 0x0079,0x030a, 0x0061,0x02be, 0x0073,0x0073, 0x03c5,0x0313, 0x1f00,0x03b9, 0x1f07,0x03b9, + 0x1f00,0x03b9, 0x1f07,0x03b9, 0x1f20,0x03b9, 0x1f27,0x03b9, 0x1f20,0x03b9, 0x1f27,0x03b9, 0x1f60,0x03b9, + 0x1f67,0x03b9, 0x1f60,0x03b9, 0x1f67,0x03b9, 0x1f70,0x03b9, 0x03b1,0x03b9, 0x03ac,0x03b9, 0x03b1,0x0342, + 0x03b1,0x03b9, 0x1f74,0x03b9, 0x03b7,0x03b9, 0x03ae,0x03b9, 0x03b7,0x0342, 0x03b7,0x03b9, 0x03b9,0x0342, + 0x03c1,0x0313, 0x03c5,0x0342, 0x1f7c,0x03b9, 0x03c9,0x03b9, 0x03ce,0x03b9, 0x03c9,0x0342, 0x03c9,0x03b9, + 0x0066,0x0066, 0x0066,0x0069, 0x0066,0x006c, 0x0073,0x0074, 0x0073,0x0074, 0x0574,0x0576, 0x0574,0x0565, + 0x0574,0x056b, 0x057e,0x0576, 0x0574,0x056d + }; + static const unsigned FOLD_MAP_3[] = { + S(0x0390), S(0x03b0), S(0x1f52), S(0x1f54), S(0x1f56), S(0x1fb7), S(0x1fc7), S(0x1fd2), S(0x1fd3), + S(0x1fd7), S(0x1fe2), S(0x1fe3), S(0x1fe7), S(0x1ff7), S(0xfb03), S(0xfb04) + }; + static const unsigned FOLD_MAP_3_DATA[] = { + 0x03b9,0x0308,0x0301, 0x03c5,0x0308,0x0301, 0x03c5,0x0313,0x0300, 0x03c5,0x0313,0x0301, + 0x03c5,0x0313,0x0342, 0x03b1,0x0342,0x03b9, 0x03b7,0x0342,0x03b9, 0x03b9,0x0308,0x0300, + 0x03b9,0x0308,0x0301, 0x03b9,0x0308,0x0342, 0x03c5,0x0308,0x0300, 0x03c5,0x0308,0x0301, + 0x03c5,0x0308,0x0342, 0x03c9,0x0342,0x03b9, 0x0066,0x0066,0x0069, 0x0066,0x0066,0x006c + }; +#undef R +#undef S + static const struct { + const unsigned* map; + const unsigned* data; + size_t map_size; + int n_codepoints; + } FOLD_MAP_LIST[] = { + { FOLD_MAP_1, FOLD_MAP_1_DATA, SIZEOF_ARRAY(FOLD_MAP_1), 1 }, + { FOLD_MAP_2, FOLD_MAP_2_DATA, SIZEOF_ARRAY(FOLD_MAP_2), 2 }, + { FOLD_MAP_3, FOLD_MAP_3_DATA, SIZEOF_ARRAY(FOLD_MAP_3), 3 } + }; + + int i; + + /* Fast path for ASCII characters. */ + if(codepoint <= 0x7f) { + info->codepoints[0] = codepoint; + if(ISUPPER_(codepoint)) + info->codepoints[0] += 'a' - 'A'; + info->n_codepoints = 1; + return; + } + + /* Try to locate the codepoint in any of the maps. */ + for(i = 0; i < (int) SIZEOF_ARRAY(FOLD_MAP_LIST); i++) { + int index; + + index = md_unicode_bsearch__(codepoint, FOLD_MAP_LIST[i].map, FOLD_MAP_LIST[i].map_size); + if(index >= 0) { + /* Found the mapping. */ + int n_codepoints = FOLD_MAP_LIST[i].n_codepoints; + const unsigned* map = FOLD_MAP_LIST[i].map; + const unsigned* codepoints = FOLD_MAP_LIST[i].data + (index * n_codepoints); + + memcpy(info->codepoints, codepoints, sizeof(unsigned) * n_codepoints); + info->n_codepoints = n_codepoints; + + if(FOLD_MAP_LIST[i].map[index] != codepoint) { + /* The found mapping maps whole range of codepoints, + * i.e. we have to offset info->codepoints[0] accordingly. */ + if((map[index] & 0x00ffffff)+1 == codepoints[0]) { + /* Alternating type of the range. */ + info->codepoints[0] = codepoint + ((codepoint & 0x1) == (map[index] & 0x1) ? 1 : 0); + } else { + /* Range to range kind of mapping. */ + info->codepoints[0] += (codepoint - (map[index] & 0x00ffffff)); + } + } + + return; + } + } + + /* No mapping found. Map the codepoint to itself. */ + info->codepoints[0] = codepoint; + info->n_codepoints = 1; + } +#endif + + +#if defined MD4C_USE_UTF16 + #define IS_UTF16_SURROGATE_HI(word) (((WORD)(word) & 0xfc00) == 0xd800) + #define IS_UTF16_SURROGATE_LO(word) (((WORD)(word) & 0xfc00) == 0xdc00) + #define UTF16_DECODE_SURROGATE(hi, lo) (0x10000 + ((((unsigned)(hi) & 0x3ff) << 10) | (((unsigned)(lo) & 0x3ff) << 0))) + + static unsigned + md_decode_utf16le__(const CHAR* str, SZ str_size, SZ* p_size) + { + if(IS_UTF16_SURROGATE_HI(str[0])) { + if(1 < str_size && IS_UTF16_SURROGATE_LO(str[1])) { + if(p_size != NULL) + *p_size = 2; + return UTF16_DECODE_SURROGATE(str[0], str[1]); + } + } + + if(p_size != NULL) + *p_size = 1; + return str[0]; + } + + static unsigned + md_decode_utf16le_before__(MD_CTX* ctx, OFF off) + { + if(off > 2 && IS_UTF16_SURROGATE_HI(CH(off-2)) && IS_UTF16_SURROGATE_LO(CH(off-1))) + return UTF16_DECODE_SURROGATE(CH(off-2), CH(off-1)); + + return CH(off); + } + + /* No whitespace uses surrogates, so no decoding needed here. */ + #define ISUNICODEWHITESPACE_(codepoint) md_is_unicode_whitespace__(codepoint) + #define ISUNICODEWHITESPACE(off) md_is_unicode_whitespace__(CH(off)) + #define ISUNICODEWHITESPACEBEFORE(off) md_is_unicode_whitespace__(CH((off)-1)) + + #define ISUNICODEPUNCT(off) md_is_unicode_punct__(md_decode_utf16le__(STR(off), ctx->size - (off), NULL)) + #define ISUNICODEPUNCTBEFORE(off) md_is_unicode_punct__(md_decode_utf16le_before__(ctx, off)) + + static inline int + md_decode_unicode(const CHAR* str, OFF off, SZ str_size, SZ* p_char_size) + { + return md_decode_utf16le__(str+off, str_size-off, p_char_size); + } +#elif defined MD4C_USE_UTF8 + #define IS_UTF8_LEAD1(byte) ((unsigned char)(byte) <= 0x7f) + #define IS_UTF8_LEAD2(byte) (((unsigned char)(byte) & 0xe0) == 0xc0) + #define IS_UTF8_LEAD3(byte) (((unsigned char)(byte) & 0xf0) == 0xe0) + #define IS_UTF8_LEAD4(byte) (((unsigned char)(byte) & 0xf8) == 0xf0) + #define IS_UTF8_TAIL(byte) (((unsigned char)(byte) & 0xc0) == 0x80) + + static unsigned + md_decode_utf8__(const CHAR* str, SZ str_size, SZ* p_size) + { + if(!IS_UTF8_LEAD1(str[0])) { + if(IS_UTF8_LEAD2(str[0])) { + if(1 < str_size && IS_UTF8_TAIL(str[1])) { + if(p_size != NULL) + *p_size = 2; + + return (((unsigned int)str[0] & 0x1f) << 6) | + (((unsigned int)str[1] & 0x3f) << 0); + } + } else if(IS_UTF8_LEAD3(str[0])) { + if(2 < str_size && IS_UTF8_TAIL(str[1]) && IS_UTF8_TAIL(str[2])) { + if(p_size != NULL) + *p_size = 3; + + return (((unsigned int)str[0] & 0x0f) << 12) | + (((unsigned int)str[1] & 0x3f) << 6) | + (((unsigned int)str[2] & 0x3f) << 0); + } + } else if(IS_UTF8_LEAD4(str[0])) { + if(3 < str_size && IS_UTF8_TAIL(str[1]) && IS_UTF8_TAIL(str[2]) && IS_UTF8_TAIL(str[3])) { + if(p_size != NULL) + *p_size = 4; + + return (((unsigned int)str[0] & 0x07) << 18) | + (((unsigned int)str[1] & 0x3f) << 12) | + (((unsigned int)str[2] & 0x3f) << 6) | + (((unsigned int)str[3] & 0x3f) << 0); + } + } + } + + if(p_size != NULL) + *p_size = 1; + return (unsigned) str[0]; + } + + static unsigned + md_decode_utf8_before__(MD_CTX* ctx, OFF off) + { + if(!IS_UTF8_LEAD1(CH(off-1))) { + if(off > 1 && IS_UTF8_LEAD2(CH(off-2)) && IS_UTF8_TAIL(CH(off-1))) + return (((unsigned int)CH(off-2) & 0x1f) << 6) | + (((unsigned int)CH(off-1) & 0x3f) << 0); + + if(off > 2 && IS_UTF8_LEAD3(CH(off-3)) && IS_UTF8_TAIL(CH(off-2)) && IS_UTF8_TAIL(CH(off-1))) + return (((unsigned int)CH(off-3) & 0x0f) << 12) | + (((unsigned int)CH(off-2) & 0x3f) << 6) | + (((unsigned int)CH(off-1) & 0x3f) << 0); + + if(off > 3 && IS_UTF8_LEAD4(CH(off-4)) && IS_UTF8_TAIL(CH(off-3)) && IS_UTF8_TAIL(CH(off-2)) && IS_UTF8_TAIL(CH(off-1))) + return (((unsigned int)CH(off-4) & 0x07) << 18) | + (((unsigned int)CH(off-3) & 0x3f) << 12) | + (((unsigned int)CH(off-2) & 0x3f) << 6) | + (((unsigned int)CH(off-1) & 0x3f) << 0); + } + + return (unsigned) CH(off-1); + } + + #define ISUNICODEWHITESPACE_(codepoint) md_is_unicode_whitespace__(codepoint) + #define ISUNICODEWHITESPACE(off) md_is_unicode_whitespace__(md_decode_utf8__(STR(off), ctx->size - (off), NULL)) + #define ISUNICODEWHITESPACEBEFORE(off) md_is_unicode_whitespace__(md_decode_utf8_before__(ctx, off)) + + #define ISUNICODEPUNCT(off) md_is_unicode_punct__(md_decode_utf8__(STR(off), ctx->size - (off), NULL)) + #define ISUNICODEPUNCTBEFORE(off) md_is_unicode_punct__(md_decode_utf8_before__(ctx, off)) + + static inline unsigned + md_decode_unicode(const CHAR* str, OFF off, SZ str_size, SZ* p_char_size) + { + return md_decode_utf8__(str+off, str_size-off, p_char_size); + } +#else + #define ISUNICODEWHITESPACE_(codepoint) ISWHITESPACE_(codepoint) + #define ISUNICODEWHITESPACE(off) ISWHITESPACE(off) + #define ISUNICODEWHITESPACEBEFORE(off) ISWHITESPACE((off)-1) + + #define ISUNICODEPUNCT(off) ISPUNCT(off) + #define ISUNICODEPUNCTBEFORE(off) ISPUNCT((off)-1) + + static inline void + md_get_unicode_fold_info(unsigned codepoint, MD_UNICODE_FOLD_INFO* info) + { + info->codepoints[0] = codepoint; + if(ISUPPER_(codepoint)) + info->codepoints[0] += 'a' - 'A'; + info->n_codepoints = 1; + } + + static inline unsigned + md_decode_unicode(const CHAR* str, OFF off, SZ str_size, SZ* p_size) + { + *p_size = 1; + return (unsigned) str[off]; + } +#endif + + +/************************************* + *** Helper string manipulations *** + *************************************/ + +/* Fill buffer with copy of the string between 'beg' and 'end' but replace any + * line breaks with given replacement character. + * + * NOTE: Caller is responsible to make sure the buffer is large enough. + * (Given the output is always shorter then input, (end - beg) is good idea + * what the caller should allocate.) + */ +static void +md_merge_lines(MD_CTX* ctx, OFF beg, OFF end, const MD_LINE* lines, int n_lines, + CHAR line_break_replacement_char, CHAR* buffer, SZ* p_size) +{ + CHAR* ptr = buffer; + int line_index = 0; + OFF off = beg; + + while(1) { + const MD_LINE* line = &lines[line_index]; + OFF line_end = line->end; + if(end < line_end) + line_end = end; + + while(off < line_end) { + *ptr = CH(off); + ptr++; + off++; + } + + if(off >= end) { + *p_size = ptr - buffer; + return; + } + + *ptr = line_break_replacement_char; + ptr++; + + line_index++; + off = lines[line_index].beg; + } +} + +/* Wrapper of md_merge_lines() which allocates new buffer for the output string. + */ +static int +md_merge_lines_alloc(MD_CTX* ctx, OFF beg, OFF end, const MD_LINE* lines, int n_lines, + CHAR line_break_replacement_char, CHAR** p_str, SZ* p_size) +{ + CHAR* buffer; + + buffer = (CHAR*) malloc(sizeof(CHAR) * (end - beg)); + if(buffer == NULL) { + MD_LOG("malloc() failed."); + return -1; + } + + md_merge_lines(ctx, beg, end, lines, n_lines, + line_break_replacement_char, buffer, p_size); + + *p_str = buffer; + return 0; +} + +static OFF +md_skip_unicode_whitespace(const CHAR* label, OFF off, SZ size) +{ + SZ char_size; + unsigned codepoint; + + while(off < size) { + codepoint = md_decode_unicode(label, off, size, &char_size); + if(!ISUNICODEWHITESPACE_(codepoint) && !ISNEWLINE_(label[off])) + break; + off += char_size; + } + + return off; +} + + +/****************************** + *** Recognizing raw HTML *** + ******************************/ + +/* md_is_html_tag() may be called when processing inlines (inline raw HTML) + * or when breaking document to blocks (checking for start of HTML block type 7). + * + * When breaking document to blocks, we do not yet know line boundaries, but + * in that case the whole tag has to live on a single line. We distinguish this + * by n_lines == 0. + */ +static int +md_is_html_tag(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + int attr_state; + OFF off = beg; + OFF line_end = (n_lines > 0) ? lines[0].end : ctx->size; + int i = 0; + + MD_ASSERT(CH(beg) == _T('<')); + + if(off + 1 >= line_end) + return FALSE; + off++; + + /* For parsing attributes, we need a little state automaton below. + * State -1: no attributes are allowed. + * State 0: attribute could follow after some whitespace. + * State 1: after a whitespace (attribute name may follow). + * State 2: after attribute name ('=' MAY follow). + * State 3: after '=' (value specification MUST follow). + * State 41: in middle of unquoted attribute value. + * State 42: in middle of single-quoted attribute value. + * State 43: in middle of double-quoted attribute value. + */ + attr_state = 0; + + if(CH(off) == _T('/')) { + /* Closer tag "</ ... >". No attributes may be present. */ + attr_state = -1; + off++; + } + + /* Tag name */ + if(off >= line_end || !ISALPHA(off)) + return FALSE; + off++; + while(off < line_end && (ISALNUM(off) || CH(off) == _T('-'))) + off++; + + /* (Optional) attributes (if not closer), (optional) '/' (if not closer) + * and final '>'. */ + while(1) { + while(off < line_end && !ISNEWLINE(off)) { + if(attr_state > 40) { + if(attr_state == 41 && (ISBLANK(off) || ISANYOF(off, _T("\"'=<>`")))) { + attr_state = 0; + off--; /* Put the char back for re-inspection in the new state. */ + } else if(attr_state == 42 && CH(off) == _T('\'')) { + attr_state = 0; + } else if(attr_state == 43 && CH(off) == _T('"')) { + attr_state = 0; + } + off++; + } else if(ISWHITESPACE(off)) { + if(attr_state == 0) + attr_state = 1; + off++; + } else if(attr_state <= 2 && CH(off) == _T('>')) { + /* End. */ + goto done; + } else if(attr_state <= 2 && CH(off) == _T('/') && off+1 < line_end && CH(off+1) == _T('>')) { + /* End with digraph '/>' */ + off++; + goto done; + } else if((attr_state == 1 || attr_state == 2) && (ISALPHA(off) || CH(off) == _T('_') || CH(off) == _T(':'))) { + off++; + /* Attribute name */ + while(off < line_end && (ISALNUM(off) || ISANYOF(off, _T("_.:-")))) + off++; + attr_state = 2; + } else if(attr_state == 2 && CH(off) == _T('=')) { + /* Attribute assignment sign */ + off++; + attr_state = 3; + } else if(attr_state == 3) { + /* Expecting start of attribute value. */ + if(CH(off) == _T('"')) + attr_state = 43; + else if(CH(off) == _T('\'')) + attr_state = 42; + else if(!ISANYOF(off, _T("\"'=<>`")) && !ISNEWLINE(off)) + attr_state = 41; + else + return FALSE; + off++; + } else { + /* Anything unexpected. */ + return FALSE; + } + } + + /* We have to be on a single line. See definition of start condition + * of HTML block, type 7. */ + if(n_lines == 0) + return FALSE; + + i++; + if(i >= n_lines) + return FALSE; + + off = lines[i].beg; + line_end = lines[i].end; + + if(attr_state == 0 || attr_state == 41) + attr_state = 1; + + if(off >= max_end) + return FALSE; + } + +done: + if(off >= max_end) + return FALSE; + + *p_end = off+1; + return TRUE; +} + +static int +md_scan_for_html_closer(MD_CTX* ctx, const MD_CHAR* str, MD_SIZE len, + const MD_LINE* lines, int n_lines, + OFF beg, OFF max_end, OFF* p_end, + OFF* p_scan_horizon) +{ + OFF off = beg; + int i = 0; + + if(off < *p_scan_horizon && *p_scan_horizon >= max_end - len) { + /* We have already scanned the range up to the max_end so we know + * there is nothing to see. */ + return FALSE; + } + + while(TRUE) { + while(off + len <= lines[i].end && off + len <= max_end) { + if(md_ascii_eq(STR(off), str, len)) { + /* Success. */ + *p_end = off + len; + return TRUE; + } + off++; + } + + i++; + if(off >= max_end || i >= n_lines) { + /* Failure. */ + *p_scan_horizon = off; + return FALSE; + } + + off = lines[i].beg; + } +} + +static int +md_is_html_comment(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + MD_ASSERT(CH(beg) == _T('<')); + + if(off + 4 >= lines[0].end) + return FALSE; + if(CH(off+1) != _T('!') || CH(off+2) != _T('-') || CH(off+3) != _T('-')) + return FALSE; + off += 4; + + /* ">" and "->" must not follow the opening. */ + if(off < lines[0].end && CH(off) == _T('>')) + return FALSE; + if(off+1 < lines[0].end && CH(off) == _T('-') && CH(off+1) == _T('>')) + return FALSE; + + /* HTML comment must not contyain "--", so we scan just for "--" instead + * of "-->" and verify manually that '>' follows. */ + if(md_scan_for_html_closer(ctx, _T("--"), 2, + lines, n_lines, off, max_end, p_end, &ctx->html_comment_horizon)) + { + if(*p_end < max_end && CH(*p_end) == _T('>')) { + *p_end = *p_end + 1; + return TRUE; + } + } + + return FALSE; +} + +static int +md_is_html_processing_instruction(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + if(off + 2 >= lines[0].end) + return FALSE; + if(CH(off+1) != _T('?')) + return FALSE; + off += 2; + + return md_scan_for_html_closer(ctx, _T("?>"), 2, + lines, n_lines, off, max_end, p_end, &ctx->html_proc_instr_horizon); +} + +static int +md_is_html_declaration(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + if(off + 2 >= lines[0].end) + return FALSE; + if(CH(off+1) != _T('!')) + return FALSE; + off += 2; + + /* Declaration name. */ + if(off >= lines[0].end || !ISALPHA(off)) + return FALSE; + off++; + while(off < lines[0].end && ISALPHA(off)) + off++; + if(off < lines[0].end && !ISWHITESPACE(off)) + return FALSE; + + return md_scan_for_html_closer(ctx, _T(">"), 1, + lines, n_lines, off, max_end, p_end, &ctx->html_decl_horizon); +} + +static int +md_is_html_cdata(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + static const CHAR open_str[] = _T("<![CDATA["); + static const SZ open_size = SIZEOF_ARRAY(open_str) - 1; + + OFF off = beg; + + if(off + open_size >= lines[0].end) + return FALSE; + if(memcmp(STR(off), open_str, open_size) != 0) + return FALSE; + off += open_size; + + if(lines[n_lines-1].end < max_end) + max_end = lines[n_lines-1].end - 2; + + return md_scan_for_html_closer(ctx, _T("]]>"), 3, + lines, n_lines, off, max_end, p_end, &ctx->html_cdata_horizon); +} + +static int +md_is_html_any(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, OFF max_end, OFF* p_end) +{ + MD_ASSERT(CH(beg) == _T('<')); + return (md_is_html_tag(ctx, lines, n_lines, beg, max_end, p_end) || + md_is_html_comment(ctx, lines, n_lines, beg, max_end, p_end) || + md_is_html_processing_instruction(ctx, lines, n_lines, beg, max_end, p_end) || + md_is_html_declaration(ctx, lines, n_lines, beg, max_end, p_end) || + md_is_html_cdata(ctx, lines, n_lines, beg, max_end, p_end)); +} + + +/**************************** + *** Recognizing Entity *** + ****************************/ + +static int +md_is_hex_entity_contents(MD_CTX* ctx, const CHAR* text, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + while(off < max_end && ISXDIGIT_(text[off]) && off - beg <= 8) + off++; + + if(1 <= off - beg && off - beg <= 6) { + *p_end = off; + return TRUE; + } else { + return FALSE; + } +} + +static int +md_is_dec_entity_contents(MD_CTX* ctx, const CHAR* text, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + while(off < max_end && ISDIGIT_(text[off]) && off - beg <= 8) + off++; + + if(1 <= off - beg && off - beg <= 7) { + *p_end = off; + return TRUE; + } else { + return FALSE; + } +} + +static int +md_is_named_entity_contents(MD_CTX* ctx, const CHAR* text, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg; + + if(off < max_end && ISALPHA_(text[off])) + off++; + else + return FALSE; + + while(off < max_end && ISALNUM_(text[off]) && off - beg <= 48) + off++; + + if(2 <= off - beg && off - beg <= 48) { + *p_end = off; + return TRUE; + } else { + return FALSE; + } +} + +static int +md_is_entity_str(MD_CTX* ctx, const CHAR* text, OFF beg, OFF max_end, OFF* p_end) +{ + int is_contents; + OFF off = beg; + + MD_ASSERT(text[off] == _T('&')); + off++; + + if(off+2 < max_end && text[off] == _T('#') && (text[off+1] == _T('x') || text[off+1] == _T('X'))) + is_contents = md_is_hex_entity_contents(ctx, text, off+2, max_end, &off); + else if(off+1 < max_end && text[off] == _T('#')) + is_contents = md_is_dec_entity_contents(ctx, text, off+1, max_end, &off); + else + is_contents = md_is_named_entity_contents(ctx, text, off, max_end, &off); + + if(is_contents && off < max_end && text[off] == _T(';')) { + *p_end = off+1; + return TRUE; + } else { + return FALSE; + } +} + +static inline int +md_is_entity(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end) +{ + return md_is_entity_str(ctx, ctx->text, beg, max_end, p_end); +} + + +/****************************** + *** Attribute Management *** + ******************************/ + +typedef struct MD_ATTRIBUTE_BUILD_tag MD_ATTRIBUTE_BUILD; +struct MD_ATTRIBUTE_BUILD_tag { + CHAR* text; + MD_TEXTTYPE* substr_types; + OFF* substr_offsets; + int substr_count; + int substr_alloc; + MD_TEXTTYPE trivial_types[1]; + OFF trivial_offsets[2]; +}; + + +#define MD_BUILD_ATTR_NO_ESCAPES 0x0001 + +static int +md_build_attr_append_substr(MD_CTX* ctx, MD_ATTRIBUTE_BUILD* build, + MD_TEXTTYPE type, OFF off) +{ + if(build->substr_count >= build->substr_alloc) { + MD_TEXTTYPE* new_substr_types; + OFF* new_substr_offsets; + + build->substr_alloc = (build->substr_alloc == 0 ? 8 : build->substr_alloc * 2); + + new_substr_types = (MD_TEXTTYPE*) realloc(build->substr_types, + build->substr_alloc * sizeof(MD_TEXTTYPE)); + if(new_substr_types == NULL) { + MD_LOG("realloc() failed."); + return -1; + } + /* Note +1 to reserve space for final offset (== raw_size). */ + new_substr_offsets = (OFF*) realloc(build->substr_offsets, + (build->substr_alloc+1) * sizeof(OFF)); + if(new_substr_offsets == NULL) { + MD_LOG("realloc() failed."); + free(new_substr_types); + return -1; + } + + build->substr_types = new_substr_types; + build->substr_offsets = new_substr_offsets; + } + + build->substr_types[build->substr_count] = type; + build->substr_offsets[build->substr_count] = off; + build->substr_count++; + return 0; +} + +static void +md_free_attribute(MD_CTX* ctx, MD_ATTRIBUTE_BUILD* build) +{ + if(build->substr_alloc > 0) { + free(build->text); + free(build->substr_types); + free(build->substr_offsets); + } +} + +static int +md_build_attribute(MD_CTX* ctx, const CHAR* raw_text, SZ raw_size, + unsigned flags, MD_ATTRIBUTE* attr, MD_ATTRIBUTE_BUILD* build) +{ + OFF raw_off, off; + int is_trivial; + int ret = 0; + + memset(build, 0, sizeof(MD_ATTRIBUTE_BUILD)); + + /* If there is no backslash and no ampersand, build trivial attribute + * without any malloc(). */ + is_trivial = TRUE; + for(raw_off = 0; raw_off < raw_size; raw_off++) { + if(ISANYOF3_(raw_text[raw_off], _T('\\'), _T('&'), _T('\0'))) { + is_trivial = FALSE; + break; + } + } + + if(is_trivial) { + build->text = (CHAR*) (raw_size ? raw_text : NULL); + build->substr_types = build->trivial_types; + build->substr_offsets = build->trivial_offsets; + build->substr_count = 1; + build->substr_alloc = 0; + build->trivial_types[0] = MD_TEXT_NORMAL; + build->trivial_offsets[0] = 0; + build->trivial_offsets[1] = raw_size; + off = raw_size; + } else { + build->text = (CHAR*) malloc(raw_size * sizeof(CHAR)); + if(build->text == NULL) { + MD_LOG("malloc() failed."); + goto abort; + } + + raw_off = 0; + off = 0; + + while(raw_off < raw_size) { + if(raw_text[raw_off] == _T('\0')) { + MD_CHECK(md_build_attr_append_substr(ctx, build, MD_TEXT_NULLCHAR, off)); + memcpy(build->text + off, raw_text + raw_off, 1); + off++; + raw_off++; + continue; + } + + if(raw_text[raw_off] == _T('&')) { + OFF ent_end; + + if(md_is_entity_str(ctx, raw_text, raw_off, raw_size, &ent_end)) { + MD_CHECK(md_build_attr_append_substr(ctx, build, MD_TEXT_ENTITY, off)); + memcpy(build->text + off, raw_text + raw_off, ent_end - raw_off); + off += ent_end - raw_off; + raw_off = ent_end; + continue; + } + } + + if(build->substr_count == 0 || build->substr_types[build->substr_count-1] != MD_TEXT_NORMAL) + MD_CHECK(md_build_attr_append_substr(ctx, build, MD_TEXT_NORMAL, off)); + + if(!(flags & MD_BUILD_ATTR_NO_ESCAPES) && + raw_text[raw_off] == _T('\\') && raw_off+1 < raw_size && + (ISPUNCT_(raw_text[raw_off+1]) || ISNEWLINE_(raw_text[raw_off+1]))) + raw_off++; + + build->text[off++] = raw_text[raw_off++]; + } + build->substr_offsets[build->substr_count] = off; + } + + attr->text = build->text; + attr->size = off; + attr->substr_offsets = build->substr_offsets; + attr->substr_types = build->substr_types; + return 0; + +abort: + md_free_attribute(ctx, build); + return -1; +} + + +/********************************************* + *** Dictionary of Reference Definitions *** + *********************************************/ + +#define MD_FNV1A_BASE 2166136261 +#define MD_FNV1A_PRIME 16777619 + +static inline unsigned +md_fnv1a(unsigned base, const void* data, size_t n) +{ + const unsigned char* buf = (const unsigned char*) data; + unsigned hash = base; + size_t i; + + for(i = 0; i < n; i++) { + hash ^= buf[i]; + hash *= MD_FNV1A_PRIME; + } + + return hash; +} + + +struct MD_REF_DEF_tag { + CHAR* label; + CHAR* title; + unsigned hash; + SZ label_size : 24; + unsigned label_needs_free : 1; + unsigned title_needs_free : 1; + SZ title_size; + OFF dest_beg; + OFF dest_end; +}; + +/* Label equivalence is quite complicated with regards to whitespace and case + * folding. This complicates computing a hash of it as well as direct comparison + * of two labels. */ + +static unsigned +md_link_label_hash(const CHAR* label, SZ size) +{ + unsigned hash = MD_FNV1A_BASE; + OFF off; + unsigned codepoint; + int is_whitespace = FALSE; + + off = md_skip_unicode_whitespace(label, 0, size); + while(off < size) { + SZ char_size; + + codepoint = md_decode_unicode(label, off, size, &char_size); + is_whitespace = ISUNICODEWHITESPACE_(codepoint) || ISNEWLINE_(label[off]); + + if(is_whitespace) { + codepoint = ' '; + hash = md_fnv1a(hash, &codepoint, sizeof(unsigned)); + off = md_skip_unicode_whitespace(label, off, size); + } else { + MD_UNICODE_FOLD_INFO fold_info; + + md_get_unicode_fold_info(codepoint, &fold_info); + hash = md_fnv1a(hash, fold_info.codepoints, fold_info.n_codepoints * sizeof(unsigned)); + off += char_size; + } + } + + return hash; +} + +static OFF +md_link_label_cmp_load_fold_info(const CHAR* label, OFF off, SZ size, + MD_UNICODE_FOLD_INFO* fold_info) +{ + unsigned codepoint; + SZ char_size; + + if(off >= size) { + /* Treat end of link label as a whitespace. */ + goto whitespace; + } + + if(ISNEWLINE_(label[off])) { + /* Treat new lines as a whitespace. */ + off++; + goto whitespace; + } + + codepoint = md_decode_unicode(label, off, size, &char_size); + off += char_size; + if(ISUNICODEWHITESPACE_(codepoint)) { + /* Treat all whitespace as equivalent */ + goto whitespace; + } + + /* Get real folding info. */ + md_get_unicode_fold_info(codepoint, fold_info); + return off; + +whitespace: + fold_info->codepoints[0] = _T(' '); + fold_info->n_codepoints = 1; + return off; +} + +static int +md_link_label_cmp(const CHAR* a_label, SZ a_size, const CHAR* b_label, SZ b_size) +{ + OFF a_off; + OFF b_off; + int a_reached_end = FALSE; + int b_reached_end = FALSE; + MD_UNICODE_FOLD_INFO a_fi = { 0 }; + MD_UNICODE_FOLD_INFO b_fi = { 0 }; + OFF a_fi_off = 0; + OFF b_fi_off = 0; + int cmp; + + a_off = md_skip_unicode_whitespace(a_label, 0, a_size); + b_off = md_skip_unicode_whitespace(b_label, 0, b_size); + while(!a_reached_end && !b_reached_end) { + /* If needed, load fold info for next char. */ + if(a_fi_off >= a_fi.n_codepoints) { + a_fi_off = 0; + a_off = md_link_label_cmp_load_fold_info(a_label, a_off, a_size, &a_fi); + a_reached_end = (a_off >= a_size); + } + if(b_fi_off >= b_fi.n_codepoints) { + b_fi_off = 0; + b_off = md_link_label_cmp_load_fold_info(b_label, b_off, b_size, &b_fi); + b_reached_end = (b_off >= b_size); + } + + cmp = b_fi.codepoints[b_fi_off] - a_fi.codepoints[a_fi_off]; + if(cmp != 0) + return cmp; + + a_fi_off++; + b_fi_off++; + } + + return 0; +} + +typedef struct MD_REF_DEF_LIST_tag MD_REF_DEF_LIST; +struct MD_REF_DEF_LIST_tag { + int n_ref_defs; + int alloc_ref_defs; + MD_REF_DEF* ref_defs[]; /* Valid items always point into ctx->ref_defs[] */ +}; + +static int +md_ref_def_cmp(const void* a, const void* b) +{ + const MD_REF_DEF* a_ref = *(const MD_REF_DEF**)a; + const MD_REF_DEF* b_ref = *(const MD_REF_DEF**)b; + + if(a_ref->hash < b_ref->hash) + return -1; + else if(a_ref->hash > b_ref->hash) + return +1; + else + return md_link_label_cmp(a_ref->label, a_ref->label_size, b_ref->label, b_ref->label_size); +} + +static int +md_ref_def_cmp_stable(const void* a, const void* b) +{ + int cmp; + + cmp = md_ref_def_cmp(a, b); + + /* Ensure stability of the sorting. */ + if(cmp == 0) { + const MD_REF_DEF* a_ref = *(const MD_REF_DEF**)a; + const MD_REF_DEF* b_ref = *(const MD_REF_DEF**)b; + + if(a_ref < b_ref) + cmp = -1; + else if(a_ref > b_ref) + cmp = +1; + else + cmp = 0; + } + + return cmp; +} + +static int +md_build_ref_def_hashtable(MD_CTX* ctx) +{ + int i, j; + + if(ctx->n_ref_defs == 0) + return 0; + + ctx->ref_def_hashtable_size = (ctx->n_ref_defs * 5) / 4; + ctx->ref_def_hashtable = malloc(ctx->ref_def_hashtable_size * sizeof(void*)); + if(ctx->ref_def_hashtable == NULL) { + MD_LOG("malloc() failed."); + goto abort; + } + memset(ctx->ref_def_hashtable, 0, ctx->ref_def_hashtable_size * sizeof(void*)); + + /* Each member of ctx->ref_def_hashtable[] can be: + * -- NULL, + * -- pointer to the MD_REF_DEF in ctx->ref_defs[], or + * -- pointer to a MD_REF_DEF_LIST, which holds multiple pointers to + * such MD_REF_DEFs. + */ + for(i = 0; i < ctx->n_ref_defs; i++) { + MD_REF_DEF* def = &ctx->ref_defs[i]; + void* bucket; + MD_REF_DEF_LIST* list; + + def->hash = md_link_label_hash(def->label, def->label_size); + bucket = ctx->ref_def_hashtable[def->hash % ctx->ref_def_hashtable_size]; + + if(bucket == NULL) { + ctx->ref_def_hashtable[def->hash % ctx->ref_def_hashtable_size] = def; + continue; + } + + if(ctx->ref_defs <= (MD_REF_DEF*) bucket && (MD_REF_DEF*) bucket < ctx->ref_defs + ctx->n_ref_defs) { + /* The bucket already contains one ref. def. Lets see whether it + * is the same label (ref. def. duplicate) or different one + * (hash conflict). */ + MD_REF_DEF* old_def = (MD_REF_DEF*) bucket; + + if(md_link_label_cmp(def->label, def->label_size, old_def->label, old_def->label_size) == 0) { + /* Ignore this ref. def. */ + continue; + } + + /* Make the bucket capable of holding more ref. defs. */ + list = (MD_REF_DEF_LIST*) malloc(sizeof(MD_REF_DEF_LIST) + 4 * sizeof(MD_REF_DEF)); + if(list == NULL) { + MD_LOG("malloc() failed."); + goto abort; + } + list->ref_defs[0] = old_def; + list->ref_defs[1] = def; + list->n_ref_defs = 2; + list->alloc_ref_defs = 4; + ctx->ref_def_hashtable[def->hash % ctx->ref_def_hashtable_size] = list; + continue; + } + + /* Append the def to the bucket list. */ + list = (MD_REF_DEF_LIST*) bucket; + if(list->n_ref_defs >= list->alloc_ref_defs) { + MD_REF_DEF_LIST* list_tmp = (MD_REF_DEF_LIST*) realloc(list, + sizeof(MD_REF_DEF_LIST) + 2 * list->alloc_ref_defs * sizeof(MD_REF_DEF)); + if(list_tmp == NULL) { + MD_LOG("realloc() failed."); + goto abort; + } + list = list_tmp; + list->alloc_ref_defs *= 2; + ctx->ref_def_hashtable[def->hash % ctx->ref_def_hashtable_size] = list; + } + + list->ref_defs[list->n_ref_defs] = def; + list->n_ref_defs++; + } + + /* Sort the complex buckets so we can use bsearch() with them. */ + for(i = 0; i < ctx->ref_def_hashtable_size; i++) { + void* bucket = ctx->ref_def_hashtable[i]; + MD_REF_DEF_LIST* list; + + if(bucket == NULL) + continue; + if(ctx->ref_defs <= (MD_REF_DEF*) bucket && (MD_REF_DEF*) bucket < ctx->ref_defs + ctx->n_ref_defs) + continue; + + list = (MD_REF_DEF_LIST*) bucket; + qsort(list->ref_defs, list->n_ref_defs, sizeof(MD_REF_DEF*), md_ref_def_cmp_stable); + + /* Disable duplicates. */ + for(j = 1; j < list->n_ref_defs; j++) { + if(md_ref_def_cmp(&list->ref_defs[j-1], &list->ref_defs[j]) == 0) + list->ref_defs[j] = list->ref_defs[j-1]; + } + } + + return 0; + +abort: + return -1; +} + +static void +md_free_ref_def_hashtable(MD_CTX* ctx) +{ + if(ctx->ref_def_hashtable != NULL) { + int i; + + for(i = 0; i < ctx->ref_def_hashtable_size; i++) { + void* bucket = ctx->ref_def_hashtable[i]; + if(bucket == NULL) + continue; + if(ctx->ref_defs <= (MD_REF_DEF*) bucket && (MD_REF_DEF*) bucket < ctx->ref_defs + ctx->n_ref_defs) + continue; + free(bucket); + } + + free(ctx->ref_def_hashtable); + } +} + +static const MD_REF_DEF* +md_lookup_ref_def(MD_CTX* ctx, const CHAR* label, SZ label_size) +{ + unsigned hash; + void* bucket; + + if(ctx->ref_def_hashtable_size == 0) + return NULL; + + hash = md_link_label_hash(label, label_size); + bucket = ctx->ref_def_hashtable[hash % ctx->ref_def_hashtable_size]; + + if(bucket == NULL) { + return NULL; + } else if(ctx->ref_defs <= (MD_REF_DEF*) bucket && (MD_REF_DEF*) bucket < ctx->ref_defs + ctx->n_ref_defs) { + const MD_REF_DEF* def = (MD_REF_DEF*) bucket; + + if(md_link_label_cmp(def->label, def->label_size, label, label_size) == 0) + return def; + else + return NULL; + } else { + MD_REF_DEF_LIST* list = (MD_REF_DEF_LIST*) bucket; + MD_REF_DEF key_buf; + const MD_REF_DEF* key = &key_buf; + const MD_REF_DEF** ret; + + key_buf.label = (CHAR*) label; + key_buf.label_size = label_size; + key_buf.hash = md_link_label_hash(key_buf.label, key_buf.label_size); + + ret = (const MD_REF_DEF**) bsearch(&key, list->ref_defs, + list->n_ref_defs, sizeof(MD_REF_DEF*), md_ref_def_cmp); + if(ret != NULL) + return *ret; + else + return NULL; + } +} + + +/*************************** + *** Recognizing Links *** + ***************************/ + +/* Note this code is partially shared between processing inlines and blocks + * as reference definitions and links share some helper parser functions. + */ + +typedef struct MD_LINK_ATTR_tag MD_LINK_ATTR; +struct MD_LINK_ATTR_tag { + OFF dest_beg; + OFF dest_end; + + CHAR* title; + SZ title_size; + int title_needs_free; +}; + + +static int +md_is_link_label(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, + OFF* p_end, int* p_beg_line_index, int* p_end_line_index, + OFF* p_contents_beg, OFF* p_contents_end) +{ + OFF off = beg; + OFF contents_beg = 0; + OFF contents_end = 0; + int line_index = 0; + int len = 0; + + if(CH(off) != _T('[')) + return FALSE; + off++; + + while(1) { + OFF line_end = lines[line_index].end; + + while(off < line_end) { + if(CH(off) == _T('\\') && off+1 < ctx->size && (ISPUNCT(off+1) || ISNEWLINE(off+1))) { + if(contents_end == 0) { + contents_beg = off; + *p_beg_line_index = line_index; + } + contents_end = off + 2; + off += 2; + } else if(CH(off) == _T('[')) { + return FALSE; + } else if(CH(off) == _T(']')) { + if(contents_beg < contents_end) { + /* Success. */ + *p_contents_beg = contents_beg; + *p_contents_end = contents_end; + *p_end = off+1; + *p_end_line_index = line_index; + return TRUE; + } else { + /* Link label must have some non-whitespace contents. */ + return FALSE; + } + } else { + unsigned codepoint; + SZ char_size; + + codepoint = md_decode_unicode(ctx->text, off, ctx->size, &char_size); + if(!ISUNICODEWHITESPACE_(codepoint)) { + if(contents_end == 0) { + contents_beg = off; + *p_beg_line_index = line_index; + } + contents_end = off + char_size; + } + + off += char_size; + } + + len++; + if(len > 999) + return FALSE; + } + + line_index++; + len++; + if(line_index < n_lines) + off = lines[line_index].beg; + else + break; + } + + return FALSE; +} + +static int +md_is_link_destination_A(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end, + OFF* p_contents_beg, OFF* p_contents_end) +{ + OFF off = beg; + + if(off >= max_end || CH(off) != _T('<')) + return FALSE; + off++; + + while(off < max_end) { + if(CH(off) == _T('\\') && off+1 < max_end && ISPUNCT(off+1)) { + off += 2; + continue; + } + + if(ISNEWLINE(off) || CH(off) == _T('<')) + return FALSE; + + if(CH(off) == _T('>')) { + /* Success. */ + *p_contents_beg = beg+1; + *p_contents_end = off; + *p_end = off+1; + return TRUE; + } + + off++; + } + + return FALSE; +} + +static int +md_is_link_destination_B(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end, + OFF* p_contents_beg, OFF* p_contents_end) +{ + OFF off = beg; + int parenthesis_level = 0; + + while(off < max_end) { + if(CH(off) == _T('\\') && off+1 < max_end && ISPUNCT(off+1)) { + off += 2; + continue; + } + + if(ISWHITESPACE(off) || ISCNTRL(off)) + break; + + /* Link destination may include balanced pairs of unescaped '(' ')'. + * Note we limit the maximal nesting level by 32 to protect us from + * https://github.com/jgm/cmark/issues/214 */ + if(CH(off) == _T('(')) { + parenthesis_level++; + if(parenthesis_level > 32) + return FALSE; + } else if(CH(off) == _T(')')) { + if(parenthesis_level == 0) + break; + parenthesis_level--; + } + + off++; + } + + if(parenthesis_level != 0 || off == beg) + return FALSE; + + /* Success. */ + *p_contents_beg = beg; + *p_contents_end = off; + *p_end = off; + return TRUE; +} + +static inline int +md_is_link_destination(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end, + OFF* p_contents_beg, OFF* p_contents_end) +{ + if(CH(beg) == _T('<')) + return md_is_link_destination_A(ctx, beg, max_end, p_end, p_contents_beg, p_contents_end); + else + return md_is_link_destination_B(ctx, beg, max_end, p_end, p_contents_beg, p_contents_end); +} + +static int +md_is_link_title(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, + OFF* p_end, int* p_beg_line_index, int* p_end_line_index, + OFF* p_contents_beg, OFF* p_contents_end) +{ + OFF off = beg; + CHAR closer_char; + int line_index = 0; + + /* White space with up to one line break. */ + while(off < lines[line_index].end && ISWHITESPACE(off)) + off++; + if(off >= lines[line_index].end) { + line_index++; + if(line_index >= n_lines) + return FALSE; + off = lines[line_index].beg; + } + if(off == beg) + return FALSE; + + *p_beg_line_index = line_index; + + /* First char determines how to detect end of it. */ + switch(CH(off)) { + case _T('"'): closer_char = _T('"'); break; + case _T('\''): closer_char = _T('\''); break; + case _T('('): closer_char = _T(')'); break; + default: return FALSE; + } + off++; + + *p_contents_beg = off; + + while(line_index < n_lines) { + OFF line_end = lines[line_index].end; + + while(off < line_end) { + if(CH(off) == _T('\\') && off+1 < ctx->size && (ISPUNCT(off+1) || ISNEWLINE(off+1))) { + off++; + } else if(CH(off) == closer_char) { + /* Success. */ + *p_contents_end = off; + *p_end = off+1; + *p_end_line_index = line_index; + return TRUE; + } else if(closer_char == _T(')') && CH(off) == _T('(')) { + /* ()-style title cannot contain (unescaped '(')) */ + return FALSE; + } + + off++; + } + + line_index++; + } + + return FALSE; +} + +/* Returns 0 if it is not a reference definition. + * + * Returns N > 0 if it is a reference definition. N then corresponds to the + * number of lines forming it). In this case the definition is stored for + * resolving any links referring to it. + * + * Returns -1 in case of an error (out of memory). + */ +static int +md_is_link_reference_definition(MD_CTX* ctx, const MD_LINE* lines, int n_lines) +{ + OFF label_contents_beg; + OFF label_contents_end; + int label_contents_line_index = -1; + int label_is_multiline; + CHAR* label; + SZ label_size; + int label_needs_free = FALSE; + OFF dest_contents_beg; + OFF dest_contents_end; + OFF title_contents_beg; + OFF title_contents_end; + int title_contents_line_index; + int title_is_multiline; + OFF off; + int line_index = 0; + int tmp_line_index; + MD_REF_DEF* def; + int ret; + + /* Link label. */ + if(!md_is_link_label(ctx, lines, n_lines, lines[0].beg, + &off, &label_contents_line_index, &line_index, + &label_contents_beg, &label_contents_end)) + return FALSE; + label_is_multiline = (label_contents_line_index != line_index); + + /* Colon. */ + if(off >= lines[line_index].end || CH(off) != _T(':')) + return FALSE; + off++; + + /* Optional white space with up to one line break. */ + while(off < lines[line_index].end && ISWHITESPACE(off)) + off++; + if(off >= lines[line_index].end) { + line_index++; + if(line_index >= n_lines) + return FALSE; + off = lines[line_index].beg; + } + + /* Link destination. */ + if(!md_is_link_destination(ctx, off, lines[line_index].end, + &off, &dest_contents_beg, &dest_contents_end)) + return FALSE; + + /* (Optional) title. Note we interpret it as an title only if nothing + * more follows on its last line. */ + if(md_is_link_title(ctx, lines + line_index, n_lines - line_index, off, + &off, &title_contents_line_index, &tmp_line_index, + &title_contents_beg, &title_contents_end) + && off >= lines[line_index + tmp_line_index].end) + { + title_is_multiline = (tmp_line_index != title_contents_line_index); + title_contents_line_index += line_index; + line_index += tmp_line_index; + } else { + /* Not a title. */ + title_is_multiline = FALSE; + title_contents_beg = off; + title_contents_end = off; + title_contents_line_index = 0; + } + + /* Nothing more can follow on the last line. */ + if(off < lines[line_index].end) + return FALSE; + + /* Construct label. */ + if(!label_is_multiline) { + label = (CHAR*) STR(label_contents_beg); + label_size = label_contents_end - label_contents_beg; + label_needs_free = FALSE; + } else { + MD_CHECK(md_merge_lines_alloc(ctx, label_contents_beg, label_contents_end, + lines + label_contents_line_index, n_lines - label_contents_line_index, + _T(' '), &label, &label_size)); + label_needs_free = TRUE; + } + + /* Store the reference definition. */ + if(ctx->n_ref_defs >= ctx->alloc_ref_defs) { + MD_REF_DEF* new_defs; + + ctx->alloc_ref_defs = (ctx->alloc_ref_defs > 0 ? ctx->alloc_ref_defs * 2 : 16); + new_defs = (MD_REF_DEF*) realloc(ctx->ref_defs, ctx->alloc_ref_defs * sizeof(MD_REF_DEF)); + if(new_defs == NULL) { + MD_LOG("realloc() failed."); + ret = -1; + goto abort; + } + + ctx->ref_defs = new_defs; + } + + def = &ctx->ref_defs[ctx->n_ref_defs]; + memset(def, 0, sizeof(MD_REF_DEF)); + + def->label = label; + def->label_size = label_size; + def->label_needs_free = label_needs_free; + + def->dest_beg = dest_contents_beg; + def->dest_end = dest_contents_end; + + if(title_contents_beg >= title_contents_end) { + def->title = NULL; + def->title_size = 0; + } else if(!title_is_multiline) { + def->title = (CHAR*) STR(title_contents_beg); + def->title_size = title_contents_end - title_contents_beg; + } else { + MD_CHECK(md_merge_lines_alloc(ctx, title_contents_beg, title_contents_end, + lines + title_contents_line_index, n_lines - title_contents_line_index, + _T('\n'), &def->title, &def->title_size)); + def->title_needs_free = TRUE; + } + + /* Success. */ + ctx->n_ref_defs++; + return line_index + 1; + +abort: + /* Failure. */ + if(label_needs_free) + free(label); + return -1; +} + +static int +md_is_link_reference(MD_CTX* ctx, const MD_LINE* lines, int n_lines, + OFF beg, OFF end, MD_LINK_ATTR* attr) +{ + const MD_REF_DEF* def; + const MD_LINE* beg_line; + const MD_LINE* end_line; + CHAR* label; + SZ label_size; + int ret; + + MD_ASSERT(CH(beg) == _T('[') || CH(beg) == _T('!')); + MD_ASSERT(CH(end-1) == _T(']')); + + beg += (CH(beg) == _T('!') ? 2 : 1); + end--; + + /* Find lines corresponding to the beg and end positions. */ + MD_ASSERT(lines[0].beg <= beg); + beg_line = lines; + while(beg >= beg_line->end) + beg_line++; + + MD_ASSERT(end <= lines[n_lines-1].end); + end_line = beg_line; + while(end >= end_line->end) + end_line++; + + if(beg_line != end_line) { + MD_CHECK(md_merge_lines_alloc(ctx, beg, end, beg_line, + n_lines - (beg_line - lines), _T(' '), &label, &label_size)); + } else { + label = (CHAR*) STR(beg); + label_size = end - beg; + } + + def = md_lookup_ref_def(ctx, label, label_size); + if(def != NULL) { + attr->dest_beg = def->dest_beg; + attr->dest_end = def->dest_end; + attr->title = def->title; + attr->title_size = def->title_size; + attr->title_needs_free = FALSE; + } + + if(beg_line != end_line) + free(label); + + ret = (def != NULL); + +abort: + return ret; +} + +static int +md_is_inline_link_spec(MD_CTX* ctx, const MD_LINE* lines, int n_lines, + OFF beg, OFF* p_end, MD_LINK_ATTR* attr) +{ + int line_index = 0; + int tmp_line_index; + OFF title_contents_beg; + OFF title_contents_end; + int title_contents_line_index; + int title_is_multiline; + OFF off = beg; + int ret = FALSE; + + while(off >= lines[line_index].end) + line_index++; + + MD_ASSERT(CH(off) == _T('(')); + off++; + + /* Optional white space with up to one line break. */ + while(off < lines[line_index].end && ISWHITESPACE(off)) + off++; + if(off >= lines[line_index].end && ISNEWLINE(off)) { + line_index++; + if(line_index >= n_lines) + return FALSE; + off = lines[line_index].beg; + } + + /* Link destination may be omitted, but only when not also having a title. */ + if(off < ctx->size && CH(off) == _T(')')) { + attr->dest_beg = off; + attr->dest_end = off; + attr->title = NULL; + attr->title_size = 0; + attr->title_needs_free = FALSE; + off++; + *p_end = off; + return TRUE; + } + + /* Link destination. */ + if(!md_is_link_destination(ctx, off, lines[line_index].end, + &off, &attr->dest_beg, &attr->dest_end)) + return FALSE; + + /* (Optional) title. */ + if(md_is_link_title(ctx, lines + line_index, n_lines - line_index, off, + &off, &title_contents_line_index, &tmp_line_index, + &title_contents_beg, &title_contents_end)) + { + title_is_multiline = (tmp_line_index != title_contents_line_index); + title_contents_line_index += line_index; + line_index += tmp_line_index; + } else { + /* Not a title. */ + title_is_multiline = FALSE; + title_contents_beg = off; + title_contents_end = off; + title_contents_line_index = 0; + } + + /* Optional whitespace followed with final ')'. */ + while(off < lines[line_index].end && ISWHITESPACE(off)) + off++; + if(off >= lines[line_index].end && ISNEWLINE(off)) { + line_index++; + if(line_index >= n_lines) + return FALSE; + off = lines[line_index].beg; + } + if(CH(off) != _T(')')) + goto abort; + off++; + + if(title_contents_beg >= title_contents_end) { + attr->title = NULL; + attr->title_size = 0; + attr->title_needs_free = FALSE; + } else if(!title_is_multiline) { + attr->title = (CHAR*) STR(title_contents_beg); + attr->title_size = title_contents_end - title_contents_beg; + attr->title_needs_free = FALSE; + } else { + MD_CHECK(md_merge_lines_alloc(ctx, title_contents_beg, title_contents_end, + lines + title_contents_line_index, n_lines - title_contents_line_index, + _T('\n'), &attr->title, &attr->title_size)); + attr->title_needs_free = TRUE; + } + + *p_end = off; + ret = TRUE; + +abort: + return ret; +} + +static void +md_free_ref_defs(MD_CTX* ctx) +{ + int i; + + for(i = 0; i < ctx->n_ref_defs; i++) { + MD_REF_DEF* def = &ctx->ref_defs[i]; + + if(def->label_needs_free) + free(def->label); + if(def->title_needs_free) + free(def->title); + } + + free(ctx->ref_defs); +} + + +/****************************************** + *** Processing Inlines (a.k.a Spans) *** + ******************************************/ + +/* We process inlines in few phases: + * + * (1) We go through the block text and collect all significant characters + * which may start/end a span or some other significant position into + * ctx->marks[]. Core of this is what md_collect_marks() does. + * + * We also do some very brief preliminary context-less analysis, whether + * it might be opener or closer (e.g. of an emphasis span). + * + * This speeds the other steps as we do not need to re-iterate over all + * characters anymore. + * + * (2) We analyze each potential mark types, in order by their precedence. + * + * In each md_analyze_XXX() function, we re-iterate list of the marks, + * skipping already resolved regions (in preceding precedences) and try to + * resolve them. + * + * (2.1) For trivial marks, which are single (e.g. HTML entity), we just mark + * them as resolved. + * + * (2.2) For range-type marks, we analyze whether the mark could be closer + * and, if yes, whether there is some preceding opener it could satisfy. + * + * If not we check whether it could be really an opener and if yes, we + * remember it so subsequent closers may resolve it. + * + * (3) Finally, when all marks were analyzed, we render the block contents + * by calling MD_RENDERER::text() callback, interrupting by ::enter_span() + * or ::close_span() whenever we reach a resolved mark. + */ + + +/* The mark structure. + * + * '\\': Maybe escape sequence. + * '\0': NULL char. + * '*': Maybe (strong) emphasis start/end. + * '_': Maybe (strong) emphasis start/end. + * '~': Maybe strikethrough start/end (needs MD_FLAG_STRIKETHROUGH). + * '`': Maybe code span start/end. + * '&': Maybe start of entity. + * ';': Maybe end of entity. + * '<': Maybe start of raw HTML or autolink. + * '>': Maybe end of raw HTML or autolink. + * '[': Maybe start of link label or link text. + * '!': Equivalent of '[' for image. + * ']': Maybe end of link label or link text. + * '@': Maybe permissive e-mail auto-link (needs MD_FLAG_PERMISSIVEEMAILAUTOLINKS). + * ':': Maybe permissive URL auto-link (needs MD_FLAG_PERMISSIVEURLAUTOLINKS). + * '.': Maybe permissive WWW auto-link (needs MD_FLAG_PERMISSIVEWWWAUTOLINKS). + * 'D': Dummy mark, it reserves a space for splitting a previous mark + * (e.g. emphasis) or to make more space for storing some special data + * related to the preceding mark (e.g. link). + * + * Note that not all instances of these chars in the text imply creation of the + * structure. Only those which have (or may have, after we see more context) + * the special meaning. + * + * (Keep this struct as small as possible to fit as much of them into CPU + * cache line.) + */ +struct MD_MARK_tag { + OFF beg; + OFF end; + + /* For unresolved openers, 'prev' and 'next' form the chain of open openers + * of given type 'ch'. + * + * During resolving, we disconnect from the chain and point to the + * corresponding counterpart so opener points to its closer and vice versa. + */ + int prev; + int next; + CHAR ch; + unsigned char flags; +}; + +/* Mark flags (these apply to ALL mark types). */ +#define MD_MARK_POTENTIAL_OPENER 0x01 /* Maybe opener. */ +#define MD_MARK_POTENTIAL_CLOSER 0x02 /* Maybe closer. */ +#define MD_MARK_OPENER 0x04 /* Definitely opener. */ +#define MD_MARK_CLOSER 0x08 /* Definitely closer. */ +#define MD_MARK_RESOLVED 0x10 /* Resolved in any definite way. */ + +/* Mark flags specific for various mark types (so they can share bits). */ +#define MD_MARK_EMPH_INTRAWORD 0x20 /* Helper for the "rule of 3". */ +#define MD_MARK_EMPH_MOD3_0 0x40 +#define MD_MARK_EMPH_MOD3_1 0x80 +#define MD_MARK_EMPH_MOD3_2 (0x40 | 0x80) +#define MD_MARK_EMPH_MOD3_MASK (0x40 | 0x80) +#define MD_MARK_AUTOLINK 0x20 /* Distinguisher for '<', '>'. */ +#define MD_MARK_VALIDPERMISSIVEAUTOLINK 0x20 /* For permissive autolinks. */ + +static MD_MARKCHAIN* +md_asterisk_chain(MD_CTX* ctx, unsigned flags) +{ + switch(flags & (MD_MARK_EMPH_INTRAWORD | MD_MARK_EMPH_MOD3_MASK)) { + case MD_MARK_EMPH_INTRAWORD | MD_MARK_EMPH_MOD3_0: return &ASTERISK_OPENERS_intraword_mod3_0; + case MD_MARK_EMPH_INTRAWORD | MD_MARK_EMPH_MOD3_1: return &ASTERISK_OPENERS_intraword_mod3_1; + case MD_MARK_EMPH_INTRAWORD | MD_MARK_EMPH_MOD3_2: return &ASTERISK_OPENERS_intraword_mod3_2; + case MD_MARK_EMPH_MOD3_0: return &ASTERISK_OPENERS_extraword_mod3_0; + case MD_MARK_EMPH_MOD3_1: return &ASTERISK_OPENERS_extraword_mod3_1; + case MD_MARK_EMPH_MOD3_2: return &ASTERISK_OPENERS_extraword_mod3_2; + default: MD_UNREACHABLE(); + } + return NULL; +} + +static MD_MARKCHAIN* +md_mark_chain(MD_CTX* ctx, int mark_index) +{ + MD_MARK* mark = &ctx->marks[mark_index]; + + switch(mark->ch) { + case _T('*'): return md_asterisk_chain(ctx, mark->flags); + case _T('_'): return &UNDERSCORE_OPENERS; + case _T('~'): return &TILDE_OPENERS; + case _T('['): return &BRACKET_OPENERS; + case _T('|'): return &TABLECELLBOUNDARIES; + default: return NULL; + } +} + +static MD_MARK* +md_push_mark(MD_CTX* ctx) +{ + if(ctx->n_marks >= ctx->alloc_marks) { + MD_MARK* new_marks; + + ctx->alloc_marks = (ctx->alloc_marks > 0 ? ctx->alloc_marks * 2 : 64); + new_marks = realloc(ctx->marks, ctx->alloc_marks * sizeof(MD_MARK)); + if(new_marks == NULL) { + MD_LOG("realloc() failed."); + return NULL; + } + + ctx->marks = new_marks; + } + + return &ctx->marks[ctx->n_marks++]; +} + +#define PUSH_MARK_() \ + do { \ + mark = md_push_mark(ctx); \ + if(mark == NULL) { \ + ret = -1; \ + goto abort; \ + } \ + } while(0) + +#define PUSH_MARK(ch_, beg_, end_, flags_) \ + do { \ + PUSH_MARK_(); \ + mark->beg = (beg_); \ + mark->end = (end_); \ + mark->prev = -1; \ + mark->next = -1; \ + mark->ch = (char)(ch_); \ + mark->flags = (flags_); \ + } while(0) + + +static void +md_mark_chain_append(MD_CTX* ctx, MD_MARKCHAIN* chain, int mark_index) +{ + if(chain->tail >= 0) + ctx->marks[chain->tail].next = mark_index; + else + chain->head = mark_index; + + ctx->marks[mark_index].prev = chain->tail; + chain->tail = mark_index; +} + +/* Sometimes, we need to store a pointer into the mark. It is quite rare + * so we do not bother to make MD_MARK use union, and it can only happen + * for dummy marks. */ +static inline void +md_mark_store_ptr(MD_CTX* ctx, int mark_index, void* ptr) +{ + MD_MARK* mark = &ctx->marks[mark_index]; + MD_ASSERT(mark->ch == 'D'); + + /* Check only members beg and end are misused for this. */ + MD_ASSERT(sizeof(void*) <= 2 * sizeof(OFF)); + memcpy(mark, &ptr, sizeof(void*)); +} + +static inline void* +md_mark_get_ptr(MD_CTX* ctx, int mark_index) +{ + void* ptr; + MD_MARK* mark = &ctx->marks[mark_index]; + MD_ASSERT(mark->ch == 'D'); + memcpy(&ptr, mark, sizeof(void*)); + return ptr; +} + +static void +md_resolve_range(MD_CTX* ctx, MD_MARKCHAIN* chain, int opener_index, int closer_index) +{ + MD_MARK* opener = &ctx->marks[opener_index]; + MD_MARK* closer = &ctx->marks[closer_index]; + + /* Remove opener from the list of openers. */ + if(chain != NULL) { + if(opener->prev >= 0) + ctx->marks[opener->prev].next = opener->next; + else + chain->head = opener->next; + + if(opener->next >= 0) + ctx->marks[opener->next].prev = opener->prev; + else + chain->tail = opener->prev; + } + + /* Interconnect opener and closer and mark both as resolved. */ + opener->next = closer_index; + opener->flags |= MD_MARK_OPENER | MD_MARK_RESOLVED; + closer->prev = opener_index; + closer->flags |= MD_MARK_CLOSER | MD_MARK_RESOLVED; +} + + +#define MD_ROLLBACK_ALL 0 +#define MD_ROLLBACK_CROSSING 1 + +/* In the range ctx->marks[opener_index] ... [closer_index], undo some or all + * resolvings accordingly to these rules: + * + * (1) All openers BEFORE the range corresponding to any closer inside the + * range are un-resolved and they are re-added to their respective chains + * of unresolved openers. This ensures we can reuse the opener for closers + * AFTER the range. + * + * (2) If 'how' is MD_ROLLBACK_ALL, then ALL resolved marks inside the range + * are discarded. + * + * (3) If 'how' is MD_ROLLBACK_CROSSING, only closers with openers handled + * in (1) are discarded. I.e. pairs of openers and closers which are both + * inside the range are retained as well as any unpaired marks. + */ +static void +md_rollback(MD_CTX* ctx, int opener_index, int closer_index, int how) +{ + int i; + int mark_index; + + /* Cut all unresolved openers at the mark index. */ + for(i = OPENERS_CHAIN_FIRST; i < OPENERS_CHAIN_LAST+1; i++) { + MD_MARKCHAIN* chain = &ctx->mark_chains[i]; + + while(chain->tail >= opener_index) + chain->tail = ctx->marks[chain->tail].prev; + + if(chain->tail >= 0) + ctx->marks[chain->tail].next = -1; + else + chain->head = -1; + } + + /* Go backwards so that un-resolved openers are re-added into their + * respective chains, in the right order. */ + mark_index = closer_index - 1; + while(mark_index > opener_index) { + MD_MARK* mark = &ctx->marks[mark_index]; + int mark_flags = mark->flags; + int discard_flag = (how == MD_ROLLBACK_ALL); + + if(mark->flags & MD_MARK_CLOSER) { + int mark_opener_index = mark->prev; + + /* Undo opener BEFORE the range. */ + if(mark_opener_index < opener_index) { + MD_MARK* mark_opener = &ctx->marks[mark_opener_index]; + MD_MARKCHAIN* chain; + + mark_opener->flags &= ~(MD_MARK_OPENER | MD_MARK_CLOSER | MD_MARK_RESOLVED); + chain = md_mark_chain(ctx, opener_index); + if(chain != NULL) { + md_mark_chain_append(ctx, chain, mark_opener_index); + discard_flag = 1; + } + } + } + + /* And reset our flags. */ + if(discard_flag) + mark->flags &= ~(MD_MARK_OPENER | MD_MARK_CLOSER | MD_MARK_RESOLVED); + + /* Jump as far as we can over unresolved or non-interesting marks. */ + switch(how) { + case MD_ROLLBACK_CROSSING: + if((mark_flags & MD_MARK_CLOSER) && mark->prev > opener_index) { + /* If we are closer with opener INSIDE the range, there may + * not be any other crosser inside the subrange. */ + mark_index = mark->prev; + break; + } + /* Pass through. */ + default: + mark_index--; + break; + } + } +} + +static void +md_build_mark_char_map(MD_CTX* ctx) +{ + memset(ctx->mark_char_map, 0, sizeof(ctx->mark_char_map)); + + ctx->mark_char_map['\\'] = 1; + ctx->mark_char_map['*'] = 1; + ctx->mark_char_map['_'] = 1; + ctx->mark_char_map['`'] = 1; + ctx->mark_char_map['&'] = 1; + ctx->mark_char_map[';'] = 1; + ctx->mark_char_map['<'] = 1; + ctx->mark_char_map['>'] = 1; + ctx->mark_char_map['['] = 1; + ctx->mark_char_map['!'] = 1; + ctx->mark_char_map[']'] = 1; + ctx->mark_char_map['\0'] = 1; + + if(ctx->parser.flags & MD_FLAG_STRIKETHROUGH) + ctx->mark_char_map['~'] = 1; + + if(ctx->parser.flags & MD_FLAG_PERMISSIVEEMAILAUTOLINKS) + ctx->mark_char_map['@'] = 1; + + if(ctx->parser.flags & MD_FLAG_PERMISSIVEURLAUTOLINKS) + ctx->mark_char_map[':'] = 1; + + if(ctx->parser.flags & MD_FLAG_PERMISSIVEWWWAUTOLINKS) + ctx->mark_char_map['.'] = 1; + + if(ctx->parser.flags & MD_FLAG_TABLES) + ctx->mark_char_map['|'] = 1; + + if(ctx->parser.flags & MD_FLAG_COLLAPSEWHITESPACE) { + int i; + + for(i = 0; i < (int) sizeof(ctx->mark_char_map); i++) { + if(ISWHITESPACE_(i)) + ctx->mark_char_map[i] = 1; + } + } +} + +/* We limit code span marks to lower then 32 backticks. This solves the + * pathologic case of too many openers, each of different length: Their + * resolving would be then O(n^2). */ +#define CODESPAN_MARK_MAXLEN 32 + +static int +md_is_code_span(MD_CTX* ctx, const MD_LINE* lines, int n_lines, OFF beg, + OFF* p_opener_beg, OFF* p_opener_end, + OFF* p_closer_beg, OFF* p_closer_end, + OFF last_potential_closers[CODESPAN_MARK_MAXLEN], + int* p_reached_paragraph_end) +{ + OFF opener_beg = beg; + OFF opener_end; + OFF closer_beg; + OFF closer_end; + SZ mark_len; + OFF line_end; + int has_space_after_opener = FALSE; + int has_eol_after_opener = FALSE; + int has_space_before_closer = FALSE; + int has_eol_before_closer = FALSE; + int has_only_space = TRUE; + int line_index = 0; + + line_end = lines[0].end; + opener_end = opener_beg; + while(opener_end < line_end && CH(opener_end) == _T('`')) + opener_end++; + has_space_after_opener = (opener_end < line_end && CH(opener_end) == _T(' ')); + has_eol_after_opener = (opener_end == line_end); + + /* The caller needs to know end of the opening mark even if we fail. */ + *p_opener_end = opener_end; + + mark_len = opener_end - opener_beg; + if(mark_len > CODESPAN_MARK_MAXLEN) + return FALSE; + + /* Check whether we already know there is no closer of this length. + * If so, re-scan does no sense. This fixes issue #59. */ + if(last_potential_closers[mark_len-1] >= lines[n_lines-1].end || + (*p_reached_paragraph_end && last_potential_closers[mark_len-1] < opener_end)) + return FALSE; + + closer_beg = opener_end; + closer_end = opener_end; + + /* Find closer mark. */ + while(TRUE) { + while(closer_beg < line_end && CH(closer_beg) != _T('`')) { + if(CH(closer_beg) != _T(' ')) + has_only_space = FALSE; + closer_beg++; + } + closer_end = closer_beg; + while(closer_end < line_end && CH(closer_end) == _T('`')) + closer_end++; + + if(closer_end - closer_beg == mark_len) { + /* Success. */ + has_space_before_closer = (closer_beg > lines[line_index].beg && CH(closer_beg-1) == _T(' ')); + has_eol_before_closer = (closer_beg == lines[line_index].beg); + break; + } + + if(closer_end - closer_beg > 0) { + /* We have found a back-tick which is not part of the closer. */ + has_only_space = FALSE; + + /* But if we eventually fail, remember it as a potential closer + * of its own length for future attempts. This mitigates needs for + * rescans. */ + if(closer_end - closer_beg < CODESPAN_MARK_MAXLEN) { + if(closer_beg > last_potential_closers[closer_end - closer_beg - 1]) + last_potential_closers[closer_end - closer_beg - 1] = closer_beg; + } + } + + if(closer_end >= line_end) { + line_index++; + if(line_index >= n_lines) { + /* Reached end of the paragraph and still nothing. */ + *p_reached_paragraph_end = TRUE; + return FALSE; + } + /* Try on the next line. */ + line_end = lines[line_index].end; + closer_beg = lines[line_index].beg; + } else { + closer_beg = closer_end; + } + } + + /* If there is a space or a new line both after and before the opener + * (and if the code span is not made of spaces only), consume one initial + * and one trailing space as part of the marks. */ + if(!has_only_space && + (has_space_after_opener || has_eol_after_opener) && + (has_space_before_closer || has_eol_before_closer)) + { + if(has_space_after_opener) + opener_end++; + else + opener_end = lines[1].beg; + + if(has_space_before_closer) + closer_beg--; + else { + closer_beg = lines[line_index-1].end; + /* We need to eat the preceding "\r\n" but not any line trailing + * spaces. */ + while(closer_beg < ctx->size && ISBLANK(closer_beg)) + closer_beg++; + } + } + + *p_opener_beg = opener_beg; + *p_opener_end = opener_end; + *p_closer_beg = closer_beg; + *p_closer_end = closer_end; + return TRUE; +} + +static int +md_is_autolink_uri(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg+1; + + MD_ASSERT(CH(beg) == _T('<')); + + /* Check for scheme. */ + if(off >= max_end || !ISASCII(off)) + return FALSE; + off++; + while(1) { + if(off >= max_end) + return FALSE; + if(off - beg > 32) + return FALSE; + if(CH(off) == _T(':') && off - beg >= 3) + break; + if(!ISALNUM(off) && CH(off) != _T('+') && CH(off) != _T('-') && CH(off) != _T('.')) + return FALSE; + off++; + } + + /* Check the path after the scheme. */ + while(off < max_end && CH(off) != _T('>')) { + if(ISWHITESPACE(off) || ISCNTRL(off) || CH(off) == _T('<')) + return FALSE; + off++; + } + + if(off >= max_end) + return FALSE; + + MD_ASSERT(CH(off) == _T('>')); + *p_end = off+1; + return TRUE; +} + +static int +md_is_autolink_email(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end) +{ + OFF off = beg + 1; + int label_len; + + MD_ASSERT(CH(beg) == _T('<')); + + /* The code should correspond to this regexp: + /^[a-zA-Z0-9.!#$%&'*+\/=?^_`{|}~-]+ + @[a-zA-Z0-9](?:[a-zA-Z0-9-]{0,61}[a-zA-Z0-9])? + (?:\.[a-zA-Z0-9](?:[a-zA-Z0-9-]{0,61}[a-zA-Z0-9])?)*$/ + */ + + /* Username (before '@'). */ + while(off < max_end && (ISALNUM(off) || ISANYOF(off, _T(".!#$%&'*+/=?^_`{|}~-")))) + off++; + if(off <= beg+1) + return FALSE; + + /* '@' */ + if(off >= max_end || CH(off) != _T('@')) + return FALSE; + off++; + + /* Labels delimited with '.'; each label is sequence of 1 - 62 alnum + * characters or '-', but '-' is not allowed as first or last char. */ + label_len = 0; + while(off < max_end) { + if(ISALNUM(off)) + label_len++; + else if(CH(off) == _T('-') && label_len > 0) + label_len++; + else if(CH(off) == _T('.') && label_len > 0 && CH(off-1) != _T('-')) + label_len = 0; + else + break; + + if(label_len > 62) + return FALSE; + + off++; + } + + if(label_len <= 0 || off >= max_end || CH(off) != _T('>') || CH(off-1) == _T('-')) + return FALSE; + + *p_end = off+1; + return TRUE; +} + +static int +md_is_autolink(MD_CTX* ctx, OFF beg, OFF max_end, OFF* p_end, int* p_missing_mailto) +{ + if(md_is_autolink_uri(ctx, beg, max_end, p_end)) { + *p_missing_mailto = FALSE; + return TRUE; + } + + if(md_is_autolink_email(ctx, beg, max_end, p_end)) { + *p_missing_mailto = TRUE; + return TRUE; + } + + return FALSE; +} + +static int +md_collect_marks(MD_CTX* ctx, const MD_LINE* lines, int n_lines, int table_mode) +{ + int i; + int ret = 0; + MD_MARK* mark; + OFF codespan_last_potential_closers[CODESPAN_MARK_MAXLEN] = { 0 }; + int codespan_scanned_till_paragraph_end = FALSE; + + for(i = 0; i < n_lines; i++) { + const MD_LINE* line = &lines[i]; + OFF off = line->beg; + OFF line_end = line->end; + + while(TRUE) { + CHAR ch; + +#ifdef MD4C_USE_UTF16 + /* For UTF-16, mark_char_map[] covers only ASCII. */ + #define IS_MARK_CHAR(off) ((CH(off) < SIZEOF_ARRAY(ctx->mark_char_map)) && \ + (ctx->mark_char_map[(unsigned char) CH(off)])) +#else + /* For 8-bit encodings, mark_char_map[] covers all 256 elements. */ + #define IS_MARK_CHAR(off) (ctx->mark_char_map[(unsigned char) CH(off)]) +#endif + + /* Optimization: Use some loop unrolling. */ + while(off + 3 < line_end && !IS_MARK_CHAR(off+0) && !IS_MARK_CHAR(off+1) + && !IS_MARK_CHAR(off+2) && !IS_MARK_CHAR(off+3)) + off += 4; + while(off < line_end && !IS_MARK_CHAR(off+0)) + off++; + + if(off >= line_end) + break; + + ch = CH(off); + + /* A backslash escape. + * It can go beyond line->end as it may involve escaped new + * line to form a hard break. */ + if(ch == _T('\\') && off+1 < ctx->size && (ISPUNCT(off+1) || ISNEWLINE(off+1))) { + /* Hard-break cannot be on the last line of the block. */ + if(!ISNEWLINE(off+1) || i+1 < n_lines) + PUSH_MARK(ch, off, off+2, MD_MARK_RESOLVED); + off += 2; + continue; + } + + /* A potential (string) emphasis start/end. */ + if(ch == _T('*') || ch == _T('_')) { + OFF tmp = off+1; + int left_level; /* What precedes: 0 = whitespace; 1 = punctuation; 2 = other char. */ + int right_level; /* What follows: 0 = whitespace; 1 = punctuation; 2 = other char. */ + + while(tmp < line_end && CH(tmp) == ch) + tmp++; + + if(off == line->beg || ISUNICODEWHITESPACEBEFORE(off)) + left_level = 0; + else if(ISUNICODEPUNCTBEFORE(off)) + left_level = 1; + else + left_level = 2; + + if(tmp == line_end || ISUNICODEWHITESPACE(tmp)) + right_level = 0; + else if(ISUNICODEPUNCT(tmp)) + right_level = 1; + else + right_level = 2; + + /* Intra-word underscore doesn't have special meaning. */ + if(ch == _T('_') && left_level == 2 && right_level == 2) { + left_level = 0; + right_level = 0; + } + + if(left_level != 0 || right_level != 0) { + unsigned flags = 0; + + if(left_level > 0 && left_level >= right_level) + flags |= MD_MARK_POTENTIAL_CLOSER; + if(right_level > 0 && right_level >= left_level) + flags |= MD_MARK_POTENTIAL_OPENER; + if(left_level == 2 && right_level == 2) + flags |= MD_MARK_EMPH_INTRAWORD; + + /* For "the rule of three" we need to remember the original + * size of the mark (modulo three), before we potentially + * split the mark when being later resolved partially by some + * shorter closer. */ + switch((tmp - off) % 3) { + case 0: flags |= MD_MARK_EMPH_MOD3_0; break; + case 1: flags |= MD_MARK_EMPH_MOD3_1; break; + case 2: flags |= MD_MARK_EMPH_MOD3_2; break; + } + + PUSH_MARK(ch, off, tmp, flags); + + /* During resolving, multiple asterisks may have to be + * split into independent span start/ends. Consider e.g. + * "**foo* bar*". Therefore we push also some empty dummy + * marks to have enough space for that. */ + off++; + while(off < tmp) { + PUSH_MARK('D', off, off, 0); + off++; + } + continue; + } + + off = tmp; + continue; + } + + /* A potential code span start/end. */ + if(ch == _T('`')) { + OFF opener_beg, opener_end; + OFF closer_beg, closer_end; + int is_code_span; + + is_code_span = md_is_code_span(ctx, lines + i, n_lines - i, off, + &opener_beg, &opener_end, &closer_beg, &closer_end, + codespan_last_potential_closers, + &codespan_scanned_till_paragraph_end); + if(is_code_span) { + PUSH_MARK(_T('`'), opener_beg, opener_end, MD_MARK_OPENER | MD_MARK_RESOLVED); + PUSH_MARK(_T('`'), closer_beg, closer_end, MD_MARK_CLOSER | MD_MARK_RESOLVED); + ctx->marks[ctx->n_marks-2].next = ctx->n_marks-1; + ctx->marks[ctx->n_marks-1].prev = ctx->n_marks-2; + + off = closer_end; + + /* Advance the current line accordingly. */ + while(off > line_end) { + i++; + line++; + line_end = line->end; + } + continue; + } + + off = opener_end; + continue; + } + + /* A potential entity start. */ + if(ch == _T('&')) { + PUSH_MARK(ch, off, off+1, MD_MARK_POTENTIAL_OPENER); + off++; + continue; + } + + /* A potential entity end. */ + if(ch == _T(';')) { + /* We surely cannot be entity unless the previous mark is '&'. */ + if(ctx->n_marks > 0 && ctx->marks[ctx->n_marks-1].ch == _T('&')) + PUSH_MARK(ch, off, off+1, MD_MARK_POTENTIAL_CLOSER); + + off++; + continue; + } + + /* A potential autolink or raw HTML start/end. */ + if(ch == _T('<')) { + int is_autolink; + OFF autolink_end; + int missing_mailto; + + if(!(ctx->parser.flags & MD_FLAG_NOHTMLSPANS)) { + int is_html; + OFF html_end; + + /* Given the nature of the raw HTML, we have to recognize + * it here. Doing so later in md_analyze_lt_gt() could + * open can of worms of quadratic complexity. */ + is_html = md_is_html_any(ctx, lines + i, n_lines - i, off, + lines[n_lines-1].end, &html_end); + if(is_html) { + PUSH_MARK(_T('<'), off, off, MD_MARK_OPENER | MD_MARK_RESOLVED); + PUSH_MARK(_T('>'), html_end, html_end, MD_MARK_CLOSER | MD_MARK_RESOLVED); + ctx->marks[ctx->n_marks-2].next = ctx->n_marks-1; + ctx->marks[ctx->n_marks-1].prev = ctx->n_marks-2; + off = html_end; + + /* Advance the current line accordingly. */ + while(off > line_end) { + i++; + line++; + line_end = line->end; + } + continue; + } + } + + is_autolink = md_is_autolink(ctx, off, lines[n_lines-1].end, + &autolink_end, &missing_mailto); + if(is_autolink) { + PUSH_MARK((missing_mailto ? _T('@') : _T('<')), off, off+1, + MD_MARK_OPENER | MD_MARK_RESOLVED | MD_MARK_AUTOLINK); + PUSH_MARK(_T('>'), autolink_end-1, autolink_end, + MD_MARK_CLOSER | MD_MARK_RESOLVED | MD_MARK_AUTOLINK); + ctx->marks[ctx->n_marks-2].next = ctx->n_marks-1; + ctx->marks[ctx->n_marks-1].prev = ctx->n_marks-2; + off = autolink_end; + continue; + } + + off++; + continue; + } + + /* A potential link or its part. */ + if(ch == _T('[') || (ch == _T('!') && off+1 < line_end && CH(off+1) == _T('['))) { + OFF tmp = (ch == _T('[') ? off+1 : off+2); + PUSH_MARK(ch, off, tmp, MD_MARK_POTENTIAL_OPENER); + off = tmp; + /* Two dummies to make enough place for data we need if it is + * a link. */ + PUSH_MARK('D', off, off, 0); + PUSH_MARK('D', off, off, 0); + continue; + } + if(ch == _T(']')) { + PUSH_MARK(ch, off, off+1, MD_MARK_POTENTIAL_CLOSER); + off++; + continue; + } + + /* A potential permissive e-mail autolink. */ + if(ch == _T('@')) { + if(line->beg + 1 <= off && ISALNUM(off-1) && + off + 3 < line->end && ISALNUM(off+1)) + { + PUSH_MARK(ch, off, off+1, MD_MARK_POTENTIAL_OPENER); + /* Push a dummy as a reserve for a closer. */ + PUSH_MARK('D', off, off, 0); + } + + off++; + continue; + } + + /* A potential permissive URL autolink. */ + if(ch == _T(':')) { + static struct { + const CHAR* scheme; + SZ scheme_size; + const CHAR* suffix; + SZ suffix_size; + } scheme_map[] = { + /* In the order from the most frequently used, arguably. */ + { _T("http"), 4, _T("//"), 2 }, + { _T("https"), 5, _T("//"), 2 }, + { _T("ftp"), 3, _T("//"), 2 } + }; + int scheme_index; + + for(scheme_index = 0; scheme_index < (int) SIZEOF_ARRAY(scheme_map); scheme_index++) { + const CHAR* scheme = scheme_map[scheme_index].scheme; + const SZ scheme_size = scheme_map[scheme_index].scheme_size; + const CHAR* suffix = scheme_map[scheme_index].suffix; + const SZ suffix_size = scheme_map[scheme_index].suffix_size; + + if(line->beg + scheme_size <= off && md_ascii_eq(STR(off-scheme_size), scheme, scheme_size) && + (line->beg + scheme_size == off || ISWHITESPACE(off-scheme_size-1) || ISANYOF(off-scheme_size-1, _T("*_~(["))) && + off + 1 + suffix_size < line->end && md_ascii_eq(STR(off+1), suffix, suffix_size)) + { + PUSH_MARK(ch, off-scheme_size, off+1+suffix_size, MD_MARK_POTENTIAL_OPENER); + /* Push a dummy as a reserve for a closer. */ + PUSH_MARK('D', off, off, 0); + off += 1 + suffix_size; + continue; + } + } + + off++; + continue; + } + + /* A potential permissive WWW autolink. */ + if(ch == _T('.')) { + if(line->beg + 3 <= off && md_ascii_eq(STR(off-3), _T("www"), 3) && + (line->beg + 3 == off || ISWHITESPACE(off-4) || ISANYOF(off-4, _T("*_~(["))) && + off + 1 < line_end) + { + PUSH_MARK(ch, off-3, off+1, MD_MARK_POTENTIAL_OPENER); + /* Push a dummy as a reserve for a closer. */ + PUSH_MARK('D', off, off, 0); + off++; + continue; + } + + off++; + continue; + } + + /* A potential table cell boundary. */ + if(table_mode && ch == _T('|')) { + PUSH_MARK(ch, off, off+1, 0); + off++; + continue; + } + + /* A potential strikethrough start/end. */ + if(ch == _T('~')) { + OFF tmp = off+1; + + while(tmp < line_end && CH(tmp) == _T('~')) + tmp++; + + PUSH_MARK(ch, off, tmp, MD_MARK_POTENTIAL_OPENER | MD_MARK_POTENTIAL_CLOSER); + off = tmp; + continue; + } + + /* Turn non-trivial whitespace into single space. */ + if(ISWHITESPACE_(ch)) { + OFF tmp = off+1; + + while(tmp < line_end && ISWHITESPACE(tmp)) + tmp++; + + if(tmp - off > 1 || ch != _T(' ')) + PUSH_MARK(ch, off, tmp, MD_MARK_RESOLVED); + + off = tmp; + continue; + } + + /* NULL character. */ + if(ch == _T('\0')) { + PUSH_MARK(ch, off, off+1, MD_MARK_RESOLVED); + off++; + continue; + } + + off++; + } + } + + /* Add a dummy mark at the end of the mark vector to simplify + * process_inlines(). */ + PUSH_MARK(127, ctx->size, ctx->size, MD_MARK_RESOLVED); + +abort: + return ret; +} + +static void +md_analyze_bracket(MD_CTX* ctx, int mark_index) +{ + /* We cannot really resolve links here as for that we would need + * more context. E.g. a following pair of brackets (reference link), + * or enclosing pair of brackets (if the inner is the link, the outer + * one cannot be.) + * + * Therefore we here only construct a list of resolved '[' ']' pairs + * ordered by position of the closer. This allows ur to analyze what is + * or is not link in the right order, from inside to outside in case + * of nested brackets. + * + * The resolving itself is deferred into md_resolve_links(). + */ + + MD_MARK* mark = &ctx->marks[mark_index]; + + if(mark->flags & MD_MARK_POTENTIAL_OPENER) { + md_mark_chain_append(ctx, &BRACKET_OPENERS, mark_index); + return; + } + + if(BRACKET_OPENERS.tail >= 0) { + /* Pop the opener from the chain. */ + int opener_index = BRACKET_OPENERS.tail; + MD_MARK* opener = &ctx->marks[opener_index]; + if(opener->prev >= 0) + ctx->marks[opener->prev].next = -1; + else + BRACKET_OPENERS.head = -1; + BRACKET_OPENERS.tail = opener->prev; + + /* Interconnect the opener and closer. */ + opener->next = mark_index; + mark->prev = opener_index; + + /* Add the pair into chain of potential links for md_resolve_links(). + * Note we misuse opener->prev for this as opener->next points to its + * closer. */ + if(ctx->unresolved_link_tail >= 0) + ctx->marks[ctx->unresolved_link_tail].prev = opener_index; + else + ctx->unresolved_link_head = opener_index; + ctx->unresolved_link_tail = opener_index; + opener->prev = -1; + } +} + +/* Forward declaration. */ +static void md_analyze_link_contents(MD_CTX* ctx, const MD_LINE* lines, int n_lines, + int mark_beg, int mark_end); + +static int +md_resolve_links(MD_CTX* ctx, const MD_LINE* lines, int n_lines) +{ + int opener_index = ctx->unresolved_link_head; + OFF last_link_beg = 0; + OFF last_link_end = 0; + OFF last_img_beg = 0; + OFF last_img_end = 0; + + while(opener_index >= 0) { + MD_MARK* opener = &ctx->marks[opener_index]; + int closer_index = opener->next; + MD_MARK* closer = &ctx->marks[closer_index]; + int next_index = opener->prev; + MD_MARK* next_opener; + MD_MARK* next_closer; + MD_LINK_ATTR attr; + int is_link = FALSE; + + if(next_index >= 0) { + next_opener = &ctx->marks[next_index]; + next_closer = &ctx->marks[next_opener->next]; + } else { + next_opener = NULL; + next_closer = NULL; + } + + /* If nested ("[ [ ] ]"), we need to make sure that: + * - The outer does not end inside of (...) belonging to the inner. + * - The outer cannot be link if the inner is link (i.e. not image). + * + * (Note we here analyze from inner to outer as the marks are ordered + * by closer->beg.) + */ + if((opener->beg < last_link_beg && closer->end < last_link_end) || + (opener->beg < last_img_beg && closer->end < last_img_end) || + (opener->beg < last_link_end && opener->ch == '[')) + { + opener_index = next_index; + continue; + } + + if(next_opener != NULL && next_opener->beg == closer->end) { + if(next_closer->beg > closer->end + 1) { + /* Might be full reference link. */ + is_link = md_is_link_reference(ctx, lines, n_lines, next_opener->beg, next_closer->end, &attr); + } else { + /* Might be shortcut reference link. */ + is_link = md_is_link_reference(ctx, lines, n_lines, opener->beg, closer->end, &attr); + } + + if(is_link < 0) + return -1; + + if(is_link) { + /* Eat the 2nd "[...]". */ + closer->end = next_closer->end; + } + } else { + if(closer->end < ctx->size && CH(closer->end) == _T('(')) { + /* Might be inline link. */ + OFF inline_link_end = UINT_MAX; + + is_link = md_is_inline_link_spec(ctx, lines, n_lines, closer->end, &inline_link_end, &attr); + if(is_link < 0) + return -1; + + /* Check the closing ')' is not inside an already resolved range + * (i.e. a range with a higher priority), e.g. a code span. */ + if(is_link) { + int i = closer_index + 1; + + while(i < ctx->n_marks) { + MD_MARK* mark = &ctx->marks[i]; + + if(mark->beg >= inline_link_end) + break; + if((mark->flags & (MD_MARK_OPENER | MD_MARK_RESOLVED)) == (MD_MARK_OPENER | MD_MARK_RESOLVED)) { + if(ctx->marks[mark->next].beg >= inline_link_end) { + /* Cancel the link status. */ + if(attr.title_needs_free) + free(attr.title); + is_link = FALSE; + break; + } + + i = mark->next + 1; + } else { + i++; + } + } + } + + if(is_link) { + /* Eat the "(...)" */ + closer->end = inline_link_end; + } + } + + if(!is_link) { + /* Might be collapsed reference link. */ + is_link = md_is_link_reference(ctx, lines, n_lines, opener->beg, closer->end, &attr); + if(is_link < 0) + return -1; + } + } + + if(is_link) { + /* Resolve the brackets as a link. */ + opener->flags |= MD_MARK_OPENER | MD_MARK_RESOLVED; + closer->flags |= MD_MARK_CLOSER | MD_MARK_RESOLVED; + + /* If it is a link, we store the destination and title in the two + * dummy marks after the opener. */ + MD_ASSERT(ctx->marks[opener_index+1].ch == 'D'); + ctx->marks[opener_index+1].beg = attr.dest_beg; + ctx->marks[opener_index+1].end = attr.dest_end; + + MD_ASSERT(ctx->marks[opener_index+2].ch == 'D'); + md_mark_store_ptr(ctx, opener_index+2, attr.title); + if(attr.title_needs_free) + md_mark_chain_append(ctx, &PTR_CHAIN, opener_index+2); + ctx->marks[opener_index+2].prev = attr.title_size; + + if(opener->ch == '[') { + last_link_beg = opener->beg; + last_link_end = closer->end; + } else { + last_img_beg = opener->beg; + last_img_end = closer->end; + } + + md_analyze_link_contents(ctx, lines, n_lines, opener_index+1, closer_index); + } + + opener_index = next_index; + } + + return 0; +} + +/* Analyze whether the mark '&' starts a HTML entity. + * If so, update its flags as well as flags of corresponding closer ';'. */ +static void +md_analyze_entity(MD_CTX* ctx, int mark_index) +{ + MD_MARK* opener = &ctx->marks[mark_index]; + MD_MARK* closer; + OFF off; + + /* Cannot be entity if there is no closer as the next mark. + * (Any other mark between would mean strange character which cannot be + * part of the entity. + * + * So we can do all the work on '&' and do not call this later for the + * closing mark ';'. + */ + if(mark_index + 1 >= ctx->n_marks) + return; + closer = &ctx->marks[mark_index+1]; + if(closer->ch != ';') + return; + + if(md_is_entity(ctx, opener->beg, closer->end, &off)) { + MD_ASSERT(off == closer->end); + + md_resolve_range(ctx, NULL, mark_index, mark_index+1); + opener->end = closer->end; + } +} + +static void +md_analyze_table_cell_boundary(MD_CTX* ctx, int mark_index) +{ + MD_MARK* mark = &ctx->marks[mark_index]; + mark->flags |= MD_MARK_RESOLVED; + + md_mark_chain_append(ctx, &TABLECELLBOUNDARIES, mark_index); + ctx->n_table_cell_boundaries++; +} + +/* Split a longer mark into two. The new mark takes the given count of + * characters. May only be called if an adequate number of dummy 'D' marks + * follows. + */ +static int +md_split_emph_mark(MD_CTX* ctx, int mark_index, SZ n) +{ + MD_MARK* mark = &ctx->marks[mark_index]; + int new_mark_index = mark_index + (mark->end - mark->beg - n); + MD_MARK* dummy = &ctx->marks[new_mark_index]; + + MD_ASSERT(mark->end - mark->beg > n); + MD_ASSERT(dummy->ch == 'D'); + + memcpy(dummy, mark, sizeof(MD_MARK)); + mark->end -= n; + dummy->beg = mark->end; + + return new_mark_index; +} + +static void +md_analyze_emph(MD_CTX* ctx, int mark_index) +{ + MD_MARK* mark = &ctx->marks[mark_index]; + MD_MARKCHAIN* chain = md_mark_chain(ctx, mark_index); + + /* If we can be a closer, try to resolve with the preceding opener. */ + if(mark->flags & MD_MARK_POTENTIAL_CLOSER) { + MD_MARK* opener = NULL; + int opener_index; + + if(mark->ch == _T('*')) { + MD_MARKCHAIN* opener_chains[6]; + int i, n_opener_chains; + unsigned flags = mark->flags; + + /* Apply "rule of three". (This is why we break asterisk opener + * marks into multiple chains.) */ + n_opener_chains = 0; + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_intraword_mod3_0; + if((flags & MD_MARK_EMPH_MOD3_MASK) != MD_MARK_EMPH_MOD3_2) + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_intraword_mod3_1; + if((flags & MD_MARK_EMPH_MOD3_MASK) != MD_MARK_EMPH_MOD3_1) + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_intraword_mod3_2; + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_extraword_mod3_0; + if(!(flags & MD_MARK_EMPH_INTRAWORD) || (flags & MD_MARK_EMPH_MOD3_MASK) != MD_MARK_EMPH_MOD3_2) + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_extraword_mod3_1; + if(!(flags & MD_MARK_EMPH_INTRAWORD) || (flags & MD_MARK_EMPH_MOD3_MASK) != MD_MARK_EMPH_MOD3_1) + opener_chains[n_opener_chains++] = &ASTERISK_OPENERS_extraword_mod3_2; + + /* Opener is the most recent mark from the allowed chains. */ + for(i = 0; i < n_opener_chains; i++) { + if(opener_chains[i]->tail >= 0) { + int tmp_index = opener_chains[i]->tail; + MD_MARK* tmp_mark = &ctx->marks[tmp_index]; + if(opener == NULL || tmp_mark->end > opener->end) { + opener_index = tmp_index; + opener = tmp_mark; + } + } + } + } else { + /* Simple emph. mark */ + if(chain->tail >= 0) { + opener_index = chain->tail; + opener = &ctx->marks[opener_index]; + } + } + + /* Resolve, if we have found matching opener. */ + if(opener != NULL) { + SZ opener_size = opener->end - opener->beg; + SZ closer_size = mark->end - mark->beg; + + if(opener_size > closer_size) { + opener_index = md_split_emph_mark(ctx, opener_index, closer_size); + md_mark_chain_append(ctx, md_mark_chain(ctx, opener_index), opener_index); + } else if(opener_size < closer_size) { + md_split_emph_mark(ctx, mark_index, closer_size - opener_size); + } + + md_rollback(ctx, opener_index, mark_index, MD_ROLLBACK_CROSSING); + md_resolve_range(ctx, chain, opener_index, mark_index); + return; + } + } + + /* If we could not resolve as closer, we may be yet be an opener. */ + if(mark->flags & MD_MARK_POTENTIAL_OPENER) + md_mark_chain_append(ctx, chain, mark_index); +} + +static void +md_analyze_tilde(MD_CTX* ctx, int mark_index) +{ + /* We attempt to be Github Flavored Markdown compatible here. GFM says + * that length of the tilde sequence is not important at all. Note that + * implies the TILDE_OPENERS chain can have at most one item. */ + + if(TILDE_OPENERS.head >= 0) { + /* The chain already contains an opener, so we may resolve the span. */ + int opener_index = TILDE_OPENERS.head; + + md_rollback(ctx, opener_index, mark_index, MD_ROLLBACK_CROSSING); + md_resolve_range(ctx, &TILDE_OPENERS, opener_index, mark_index); + } else { + /* We can only be opener. */ + md_mark_chain_append(ctx, &TILDE_OPENERS, mark_index); + } +} + +static void +md_analyze_permissive_url_autolink(MD_CTX* ctx, int mark_index) +{ + MD_MARK* opener = &ctx->marks[mark_index]; + int closer_index = mark_index + 1; + MD_MARK* closer = &ctx->marks[closer_index]; + MD_MARK* next_resolved_mark; + OFF off = opener->end; + int n_dots = FALSE; + int has_underscore_in_last_seg = FALSE; + int has_underscore_in_next_to_last_seg = FALSE; + int n_opened_parenthesis = 0; + + /* Check for domain. */ + while(off < ctx->size) { + if(ISALNUM(off) || CH(off) == _T('-')) { + off++; + } else if(CH(off) == _T('.')) { + /* We must see at least one period. */ + n_dots++; + has_underscore_in_next_to_last_seg = has_underscore_in_last_seg; + has_underscore_in_last_seg = FALSE; + off++; + } else if(CH(off) == _T('_')) { + /* No underscore may be present in the last two domain segments. */ + has_underscore_in_last_seg = TRUE; + off++; + } else { + break; + } + } + if(off > opener->end && CH(off-1) == _T('.')) { + off--; + n_dots--; + } + if(off <= opener->end || n_dots == 0 || has_underscore_in_next_to_last_seg || has_underscore_in_last_seg) + return; + + /* Check for path. */ + next_resolved_mark = closer + 1; + while(next_resolved_mark->ch == 'D' || !(next_resolved_mark->flags & MD_MARK_RESOLVED)) + next_resolved_mark++; + while(off < next_resolved_mark->beg && CH(off) != _T('<') && !ISWHITESPACE(off) && !ISNEWLINE(off)) { + /* Parenthesis must be balanced. */ + if(CH(off) == _T('(')) { + n_opened_parenthesis++; + } else if(CH(off) == _T(')')) { + if(n_opened_parenthesis > 0) + n_opened_parenthesis--; + else + break; + } + + off++; + } + /* These cannot be last char In such case they are more likely normal + * punctuation. */ + if(ISANYOF(off-1, _T("?!.,:*_~"))) + off--; + + /* Ok. Lets call it auto-link. Adapt opener and create closer to zero + * length so all the contents becomes the link text. */ + MD_ASSERT(closer->ch == 'D'); + opener->end = opener->beg; + closer->ch = opener->ch; + closer->beg = off; + closer->end = off; + md_resolve_range(ctx, NULL, mark_index, closer_index); +} + +/* The permissive autolinks do not have to be enclosed in '<' '>' but we + * instead impose stricter rules what is understood as an e-mail address + * here. Actually any non-alphanumeric characters with exception of '.' + * are prohibited both in username and after '@'. */ +static void +md_analyze_permissive_email_autolink(MD_CTX* ctx, int mark_index) +{ + MD_MARK* opener = &ctx->marks[mark_index]; + int closer_index; + MD_MARK* closer; + OFF beg = opener->beg; + OFF end = opener->end; + int dot_count = 0; + + MD_ASSERT(CH(beg) == _T('@')); + + /* Scan for name before '@'. */ + while(beg > 0 && (ISALNUM(beg-1) || ISANYOF(beg-1, _T(".-_+")))) + beg--; + + /* Scan for domain after '@'. */ + while(end < ctx->size && (ISALNUM(end) || ISANYOF(end, _T(".-_")))) { + if(CH(end) == _T('.')) + dot_count++; + end++; + } + if(CH(end-1) == _T('.')) { /* Final '.' not part of it. */ + dot_count--; + end--; + } + else if(ISANYOF2(end-1, _T('-'), _T('_'))) /* These are forbidden at the end. */ + return; + if(CH(end-1) == _T('@') || dot_count == 0) + return; + + /* Ok. Lets call it auto-link. Adapt opener and create closer to zero + * length so all the contents becomes the link text. */ + closer_index = mark_index + 1; + closer = &ctx->marks[closer_index]; + MD_ASSERT(closer->ch == 'D'); + + opener->beg = beg; + opener->end = beg; + closer->ch = opener->ch; + closer->beg = end; + closer->end = end; + md_resolve_range(ctx, NULL, mark_index, closer_index); +} + +static inline void +md_analyze_marks(MD_CTX* ctx, const MD_LINE* lines, int n_lines, + int mark_beg, int mark_end, const CHAR* mark_chars) +{ + int i = mark_beg; + + while(i < mark_end) { + MD_MARK* mark = &ctx->marks[i]; + + /* Skip resolved spans. */ + if(mark->flags & MD_MARK_RESOLVED) { + if(mark->flags & MD_MARK_OPENER) { + MD_ASSERT(i < mark->next); + i = mark->next + 1; + } else { + i++; + } + continue; + } + + /* Skip marks we do not want to deal with. */ + if(!ISANYOF_(mark->ch, mark_chars)) { + i++; + continue; + } + + /* Analyze the mark. */ + switch(mark->ch) { + case '[': /* Pass through. */ + case '!': /* Pass through. */ + case ']': md_analyze_bracket(ctx, i); break; + case '&': md_analyze_entity(ctx, i); break; + case '|': md_analyze_table_cell_boundary(ctx, i); break; + case '_': /* Pass through. */ + case '*': md_analyze_emph(ctx, i); break; + case '~': md_analyze_tilde(ctx, i); break; + case '.': /* Pass through. */ + case ':': md_analyze_permissive_url_autolink(ctx, i); break; + case '@': md_analyze_permissive_email_autolink(ctx, i); break; + } + + i++; + } +} + +/* Analyze marks (build ctx->marks). */ +static int +md_analyze_inlines(MD_CTX* ctx, const MD_LINE* lines, int n_lines, int table_mode) +{ + int ret; + + /* Reset the previously collected stack of marks. */ + ctx->n_marks = 0; + + /* Collect all marks. */ + MD_CHECK(md_collect_marks(ctx, lines, n_lines, table_mode)); + + /* We analyze marks in few groups to handle their precedence. */ + /* (1) Entities; code spans; autolinks; raw HTML. */ + md_analyze_marks(ctx, lines, n_lines, 0, ctx->n_marks, _T("&")); + + if(table_mode) { + /* (2) Analyze table cell boundaries. + * Note we reset TABLECELLBOUNDARIES chain prior to the call md_analyze_marks(), + * not after, because caller may need it. */ + MD_ASSERT(n_lines == 1); + TABLECELLBOUNDARIES.head = -1; + TABLECELLBOUNDARIES.tail = -1; + ctx->n_table_cell_boundaries = 0; + md_analyze_marks(ctx, lines, n_lines, 0, ctx->n_marks, _T("|")); + return ret; + } + + /* (3) Links. */ + md_analyze_marks(ctx, lines, n_lines, 0, ctx->n_marks, _T("[]!")); + MD_CHECK(md_resolve_links(ctx, lines, n_lines)); + BRACKET_OPENERS.head = -1; + BRACKET_OPENERS.tail = -1; + ctx->unresolved_link_head = -1; + ctx->unresolved_link_tail = -1; + + /* (4) Emphasis and strong emphasis; permissive autolinks. */ + md_analyze_link_contents(ctx, lines, n_lines, 0, ctx->n_marks); + +abort: + return ret; +} + +static void +md_analyze_link_contents(MD_CTX* ctx, const MD_LINE* lines, int n_lines, + int mark_beg, int mark_end) +{ + md_analyze_marks(ctx, lines, n_lines, mark_beg, mark_end, _T("*_~@:.")); + ASTERISK_OPENERS_extraword_mod3_0.head = -1; + ASTERISK_OPENERS_extraword_mod3_0.tail = -1; + ASTERISK_OPENERS_extraword_mod3_1.head = -1; + ASTERISK_OPENERS_extraword_mod3_1.tail = -1; + ASTERISK_OPENERS_extraword_mod3_2.head = -1; + ASTERISK_OPENERS_extraword_mod3_2.tail = -1; + ASTERISK_OPENERS_intraword_mod3_0.head = -1; + ASTERISK_OPENERS_intraword_mod3_0.tail = -1; + ASTERISK_OPENERS_intraword_mod3_1.head = -1; + ASTERISK_OPENERS_intraword_mod3_1.tail = -1; + ASTERISK_OPENERS_intraword_mod3_2.head = -1; + ASTERISK_OPENERS_intraword_mod3_2.tail = -1; + UNDERSCORE_OPENERS.head = -1; + UNDERSCORE_OPENERS.tail = -1; + TILDE_OPENERS.head = -1; + TILDE_OPENERS.tail = -1; +} + +static int +md_enter_leave_span_a(MD_CTX* ctx, int enter, MD_SPANTYPE type, + const CHAR* dest, SZ dest_size, int prohibit_escapes_in_dest, + const CHAR* title, SZ title_size) +{ + MD_ATTRIBUTE_BUILD href_build = { 0 }; + MD_ATTRIBUTE_BUILD title_build = { 0 }; + MD_SPAN_A_DETAIL det; + int ret = 0; + + /* Note we here rely on fact that MD_SPAN_A_DETAIL and + * MD_SPAN_IMG_DETAIL are binary-compatible. */ + memset(&det, 0, sizeof(MD_SPAN_A_DETAIL)); + MD_CHECK(md_build_attribute(ctx, dest, dest_size, + (prohibit_escapes_in_dest ? MD_BUILD_ATTR_NO_ESCAPES : 0), + &det.href, &href_build)); + MD_CHECK(md_build_attribute(ctx, title, title_size, 0, &det.title, &title_build)); + + if(enter) + MD_ENTER_SPAN(type, &det); + else + MD_LEAVE_SPAN(type, &det); + +abort: + md_free_attribute(ctx, &href_build); + md_free_attribute(ctx, &title_build); + return ret; +} + +/* Render the output, accordingly to the analyzed ctx->marks. */ +static int +md_process_inlines(MD_CTX* ctx, const MD_LINE* lines, int n_lines) +{ + MD_TEXTTYPE text_type; + const MD_LINE* line = lines; + MD_MARK* prev_mark = NULL; + MD_MARK* mark; + OFF off = lines[0].beg; + OFF end = lines[n_lines-1].end; + int enforce_hardbreak = 0; + int ret = 0; + + /* Find first resolved mark. Note there is always at least one resolved + * mark, the dummy last one after the end of the latest line we actually + * never really reach. This saves us of a lot of special checks and cases + * in this function. */ + mark = ctx->marks; + while(!(mark->flags & MD_MARK_RESOLVED)) + mark++; + + text_type = MD_TEXT_NORMAL; + + while(1) { + /* Process the text up to the next mark or end-of-line. */ + OFF tmp = (line->end < mark->beg ? line->end : mark->beg); + if(tmp > off) { + MD_TEXT(text_type, STR(off), tmp - off); + off = tmp; + } + + /* If reached the mark, process it and move to next one. */ + if(off >= mark->beg) { + switch(mark->ch) { + case '\\': /* Backslash escape. */ + if(ISNEWLINE(mark->beg+1)) + enforce_hardbreak = 1; + else + MD_TEXT(text_type, STR(mark->beg+1), 1); + break; + + case ' ': /* Non-trivial space. */ + MD_TEXT(text_type, _T(" "), 1); + break; + + case '`': /* Code span. */ + if(mark->flags & MD_MARK_OPENER) { + MD_ENTER_SPAN(MD_SPAN_CODE, NULL); + text_type = MD_TEXT_CODE; + } else { + MD_LEAVE_SPAN(MD_SPAN_CODE, NULL); + text_type = MD_TEXT_NORMAL; + } + break; + + case '_': + case '*': /* Emphasis, strong emphasis. */ + if(mark->flags & MD_MARK_OPENER) { + if((mark->end - off) % 2) { + MD_ENTER_SPAN(MD_SPAN_EM, NULL); + off++; + } + while(off + 1 < mark->end) { + MD_ENTER_SPAN(MD_SPAN_STRONG, NULL); + off += 2; + } + } else { + while(off + 1 < mark->end) { + MD_LEAVE_SPAN(MD_SPAN_STRONG, NULL); + off += 2; + } + if((mark->end - off) % 2) { + MD_LEAVE_SPAN(MD_SPAN_EM, NULL); + off++; + } + } + break; + + case '~': + if(mark->flags & MD_MARK_OPENER) + MD_ENTER_SPAN(MD_SPAN_DEL, NULL); + else + MD_LEAVE_SPAN(MD_SPAN_DEL, NULL); + break; + + case '[': /* Link, image. */ + case '!': + case ']': + { + const MD_MARK* opener = (mark->ch != ']' ? mark : &ctx->marks[mark->prev]); + const MD_MARK* dest_mark = opener+1; + const MD_MARK* title_mark = opener+2; + + MD_ASSERT(dest_mark->ch == 'D'); + MD_ASSERT(title_mark->ch == 'D'); + + MD_CHECK(md_enter_leave_span_a(ctx, (mark->ch != ']'), + (opener->ch == '!' ? MD_SPAN_IMG : MD_SPAN_A), + STR(dest_mark->beg), dest_mark->end - dest_mark->beg, FALSE, + md_mark_get_ptr(ctx, title_mark - ctx->marks), title_mark->prev)); + + /* link/image closer may span multiple lines. */ + if(mark->ch == ']') { + while(mark->end > line->end) + line++; + } + + break; + } + + case '<': + case '>': /* Autolink or raw HTML. */ + if(!(mark->flags & MD_MARK_AUTOLINK)) { + /* Raw HTML. */ + if(mark->flags & MD_MARK_OPENER) + text_type = MD_TEXT_HTML; + else + text_type = MD_TEXT_NORMAL; + break; + } + /* Pass through, if auto-link. */ + + case '@': /* Permissive e-mail autolink. */ + case ':': /* Permissive URL autolink. */ + case '.': /* Permissive WWW autolink. */ + { + MD_MARK* opener = ((mark->flags & MD_MARK_OPENER) ? mark : &ctx->marks[mark->prev]); + MD_MARK* closer = &ctx->marks[opener->next]; + const CHAR* dest = STR(opener->end); + SZ dest_size = closer->beg - opener->end; + + /* For permissive auto-links we do not know closer mark + * position at the time of md_collect_marks(), therefore + * it can be out-of-order in ctx->marks[]. + * + * With this flag, we make sure that we output the closer + * only if we processed the opener. */ + if(mark->flags & MD_MARK_OPENER) + closer->flags |= MD_MARK_VALIDPERMISSIVEAUTOLINK; + + if(opener->ch == '@' || opener->ch == '.') { + dest_size += 7; + MD_TEMP_BUFFER(dest_size * sizeof(CHAR)); + memcpy(ctx->buffer, + (opener->ch == '@' ? _T("mailto:") : _T("http://")), + 7 * sizeof(CHAR)); + memcpy(ctx->buffer + 7, dest, (dest_size-7) * sizeof(CHAR)); + dest = ctx->buffer; + } + + if(closer->flags & MD_MARK_VALIDPERMISSIVEAUTOLINK) + MD_CHECK(md_enter_leave_span_a(ctx, (mark->flags & MD_MARK_OPENER), + MD_SPAN_A, dest, dest_size, TRUE, NULL, 0)); + break; + } + + case '&': /* Entity. */ + MD_TEXT(MD_TEXT_ENTITY, STR(mark->beg), mark->end - mark->beg); + break; + + case '\0': + MD_TEXT(MD_TEXT_NULLCHAR, _T(""), 1); + break; + + case 127: + goto abort; + } + + off = mark->end; + + /* Move to next resolved mark. */ + prev_mark = mark; + mark++; + while(!(mark->flags & MD_MARK_RESOLVED) || mark->beg < off) + mark++; + } + + /* If reached end of line, move to next one. */ + if(off >= line->end) { + /* If it is the last line, we are done. */ + if(off >= end) + break; + + if(text_type == MD_TEXT_CODE) { + OFF tmp; + + MD_ASSERT(prev_mark != NULL); + MD_ASSERT(prev_mark->ch == '`' && (prev_mark->flags & MD_MARK_OPENER)); + MD_ASSERT(mark->ch == '`' && (mark->flags & MD_MARK_CLOSER)); + + /* Inside a code span, trailing line whitespace has to be + * outputted. */ + tmp = off; + while(off < ctx->size && ISBLANK(off)) + off++; + if(off > tmp) + MD_TEXT(MD_TEXT_CODE, STR(tmp), off-tmp); + + /* and new lines are transformed into single spaces. */ + if(prev_mark->end < off && off < mark->beg) + MD_TEXT(MD_TEXT_CODE, _T(" "), 1); + + } else if(text_type == MD_TEXT_HTML) { + /* Inside raw HTML, we output the new line verbatim, including + * any trailing spaces. */ + OFF tmp = off; + + while(tmp < end && ISBLANK(tmp)) + tmp++; + if(tmp > off) + MD_TEXT(MD_TEXT_HTML, STR(off), tmp - off); + MD_TEXT(MD_TEXT_HTML, _T("\n"), 1); + } else { + /* Output soft or hard line break. */ + MD_TEXTTYPE break_type = MD_TEXT_SOFTBR; + + if(text_type == MD_TEXT_NORMAL) { + if(enforce_hardbreak) + break_type = MD_TEXT_BR; + else if((CH(line->end) == _T(' ') && CH(line->end+1) == _T(' '))) + break_type = MD_TEXT_BR; + } + + MD_TEXT(break_type, _T("\n"), 1); + } + + /* Move to the next line. */ + line++; + off = line->beg; + + enforce_hardbreak = 0; + } + } + +abort: + return ret; +} + + +/*************************** + *** Processing Tables *** + ***************************/ + +static void +md_analyze_table_alignment(MD_CTX* ctx, OFF beg, OFF end, MD_ALIGN* align, int n_align) +{ + static const MD_ALIGN align_map[] = { MD_ALIGN_DEFAULT, MD_ALIGN_LEFT, MD_ALIGN_RIGHT, MD_ALIGN_CENTER }; + OFF off = beg; + + while(n_align > 0) { + int index = 0; /* index into align_map[] */ + + while(CH(off) != _T('-')) + off++; + if(off > beg && CH(off-1) == _T(':')) + index |= 1; + while(off < end && CH(off) == _T('-')) + off++; + if(off < end && CH(off) == _T(':')) + index |= 2; + + *align = align_map[index]; + align++; + n_align--; + } + +} + +/* Forward declaration. */ +static int md_process_normal_block_contents(MD_CTX* ctx, const MD_LINE* lines, int n_lines); + +static int +md_process_table_cell(MD_CTX* ctx, MD_BLOCKTYPE cell_type, MD_ALIGN align, OFF beg, OFF end) +{ + MD_LINE line; + MD_BLOCK_TD_DETAIL det; + int ret = 0; + + while(beg < end && ISWHITESPACE(beg)) + beg++; + while(end > beg && ISWHITESPACE(end-1)) + end--; + + det.align = align; + line.beg = beg; + line.end = end; + + MD_ENTER_BLOCK(cell_type, &det); + MD_CHECK(md_process_normal_block_contents(ctx, &line, 1)); + MD_LEAVE_BLOCK(cell_type, &det); + +abort: + return ret; +} + +static int +md_process_table_row(MD_CTX* ctx, MD_BLOCKTYPE cell_type, OFF beg, OFF end, + const MD_ALIGN* align, int col_count) +{ + MD_LINE line; + OFF* pipe_offs = NULL; + int i, j, n; + int ret = 0; + + line.beg = beg; + line.end = end; + + /* Break the line into table cells by identifying pipe characters who + * form the cell boundary. */ + MD_CHECK(md_analyze_inlines(ctx, &line, 1, TRUE)); + + /* We have to remember the cell boundaries in local buffer because + * ctx->marks[] shall be reused during cell contents processing. */ + n = ctx->n_table_cell_boundaries; + pipe_offs = (OFF*) malloc(n * sizeof(OFF)); + if(pipe_offs == NULL) { + MD_LOG("malloc() failed."); + ret = -1; + goto abort; + } + for(i = TABLECELLBOUNDARIES.head, j = 0; i >= 0; i = ctx->marks[i].next) { + MD_MARK* mark = &ctx->marks[i]; + pipe_offs[j++] = mark->beg; + } + + /* Process cells. */ + MD_ENTER_BLOCK(MD_BLOCK_TR, NULL); + j = 0; + if(beg < pipe_offs[0] && j < col_count) + MD_CHECK(md_process_table_cell(ctx, cell_type, align[j++], beg, pipe_offs[0])); + for(i = 0; i < n-1 && j < col_count; i++) + MD_CHECK(md_process_table_cell(ctx, cell_type, align[j++], pipe_offs[i]+1, pipe_offs[i+1])); + if(pipe_offs[n-1] < end-1 && j < col_count) + MD_CHECK(md_process_table_cell(ctx, cell_type, align[j++], pipe_offs[n-1]+1, end)); + /* Make sure we call enough table cells even if the current table contains + * too few of them. */ + while(j < col_count) + MD_CHECK(md_process_table_cell(ctx, cell_type, align[j++], 0, 0)); + + MD_LEAVE_BLOCK(MD_BLOCK_TR, NULL); + +abort: + free(pipe_offs); + + /* Free any temporary memory blocks stored within some dummy marks. */ + for(i = PTR_CHAIN.head; i >= 0; i = ctx->marks[i].next) + free(md_mark_get_ptr(ctx, i)); + PTR_CHAIN.head = -1; + PTR_CHAIN.tail = -1; + + return ret; +} + +static int +md_process_table_block_contents(MD_CTX* ctx, int col_count, const MD_LINE* lines, int n_lines) +{ + MD_ALIGN* align; + int i; + int ret = 0; + + /* At least two lines have to be present: The column headers and the line + * with the underlines. */ + MD_ASSERT(n_lines >= 2); + + align = malloc(col_count * sizeof(MD_ALIGN)); + if(align == NULL) { + MD_LOG("malloc() failed."); + ret = -1; + goto abort; + } + + md_analyze_table_alignment(ctx, lines[1].beg, lines[1].end, align, col_count); + + MD_ENTER_BLOCK(MD_BLOCK_THEAD, NULL); + MD_CHECK(md_process_table_row(ctx, MD_BLOCK_TH, + lines[0].beg, lines[0].end, align, col_count)); + MD_LEAVE_BLOCK(MD_BLOCK_THEAD, NULL); + + MD_ENTER_BLOCK(MD_BLOCK_TBODY, NULL); + for(i = 2; i < n_lines; i++) { + MD_CHECK(md_process_table_row(ctx, MD_BLOCK_TD, + lines[i].beg, lines[i].end, align, col_count)); + } + MD_LEAVE_BLOCK(MD_BLOCK_TBODY, NULL); + +abort: + free(align); + return ret; +} + +static int +md_is_table_row(MD_CTX* ctx, OFF beg, OFF* p_end) +{ + MD_LINE line; + int i; + int ret = FALSE; + + line.beg = beg; + line.end = beg; + + /* Find end of line. */ + while(line.end < ctx->size && !ISNEWLINE(line.end)) + line.end++; + + MD_CHECK(md_analyze_inlines(ctx, &line, 1, TRUE)); + + if(TABLECELLBOUNDARIES.head >= 0) { + if(p_end != NULL) + *p_end = line.end; + ret = TRUE; + } + +abort: + /* Free any temporary memory blocks stored within some dummy marks. */ + for(i = PTR_CHAIN.head; i >= 0; i = ctx->marks[i].next) + free(md_mark_get_ptr(ctx, i)); + PTR_CHAIN.head = -1; + PTR_CHAIN.tail = -1; + + return ret; +} + + +/************************** + *** Processing Block *** + **************************/ + +#define MD_BLOCK_CONTAINER_OPENER 0x01 +#define MD_BLOCK_CONTAINER_CLOSER 0x02 +#define MD_BLOCK_CONTAINER (MD_BLOCK_CONTAINER_OPENER | MD_BLOCK_CONTAINER_CLOSER) +#define MD_BLOCK_LOOSE_LIST 0x04 +#define MD_BLOCK_SETEXT_HEADER 0x08 + +struct MD_BLOCK_tag { + MD_BLOCKTYPE type : 8; + unsigned flags : 8; + + /* MD_BLOCK_H: Header level (1 - 6) + * MD_BLOCK_CODE: Non-zero if fenced, zero if indented. + * MD_BLOCK_LI: Task mark character (0 if not task list item, 'x', 'X' or ' '). + * MD_BLOCK_TABLE: Column count (as determined by the table underline). + */ + unsigned data : 16; + + /* Leaf blocks: Count of lines (MD_LINE or MD_VERBATIMLINE) on the block. + * MD_BLOCK_LI: Task mark offset in the input doc. + * MD_BLOCK_OL: Start item number. + */ + unsigned n_lines; +}; + +struct MD_CONTAINER_tag { + CHAR ch; + unsigned is_loose : 8; + unsigned is_task : 8; + unsigned start; + unsigned mark_indent; + unsigned contents_indent; + OFF block_byte_off; + OFF task_mark_off; +}; + + +static int +md_process_normal_block_contents(MD_CTX* ctx, const MD_LINE* lines, int n_lines) +{ + int i; + int ret; + + MD_CHECK(md_analyze_inlines(ctx, lines, n_lines, FALSE)); + MD_CHECK(md_process_inlines(ctx, lines, n_lines)); + +abort: + /* Free any temporary memory blocks stored within some dummy marks. */ + for(i = PTR_CHAIN.head; i >= 0; i = ctx->marks[i].next) + free(md_mark_get_ptr(ctx, i)); + PTR_CHAIN.head = -1; + PTR_CHAIN.tail = -1; + + return ret; +} + +static int +md_process_verbatim_block_contents(MD_CTX* ctx, MD_TEXTTYPE text_type, const MD_VERBATIMLINE* lines, int n_lines) +{ + static const CHAR indent_chunk_str[] = _T(" "); + static const SZ indent_chunk_size = SIZEOF_ARRAY(indent_chunk_str) - 1; + + int i; + int ret = 0; + + for(i = 0; i < n_lines; i++) { + const MD_VERBATIMLINE* line = &lines[i]; + int indent = line->indent; + + MD_ASSERT(indent >= 0); + + /* Output code indentation. */ + while(indent > (int) SIZEOF_ARRAY(indent_chunk_str)) { + MD_TEXT(text_type, indent_chunk_str, indent_chunk_size); + indent -= SIZEOF_ARRAY(indent_chunk_str); + } + if(indent > 0) + MD_TEXT(text_type, indent_chunk_str, indent); + + /* Output the code line itself. */ + MD_TEXT_INSECURE(text_type, STR(line->beg), line->end - line->beg); + + /* Enforce end-of-line. */ + MD_TEXT(text_type, _T("\n"), 1); + } + +abort: + return ret; +} + +static int +md_process_code_block_contents(MD_CTX* ctx, int is_fenced, const MD_VERBATIMLINE* lines, int n_lines) +{ + if(is_fenced) { + /* Skip the first line in case of fenced code: It is the fence. + * (Only the starting fence is present due to logic in md_analyze_line().) */ + lines++; + n_lines--; + } else { + /* Ignore blank lines at start/end of indented code block. */ + while(n_lines > 0 && lines[0].beg == lines[0].end) { + lines++; + n_lines--; + } + while(n_lines > 0 && lines[n_lines-1].beg == lines[n_lines-1].end) { + n_lines--; + } + } + + if(n_lines == 0) + return 0; + + return md_process_verbatim_block_contents(ctx, MD_TEXT_CODE, lines, n_lines); +} + +static int +md_setup_fenced_code_detail(MD_CTX* ctx, const MD_BLOCK* block, MD_BLOCK_CODE_DETAIL* det, + MD_ATTRIBUTE_BUILD* info_build, MD_ATTRIBUTE_BUILD* lang_build) +{ + const MD_VERBATIMLINE* fence_line = (const MD_VERBATIMLINE*)(block + 1); + OFF beg = fence_line->beg; + OFF end = fence_line->end; + OFF lang_end; + CHAR fence_ch = CH(fence_line->beg); + int ret = 0; + + /* Skip the fence itself. */ + while(beg < ctx->size && CH(beg) == fence_ch) + beg++; + /* Trim initial spaces. */ + while(beg < ctx->size && CH(beg) == _T(' ')) + beg++; + + /* Trim trailing spaces. */ + while(end > beg && CH(end-1) == _T(' ')) + end--; + + /* Build info string attribute. */ + MD_CHECK(md_build_attribute(ctx, STR(beg), end - beg, 0, &det->info, info_build)); + + /* Build info string attribute. */ + lang_end = beg; + while(lang_end < end && !ISWHITESPACE(lang_end)) + lang_end++; + MD_CHECK(md_build_attribute(ctx, STR(beg), lang_end - beg, 0, &det->lang, lang_build)); + + det->fence_char = fence_ch; + +abort: + return ret; +} + +static int +md_process_leaf_block(MD_CTX* ctx, const MD_BLOCK* block) +{ + union { + MD_BLOCK_H_DETAIL header; + MD_BLOCK_CODE_DETAIL code; + } det; + MD_ATTRIBUTE_BUILD info_build; + MD_ATTRIBUTE_BUILD lang_build; + int is_in_tight_list; + int clean_fence_code_detail = FALSE; + int ret = 0; + + memset(&det, 0, sizeof(det)); + + if(ctx->n_containers == 0) + is_in_tight_list = FALSE; + else + is_in_tight_list = !ctx->containers[ctx->n_containers-1].is_loose; + + switch(block->type) { + case MD_BLOCK_H: + det.header.level = block->data; + break; + + case MD_BLOCK_CODE: + /* For fenced code block, we may need to set the info string. */ + if(block->data != 0) { + memset(&det.code, 0, sizeof(MD_BLOCK_CODE_DETAIL)); + clean_fence_code_detail = TRUE; + MD_CHECK(md_setup_fenced_code_detail(ctx, block, &det.code, &info_build, &lang_build)); + } + break; + + default: + /* Noop. */ + break; + } + + if(!is_in_tight_list || block->type != MD_BLOCK_P) + MD_ENTER_BLOCK(block->type, (void*) &det); + + /* Process the block contents accordingly to is type. */ + switch(block->type) { + case MD_BLOCK_HR: + /* noop */ + break; + + case MD_BLOCK_CODE: + MD_CHECK(md_process_code_block_contents(ctx, (block->data != 0), + (const MD_VERBATIMLINE*)(block + 1), block->n_lines)); + break; + + case MD_BLOCK_HTML: + MD_CHECK(md_process_verbatim_block_contents(ctx, MD_TEXT_HTML, + (const MD_VERBATIMLINE*)(block + 1), block->n_lines)); + break; + + case MD_BLOCK_TABLE: + MD_CHECK(md_process_table_block_contents(ctx, block->data, + (const MD_LINE*)(block + 1), block->n_lines)); + break; + + default: + MD_CHECK(md_process_normal_block_contents(ctx, + (const MD_LINE*)(block + 1), block->n_lines)); + break; + } + + if(!is_in_tight_list || block->type != MD_BLOCK_P) + MD_LEAVE_BLOCK(block->type, (void*) &det); + +abort: + if(clean_fence_code_detail) { + md_free_attribute(ctx, &info_build); + md_free_attribute(ctx, &lang_build); + } + return ret; +} + +static int +md_process_all_blocks(MD_CTX* ctx) +{ + int byte_off = 0; + int ret = 0; + + /* ctx->containers now is not needed for detection of lists and list items + * so we reuse it for tracking what lists are loose or tight. We rely + * on the fact the vector is large enough to hold the deepest nesting + * level of lists. */ + ctx->n_containers = 0; + + while(byte_off < ctx->n_block_bytes) { + MD_BLOCK* block = (MD_BLOCK*)((char*)ctx->block_bytes + byte_off); + union { + MD_BLOCK_UL_DETAIL ul; + MD_BLOCK_OL_DETAIL ol; + MD_BLOCK_LI_DETAIL li; + } det; + + switch(block->type) { + case MD_BLOCK_UL: + det.ul.is_tight = (block->flags & MD_BLOCK_LOOSE_LIST) ? FALSE : TRUE; + det.ul.mark = (CHAR) block->data; + break; + + case MD_BLOCK_OL: + det.ol.start = block->n_lines; + det.ol.is_tight = (block->flags & MD_BLOCK_LOOSE_LIST) ? FALSE : TRUE; + det.ol.mark_delimiter = (CHAR) block->data; + break; + + case MD_BLOCK_LI: + det.li.is_task = (block->data != 0); + det.li.task_mark = (CHAR) block->data; + det.li.task_mark_offset = (OFF) block->n_lines; + break; + + default: + /* noop */ + break; + } + + if(block->flags & MD_BLOCK_CONTAINER) { + if(block->flags & MD_BLOCK_CONTAINER_CLOSER) { + MD_LEAVE_BLOCK(block->type, &det); + + if(block->type == MD_BLOCK_UL || block->type == MD_BLOCK_OL || block->type == MD_BLOCK_QUOTE) + ctx->n_containers--; + } + + if(block->flags & MD_BLOCK_CONTAINER_OPENER) { + MD_ENTER_BLOCK(block->type, &det); + + if(block->type == MD_BLOCK_UL || block->type == MD_BLOCK_OL) { + ctx->containers[ctx->n_containers].is_loose = (block->flags & MD_BLOCK_LOOSE_LIST); + ctx->n_containers++; + } else if(block->type == MD_BLOCK_QUOTE) { + /* This causes that any text in a block quote, even if + * nested inside a tight list item, is wrapped with + * <p>...</p>. */ + ctx->containers[ctx->n_containers].is_loose = TRUE; + ctx->n_containers++; + } + } + } else { + MD_CHECK(md_process_leaf_block(ctx, block)); + + if(block->type == MD_BLOCK_CODE || block->type == MD_BLOCK_HTML) + byte_off += block->n_lines * sizeof(MD_VERBATIMLINE); + else + byte_off += block->n_lines * sizeof(MD_LINE); + } + + byte_off += sizeof(MD_BLOCK); + } + + ctx->n_block_bytes = 0; + +abort: + return ret; +} + + +/************************************ + *** Grouping Lines into Blocks *** + ************************************/ + +static void* +md_push_block_bytes(MD_CTX* ctx, int n_bytes) +{ + void* ptr; + + if(ctx->n_block_bytes + n_bytes > ctx->alloc_block_bytes) { + void* new_block_bytes; + + ctx->alloc_block_bytes = (ctx->alloc_block_bytes > 0 ? ctx->alloc_block_bytes * 2 : 512); + new_block_bytes = realloc(ctx->block_bytes, ctx->alloc_block_bytes); + if(new_block_bytes == NULL) { + MD_LOG("realloc() failed."); + return NULL; + } + + /* Fix the ->current_block after the reallocation. */ + if(ctx->current_block != NULL) { + OFF off_current_block = (char*) ctx->current_block - (char*) ctx->block_bytes; + ctx->current_block = (MD_BLOCK*) ((char*) new_block_bytes + off_current_block); + } + + ctx->block_bytes = new_block_bytes; + } + + ptr = (char*)ctx->block_bytes + ctx->n_block_bytes; + ctx->n_block_bytes += n_bytes; + return ptr; +} + +static int +md_start_new_block(MD_CTX* ctx, const MD_LINE_ANALYSIS* line) +{ + MD_BLOCK* block; + + MD_ASSERT(ctx->current_block == NULL); + + block = (MD_BLOCK*) md_push_block_bytes(ctx, sizeof(MD_BLOCK)); + if(block == NULL) + return -1; + + switch(line->type) { + case MD_LINE_HR: + block->type = MD_BLOCK_HR; + break; + + case MD_LINE_ATXHEADER: + case MD_LINE_SETEXTHEADER: + block->type = MD_BLOCK_H; + break; + + case MD_LINE_FENCEDCODE: + case MD_LINE_INDENTEDCODE: + block->type = MD_BLOCK_CODE; + break; + + case MD_LINE_TEXT: + block->type = MD_BLOCK_P; + break; + + case MD_LINE_HTML: + block->type = MD_BLOCK_HTML; + break; + + case MD_LINE_BLANK: + case MD_LINE_SETEXTUNDERLINE: + case MD_LINE_TABLEUNDERLINE: + default: + MD_UNREACHABLE(); + break; + } + + block->flags = 0; + block->data = line->data; + block->n_lines = 0; + + ctx->current_block = block; + return 0; +} + +/* Eat from start of current (textual) block any reference definitions and + * remember them so we can resolve any links referring to them. + * + * (Reference definitions can only be at start of it as they cannot break + * a paragraph.) + */ +static int +md_consume_link_reference_definitions(MD_CTX* ctx) +{ + MD_LINE* lines = (MD_LINE*) (ctx->current_block + 1); + int n_lines = ctx->current_block->n_lines; + int n = 0; + + /* Compute how many lines at the start of the block form one or more + * reference definitions. */ + while(n < n_lines) { + int n_link_ref_lines; + + n_link_ref_lines = md_is_link_reference_definition(ctx, + lines + n, n_lines - n); + /* Not a reference definition? */ + if(n_link_ref_lines == 0) + break; + + /* We fail if it is the ref. def. but it could not be stored due + * a memory allocation error. */ + if(n_link_ref_lines < 0) + return -1; + + n += n_link_ref_lines; + } + + /* If there was at least one reference definition, we need to remove + * its lines from the block, or perhaps even the whole block. */ + if(n > 0) { + if(n == n_lines) { + /* Remove complete block. */ + ctx->n_block_bytes -= n * sizeof(MD_LINE); + ctx->n_block_bytes -= sizeof(MD_BLOCK); + ctx->current_block = NULL; + } else { + /* Remove just some initial lines from the block. */ + memmove(lines, lines + n, (n_lines - n) * sizeof(MD_LINE)); + ctx->current_block->n_lines -= n; + ctx->n_block_bytes -= n * sizeof(MD_LINE); + } + } + + return 0; +} + +static int +md_end_current_block(MD_CTX* ctx) +{ + int ret = 0; + + if(ctx->current_block == NULL) + return ret; + + /* Check whether there is a reference definition. (We do this here instead + * of in md_analyze_line() because reference definition can take multiple + * lines.) */ + if(ctx->current_block->type == MD_BLOCK_P || + (ctx->current_block->type == MD_BLOCK_H && (ctx->current_block->flags & MD_BLOCK_SETEXT_HEADER))) + { + MD_LINE* lines = (MD_LINE*) (ctx->current_block + 1); + if(CH(lines[0].beg) == _T('[')) { + MD_CHECK(md_consume_link_reference_definitions(ctx)); + if(ctx->current_block == NULL) + return ret; + } + } + + if(ctx->current_block->type == MD_BLOCK_H && (ctx->current_block->flags & MD_BLOCK_SETEXT_HEADER)) { + int n_lines = ctx->current_block->n_lines; + + if(n_lines > 1) { + /* Get rid of the underline. */ + ctx->current_block->n_lines--; + ctx->n_block_bytes -= sizeof(MD_LINE); + } else { + /* Only the underline has left after eating the ref. defs. + * Keep the line as beginning of a new ordinary paragraph. */ + ctx->current_block->type = MD_BLOCK_P; + return 0; + } + } + + /* Mark we are not building any block anymore. */ + ctx->current_block = NULL; + +abort: + return ret; +} + +static int +md_add_line_into_current_block(MD_CTX* ctx, const MD_LINE_ANALYSIS* analysis) +{ + MD_ASSERT(ctx->current_block != NULL); + + if(ctx->current_block->type == MD_BLOCK_CODE || ctx->current_block->type == MD_BLOCK_HTML) { + MD_VERBATIMLINE* line; + + line = (MD_VERBATIMLINE*) md_push_block_bytes(ctx, sizeof(MD_VERBATIMLINE)); + if(line == NULL) + return -1; + + line->indent = analysis->indent; + line->beg = analysis->beg; + line->end = analysis->end; + } else { + MD_LINE* line; + + line = (MD_LINE*) md_push_block_bytes(ctx, sizeof(MD_LINE)); + if(line == NULL) + return -1; + + line->beg = analysis->beg; + line->end = analysis->end; + } + ctx->current_block->n_lines++; + + return 0; +} + +static int +md_push_container_bytes(MD_CTX* ctx, MD_BLOCKTYPE type, unsigned start, + unsigned data, unsigned flags) +{ + MD_BLOCK* block; + int ret = 0; + + MD_CHECK(md_end_current_block(ctx)); + + block = (MD_BLOCK*) md_push_block_bytes(ctx, sizeof(MD_BLOCK)); + if(block == NULL) + return -1; + + block->type = type; + block->flags = flags; + block->data = data; + block->n_lines = start; + +abort: + return ret; +} + + + +/*********************** + *** Line Analysis *** + ***********************/ + +static int +md_is_hr_line(MD_CTX* ctx, OFF beg, OFF* p_end, OFF* p_killer) +{ + OFF off = beg + 1; + int n = 1; + + while(off < ctx->size && (CH(off) == CH(beg) || CH(off) == _T(' ') || CH(off) == _T('\t'))) { + if(CH(off) == CH(beg)) + n++; + off++; + } + + if(n < 3) { + *p_killer = off; + return FALSE; + } + + /* Nothing else can be present on the line. */ + if(off < ctx->size && !ISNEWLINE(off)) { + *p_killer = off; + return FALSE; + } + + *p_end = off; + return TRUE; +} + +static int +md_is_atxheader_line(MD_CTX* ctx, OFF beg, OFF* p_beg, OFF* p_end, unsigned* p_level) +{ + int n; + OFF off = beg + 1; + + while(off < ctx->size && CH(off) == _T('#') && off - beg < 7) + off++; + n = off - beg; + + if(n > 6) + return FALSE; + *p_level = n; + + if(!(ctx->parser.flags & MD_FLAG_PERMISSIVEATXHEADERS) && off < ctx->size && + CH(off) != _T(' ') && CH(off) != _T('\t') && !ISNEWLINE(off)) + return FALSE; + + while(off < ctx->size && CH(off) == _T(' ')) + off++; + *p_beg = off; + *p_end = off; + return TRUE; +} + +static int +md_is_setext_underline(MD_CTX* ctx, OFF beg, OFF* p_end, unsigned* p_level) +{ + OFF off = beg + 1; + + while(off < ctx->size && CH(off) == CH(beg)) + off++; + + /* Optionally, space(s) can follow. */ + while(off < ctx->size && CH(off) == _T(' ')) + off++; + + /* But nothing more is allowed on the line. */ + if(off < ctx->size && !ISNEWLINE(off)) + return FALSE; + + *p_level = (CH(beg) == _T('=') ? 1 : 2); + *p_end = off; + return TRUE; +} + +static int +md_is_table_underline(MD_CTX* ctx, OFF beg, OFF* p_end, unsigned* p_col_count) +{ + OFF off = beg; + int found_pipe = FALSE; + unsigned col_count = 0; + + if(off < ctx->size && CH(off) == _T('|')) { + found_pipe = TRUE; + off++; + while(off < ctx->size && ISWHITESPACE(off)) + off++; + } + + while(1) { + OFF cell_beg; + int delimited = FALSE; + + /* Cell underline ("-----", ":----", "----:" or ":----:") */ + cell_beg = off; + if(off < ctx->size && CH(off) == _T(':')) + off++; + while(off < ctx->size && CH(off) == _T('-')) + off++; + if(off < ctx->size && CH(off) == _T(':')) + off++; + if(off - cell_beg < 3) + return FALSE; + + col_count++; + + /* Pipe delimiter (optional at the end of line). */ + while(off < ctx->size && ISWHITESPACE(off)) + off++; + if(off < ctx->size && CH(off) == _T('|')) { + delimited = TRUE; + found_pipe = TRUE; + off++; + while(off < ctx->size && ISWHITESPACE(off)) + off++; + } + + /* Success, if we reach end of line. */ + if(off >= ctx->size || ISNEWLINE(off)) + break; + + if(!delimited) + return FALSE; + } + + if(!found_pipe) + return FALSE; + + *p_end = off; + *p_col_count = col_count; + return TRUE; +} + +static int +md_is_opening_code_fence(MD_CTX* ctx, OFF beg, OFF* p_end) +{ + OFF off = beg; + + while(off < ctx->size && CH(off) == CH(beg)) + off++; + + /* Fence must have at least three characters. */ + if(off - beg < 3) + return FALSE; + + ctx->code_fence_length = off - beg; + + /* Optionally, space(s) can follow. */ + while(off < ctx->size && CH(off) == _T(' ')) + off++; + + /* Optionally, an info string can follow. */ + while(off < ctx->size && !ISNEWLINE(off)) { + /* Backtick-based fence must not contain '`' in the info string. */ + if(CH(beg) == _T('`') && CH(off) == _T('`')) + return FALSE; + off++; + } + + *p_end = off; + return TRUE; +} + +static int +md_is_closing_code_fence(MD_CTX* ctx, CHAR ch, OFF beg, OFF* p_end) +{ + OFF off = beg; + int ret = FALSE; + + /* Closing fence must have at least the same length and use same char as + * opening one. */ + while(off < ctx->size && CH(off) == ch) + off++; + if(off - beg < ctx->code_fence_length) + goto out; + + /* Optionally, space(s) can follow */ + while(off < ctx->size && CH(off) == _T(' ')) + off++; + + /* But nothing more is allowed on the line. */ + if(off < ctx->size && !ISNEWLINE(off)) + goto out; + + ret = TRUE; + +out: + /* Note we set *p_end even on failure: If we are not closing fence, caller + * would eat the line anyway without any parsing. */ + *p_end = off; + return ret; +} + +/* Returns type of the raw HTML block, or FALSE if it is not HTML block. + * (Refer to CommonMark specification for details about the types.) + */ +static int +md_is_html_block_start_condition(MD_CTX* ctx, OFF beg) +{ + typedef struct TAG_tag TAG; + struct TAG_tag { + const CHAR* name; + unsigned len : 8; + }; + + /* Type 6 is started by a long list of allowed tags. We use two-level + * tree to speed-up the search. */ +#ifdef X + #undef X +#endif +#define X(name) { _T(name), sizeof(name)-1 } +#define Xend { NULL, 0 } + static const TAG t1[] = { X("script"), X("pre"), X("style"), Xend }; + + static const TAG a6[] = { X("address"), X("article"), X("aside"), Xend }; + static const TAG b6[] = { X("base"), X("basefont"), X("blockquote"), X("body"), Xend }; + static const TAG c6[] = { X("caption"), X("center"), X("col"), X("colgroup"), Xend }; + static const TAG d6[] = { X("dd"), X("details"), X("dialog"), X("dir"), + X("div"), X("dl"), X("dt"), Xend }; + static const TAG f6[] = { X("fieldset"), X("figcaption"), X("figure"), X("footer"), + X("form"), X("frame"), X("frameset"), Xend }; + static const TAG h6[] = { X("h1"), X("head"), X("header"), X("hr"), X("html"), Xend }; + static const TAG i6[] = { X("iframe"), Xend }; + static const TAG l6[] = { X("legend"), X("li"), X("link"), Xend }; + static const TAG m6[] = { X("main"), X("menu"), X("menuitem"), Xend }; + static const TAG n6[] = { X("nav"), X("noframes"), Xend }; + static const TAG o6[] = { X("ol"), X("optgroup"), X("option"), Xend }; + static const TAG p6[] = { X("p"), X("param"), Xend }; + static const TAG s6[] = { X("section"), X("source"), X("summary"), Xend }; + static const TAG t6[] = { X("table"), X("tbody"), X("td"), X("tfoot"), X("th"), + X("thead"), X("title"), X("tr"), X("track"), Xend }; + static const TAG u6[] = { X("ul"), Xend }; + static const TAG xx[] = { Xend }; +#undef X + + static const TAG* map6[26] = { + a6, b6, c6, d6, xx, f6, xx, h6, i6, xx, xx, l6, m6, + n6, o6, p6, xx, xx, s6, t6, u6, xx, xx, xx, xx, xx + }; + OFF off = beg + 1; + int i; + + /* Check for type 1: <script, <pre, or <style */ + for(i = 0; t1[i].name != NULL; i++) { + if(off + t1[i].len <= ctx->size) { + if(md_ascii_case_eq(STR(off), t1[i].name, t1[i].len)) + return 1; + } + } + + /* Check for type 2: <!-- */ + if(off + 3 < ctx->size && CH(off) == _T('!') && CH(off+1) == _T('-') && CH(off+2) == _T('-')) + return 2; + + /* Check for type 3: <? */ + if(off < ctx->size && CH(off) == _T('?')) + return 3; + + /* Check for type 4 or 5: <! */ + if(off < ctx->size && CH(off) == _T('!')) { + /* Check for type 4: <! followed by uppercase letter. */ + if(off + 1 < ctx->size && ISUPPER(off+1)) + return 4; + + /* Check for type 5: <![CDATA[ */ + if(off + 8 < ctx->size) { + if(md_ascii_eq(STR(off), _T("![CDATA["), 8 * sizeof(CHAR))) + return 5; + } + } + + /* Check for type 6: Many possible starting tags listed above. */ + if(off + 1 < ctx->size && (ISALPHA(off) || (CH(off) == _T('/') && ISALPHA(off+1)))) { + int slot; + const TAG* tags; + + if(CH(off) == _T('/')) + off++; + + slot = (ISUPPER(off) ? CH(off) - 'A' : CH(off) - 'a'); + tags = map6[slot]; + + for(i = 0; tags[i].name != NULL; i++) { + if(off + tags[i].len <= ctx->size) { + if(md_ascii_case_eq(STR(off), tags[i].name, tags[i].len)) { + OFF tmp = off + tags[i].len; + if(tmp >= ctx->size) + return 6; + if(ISBLANK(tmp) || ISNEWLINE(tmp) || CH(tmp) == _T('>')) + return 6; + if(tmp+1 < ctx->size && CH(tmp) == _T('/') && CH(tmp+1) == _T('>')) + return 6; + break; + } + } + } + } + + /* Check for type 7: any COMPLETE other opening or closing tag. */ + if(off + 1 < ctx->size) { + OFF end; + + if(md_is_html_tag(ctx, NULL, 0, beg, ctx->size, &end)) { + /* Only optional whitespace and new line may follow. */ + while(end < ctx->size && ISWHITESPACE(end)) + end++; + if(end >= ctx->size || ISNEWLINE(end)) + return 7; + } + } + + return FALSE; +} + +/* Case sensitive check whether there is a substring 'what' between 'beg' + * and end of line. */ +static int +md_line_contains(MD_CTX* ctx, OFF beg, const CHAR* what, SZ what_len, OFF* p_end) +{ + OFF i; + for(i = beg; i + what_len < ctx->size; i++) { + if(ISNEWLINE(i)) + break; + if(memcmp(STR(i), what, what_len * sizeof(CHAR)) == 0) { + *p_end = i + what_len; + return TRUE; + } + } + + *p_end = i; + return FALSE; +} + +/* Returns type of HTML block end condition or FALSE if not an end condition. + * + * Note it fills p_end even when it is not end condition as the caller + * does not need to analyze contents of a raw HTML block. + */ +static int +md_is_html_block_end_condition(MD_CTX* ctx, OFF beg, OFF* p_end) +{ + switch(ctx->html_block_type) { + case 1: + { + OFF off = beg; + + while(off < ctx->size && !ISNEWLINE(off)) { + if(CH(off) == _T('<')) { + if(md_ascii_case_eq(STR(off), _T("</script>"), 9)) { + *p_end = off + 9; + return TRUE; + } + + if(md_ascii_case_eq(STR(off), _T("</style>"), 8)) { + *p_end = off + 8; + return TRUE; + } + + if(md_ascii_case_eq(STR(off), _T("</pre>"), 6)) { + *p_end = off + 6; + return TRUE; + } + } + + off++; + } + *p_end = off; + return FALSE; + } + + case 2: + return (md_line_contains(ctx, beg, _T("-->"), 3, p_end) ? 2 : FALSE); + + case 3: + return (md_line_contains(ctx, beg, _T("?>"), 2, p_end) ? 3 : FALSE); + + case 4: + return (md_line_contains(ctx, beg, _T(">"), 1, p_end) ? 4 : FALSE); + + case 5: + return (md_line_contains(ctx, beg, _T("]]>"), 3, p_end) ? 5 : FALSE); + + case 6: /* Pass through */ + case 7: + *p_end = beg; + return (ISNEWLINE(beg) ? ctx->html_block_type : FALSE); + + default: + MD_UNREACHABLE(); + } + return FALSE; +} + + +static int +md_is_container_compatible(const MD_CONTAINER* pivot, const MD_CONTAINER* container) +{ + /* Block quote has no "items" like lists. */ + if(container->ch == _T('>')) + return FALSE; + + if(container->ch != pivot->ch) + return FALSE; + if(container->mark_indent > pivot->contents_indent) + return FALSE; + + return TRUE; +} + +static int +md_push_container(MD_CTX* ctx, const MD_CONTAINER* container) +{ + if(ctx->n_containers >= ctx->alloc_containers) { + MD_CONTAINER* new_containers; + + ctx->alloc_containers = (ctx->alloc_containers > 0 ? ctx->alloc_containers * 2 : 16); + new_containers = realloc(ctx->containers, ctx->alloc_containers * sizeof(MD_CONTAINER)); + if(new_containers == NULL) { + MD_LOG("realloc() failed."); + return -1; + } + + ctx->containers = new_containers; + } + + memcpy(&ctx->containers[ctx->n_containers++], container, sizeof(MD_CONTAINER)); + return 0; +} + +static int +md_enter_child_containers(MD_CTX* ctx, int n_children, unsigned data) +{ + int i; + int ret = 0; + + for(i = ctx->n_containers - n_children; i < ctx->n_containers; i++) { + MD_CONTAINER* c = &ctx->containers[i]; + int is_ordered_list = FALSE; + + switch(c->ch) { + case _T(')'): + case _T('.'): + is_ordered_list = TRUE; + /* Pass through */ + + case _T('-'): + case _T('+'): + case _T('*'): + /* Remember offset in ctx->block_bytes so we can revisit the + * block if we detect it is a loose list. */ + md_end_current_block(ctx); + c->block_byte_off = ctx->n_block_bytes; + + MD_CHECK(md_push_container_bytes(ctx, + (is_ordered_list ? MD_BLOCK_OL : MD_BLOCK_UL), + c->start, data, MD_BLOCK_CONTAINER_OPENER)); + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_LI, + c->task_mark_off, + (c->is_task ? CH(c->task_mark_off) : 0), + MD_BLOCK_CONTAINER_OPENER)); + break; + + case _T('>'): + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_QUOTE, 0, 0, MD_BLOCK_CONTAINER_OPENER)); + break; + + default: + MD_UNREACHABLE(); + break; + } + } + +abort: + return ret; +} + +static int +md_leave_child_containers(MD_CTX* ctx, int n_keep) +{ + int ret = 0; + + while(ctx->n_containers > n_keep) { + MD_CONTAINER* c = &ctx->containers[ctx->n_containers-1]; + int is_ordered_list = FALSE; + + switch(c->ch) { + case _T(')'): + case _T('.'): + is_ordered_list = TRUE; + /* Pass through */ + + case _T('-'): + case _T('+'): + case _T('*'): + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_LI, + c->task_mark_off, (c->is_task ? CH(c->task_mark_off) : 0), + MD_BLOCK_CONTAINER_CLOSER)); + MD_CHECK(md_push_container_bytes(ctx, + (is_ordered_list ? MD_BLOCK_OL : MD_BLOCK_UL), 0, + c->ch, MD_BLOCK_CONTAINER_CLOSER)); + break; + + case _T('>'): + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_QUOTE, 0, + 0, MD_BLOCK_CONTAINER_CLOSER)); + break; + + default: + MD_UNREACHABLE(); + break; + } + + ctx->n_containers--; + } + +abort: + return ret; +} + +static int +md_is_container_mark(MD_CTX* ctx, unsigned indent, OFF beg, OFF* p_end, MD_CONTAINER* p_container) +{ + OFF off = beg; + OFF max_end; + + if(indent >= ctx->code_indent_offset) + return FALSE; + + /* Check for block quote mark. */ + if(off < ctx->size && CH(off) == _T('>')) { + off++; + p_container->ch = _T('>'); + p_container->is_loose = FALSE; + p_container->is_task = FALSE; + p_container->mark_indent = indent; + p_container->contents_indent = indent + 1; + *p_end = off; + return TRUE; + } + + /* Check for list item bullet mark. */ + if(off+1 < ctx->size && ISANYOF(off, _T("-+*")) && (ISBLANK(off+1) || ISNEWLINE(off+1))) { + p_container->ch = CH(off); + p_container->is_loose = FALSE; + p_container->is_task = FALSE; + p_container->mark_indent = indent; + p_container->contents_indent = indent + 1; + *p_end = off + 1; + return TRUE; + } + + /* Check for ordered list item marks. */ + max_end = off + 9; + if(max_end > ctx->size) + max_end = ctx->size; + p_container->start = 0; + while(off < max_end && ISDIGIT(off)) { + p_container->start = p_container->start * 10 + CH(off) - _T('0'); + off++; + } + if(off+1 < ctx->size && (CH(off) == _T('.') || CH(off) == _T(')')) && (ISBLANK(off+1) || ISNEWLINE(off+1))) { + p_container->ch = CH(off); + p_container->is_loose = FALSE; + p_container->is_task = FALSE; + p_container->mark_indent = indent; + p_container->contents_indent = indent + off - beg + 1; + *p_end = off + 1; + return TRUE; + } + + return FALSE; +} + +static unsigned +md_line_indentation(MD_CTX* ctx, unsigned total_indent, OFF beg, OFF* p_end) +{ + OFF off = beg; + unsigned indent = total_indent; + + while(off < ctx->size && ISBLANK(off)) { + if(CH(off) == _T('\t')) + indent = (indent + 4) & ~3; + else + indent++; + off++; + } + + *p_end = off; + return indent - total_indent; +} + +static const MD_LINE_ANALYSIS md_dummy_blank_line = { MD_LINE_BLANK, 0 }; + +/* Analyze type of the line and find some its properties. This serves as a + * main input for determining type and boundaries of a block. */ +static int +md_analyze_line(MD_CTX* ctx, OFF beg, OFF* p_end, + const MD_LINE_ANALYSIS* pivot_line, MD_LINE_ANALYSIS* line) +{ + unsigned total_indent = 0; + int n_parents = 0; + int n_brothers = 0; + int n_children = 0; + MD_CONTAINER container = { 0 }; + int prev_line_has_list_loosening_effect = ctx->last_line_has_list_loosening_effect; + OFF off = beg; + OFF hr_killer = 0; + int ret = 0; + + line->indent = md_line_indentation(ctx, total_indent, off, &off); + total_indent += line->indent; + line->beg = off; + + /* Given the indentation and block quote marks '>', determine how many of + * the current containers are our parents. */ + while(n_parents < ctx->n_containers) { + MD_CONTAINER* c = &ctx->containers[n_parents]; + + if(c->ch == _T('>') && line->indent < ctx->code_indent_offset && + off < ctx->size && CH(off) == _T('>')) + { + /* Block quote mark. */ + off++; + total_indent++; + line->indent = md_line_indentation(ctx, total_indent, off, &off); + total_indent += line->indent; + + /* The optional 1st space after '>' is part of the block quote mark. */ + if(line->indent > 0) + line->indent--; + + line->beg = off; + } else if(c->ch != _T('>') && line->indent >= c->contents_indent) { + /* List. */ + line->indent -= c->contents_indent; + } else { + break; + } + + n_parents++; + } + + if(off >= ctx->size || ISNEWLINE(off)) { + /* Blank line does not need any real indentation to be nested inside + * a list. */ + if(n_brothers + n_children == 0) { + while(n_parents < ctx->n_containers && ctx->containers[n_parents].ch != _T('>')) + n_parents++; + } + } + + while(TRUE) { + /* Check whether we are fenced code continuation. */ + if(pivot_line->type == MD_LINE_FENCEDCODE) { + line->beg = off; + + /* We are another MD_LINE_FENCEDCODE unless we are closing fence + * which we transform into MD_LINE_BLANK. */ + if(line->indent < ctx->code_indent_offset) { + if(md_is_closing_code_fence(ctx, CH(pivot_line->beg), off, &off)) { + line->type = MD_LINE_BLANK; + ctx->last_line_has_list_loosening_effect = FALSE; + break; + } + } + + /* Change indentation accordingly to the initial code fence. */ + if(n_parents == ctx->n_containers) { + if(line->indent > pivot_line->indent) + line->indent -= pivot_line->indent; + else + line->indent = 0; + + line->type = MD_LINE_FENCEDCODE; + break; + } + } + + /* Check whether we are HTML block continuation. */ + if(pivot_line->type == MD_LINE_HTML && ctx->html_block_type > 0) { + int html_block_type; + + html_block_type = md_is_html_block_end_condition(ctx, off, &off); + if(html_block_type > 0) { + MD_ASSERT(html_block_type == ctx->html_block_type); + + /* Make sure this is the last line of the block. */ + ctx->html_block_type = 0; + + /* Some end conditions serve as blank lines at the same time. */ + if(html_block_type == 6 || html_block_type == 7) { + line->type = MD_LINE_BLANK; + line->indent = 0; + break; + } + } + + if(n_parents == ctx->n_containers) { + line->type = MD_LINE_HTML; + break; + } + } + + /* Check for blank line. */ + if(off >= ctx->size || ISNEWLINE(off)) { + if(pivot_line->type == MD_LINE_INDENTEDCODE && n_parents == ctx->n_containers) { + line->type = MD_LINE_INDENTEDCODE; + if(line->indent > ctx->code_indent_offset) + line->indent -= ctx->code_indent_offset; + else + line->indent = 0; + ctx->last_line_has_list_loosening_effect = FALSE; + } else { + line->type = MD_LINE_BLANK; + ctx->last_line_has_list_loosening_effect = (n_parents > 0 && + n_brothers + n_children == 0 && + ctx->containers[n_parents-1].ch != _T('>')); + + #if 1 + /* See https://github.com/mity/md4c/issues/6 + * + * This ugly checking tests we are in (yet empty) list item but not + * its very first line (with the list item mark). + * + * If we are such blank line, then any following non-blank line + * which would be part of this list item actually ends the list + * because "a list item can begin with at most one blank line." + */ + if(n_parents > 0 && ctx->containers[n_parents-1].ch != _T('>') && + n_brothers + n_children == 0 && ctx->current_block == NULL && + ctx->n_block_bytes > (int) sizeof(MD_BLOCK)) + { + MD_BLOCK* top_block = (MD_BLOCK*) ((char*)ctx->block_bytes + ctx->n_block_bytes - sizeof(MD_BLOCK)); + if(top_block->type == MD_BLOCK_LI) + ctx->last_list_item_starts_with_two_blank_lines = TRUE; + } + #endif + } + break; + } else { + #if 1 + /* This is 2nd half of the hack. If the flag is set (that is there + * were 2nd blank line at the start of the list item) and we would also + * belonging to such list item, then interrupt the list. */ + ctx->last_line_has_list_loosening_effect = FALSE; + if(ctx->last_list_item_starts_with_two_blank_lines) { + if(n_parents > 0 && ctx->containers[n_parents-1].ch != _T('>') && + n_brothers + n_children == 0 && ctx->current_block == NULL && + ctx->n_block_bytes > (int) sizeof(MD_BLOCK)) + { + MD_BLOCK* top_block = (MD_BLOCK*) ((char*)ctx->block_bytes + ctx->n_block_bytes - sizeof(MD_BLOCK)); + if(top_block->type == MD_BLOCK_LI) + n_parents--; + } + + ctx->last_list_item_starts_with_two_blank_lines = FALSE; + } + #endif + } + + /* Check whether we are Setext underline. */ + if(line->indent < ctx->code_indent_offset && pivot_line->type == MD_LINE_TEXT + && (CH(off) == _T('=') || CH(off) == _T('-')) + && (n_parents == ctx->n_containers)) + { + unsigned level; + + if(md_is_setext_underline(ctx, off, &off, &level)) { + line->type = MD_LINE_SETEXTUNDERLINE; + line->data = level; + break; + } + } + + /* Check for thematic break line. */ + if(line->indent < ctx->code_indent_offset && ISANYOF(off, _T("-_*")) && off >= hr_killer) { + if(md_is_hr_line(ctx, off, &off, &hr_killer)) { + line->type = MD_LINE_HR; + break; + } + } + + /* Check for "brother" container. I.e. whether we are another list item + * in already started list. */ + if(n_parents < ctx->n_containers && n_brothers + n_children == 0) { + OFF tmp; + + if(md_is_container_mark(ctx, line->indent, off, &tmp, &container) && + md_is_container_compatible(&ctx->containers[n_parents], &container)) + { + pivot_line = &md_dummy_blank_line; + + off = tmp; + + total_indent += container.contents_indent - container.mark_indent; + line->indent = md_line_indentation(ctx, total_indent, off, &off); + total_indent += line->indent; + line->beg = off; + + /* Some of the following whitespace actually still belongs to the mark. */ + if(off >= ctx->size || ISNEWLINE(off)) { + container.contents_indent++; + } else if(line->indent <= ctx->code_indent_offset) { + container.contents_indent += line->indent; + line->indent = 0; + } else { + container.contents_indent += 1; + line->indent--; + } + + ctx->containers[n_parents].mark_indent = container.mark_indent; + ctx->containers[n_parents].contents_indent = container.contents_indent; + + n_brothers++; + continue; + } + } + + /* Check for indented code. + * Note indented code block cannot interrupt a paragraph. */ + if(line->indent >= ctx->code_indent_offset && + (pivot_line->type == MD_LINE_BLANK || pivot_line->type == MD_LINE_INDENTEDCODE)) + { + line->type = MD_LINE_INDENTEDCODE; + MD_ASSERT(line->indent >= ctx->code_indent_offset); + line->indent -= ctx->code_indent_offset; + line->data = 0; + break; + } + + /* Check for start of a new container block. */ + if(line->indent < ctx->code_indent_offset && + md_is_container_mark(ctx, line->indent, off, &off, &container)) + { + if(pivot_line->type == MD_LINE_TEXT && n_parents == ctx->n_containers && + (off >= ctx->size || ISNEWLINE(off))) + { + /* Noop. List mark followed by a blank line cannot interrupt a paragraph. */ + } else if(pivot_line->type == MD_LINE_TEXT && n_parents == ctx->n_containers && + (container.ch == _T('.') || container.ch == _T(')')) && container.start != 1) + { + /* Noop. Ordered list cannot interrupt a paragraph unless the start index is 1. */ + } else { + total_indent += container.contents_indent - container.mark_indent; + line->indent = md_line_indentation(ctx, total_indent, off, &off); + total_indent += line->indent; + + line->beg = off; + line->data = container.ch; + + /* Some of the following whitespace actually still belongs to the mark. */ + if(off >= ctx->size || ISNEWLINE(off)) { + container.contents_indent++; + } else if(line->indent <= ctx->code_indent_offset) { + container.contents_indent += line->indent; + line->indent = 0; + } else { + container.contents_indent += 1; + line->indent--; + } + + if(n_brothers + n_children == 0) + pivot_line = &md_dummy_blank_line; + + if(n_children == 0) + MD_CHECK(md_leave_child_containers(ctx, n_parents + n_brothers)); + + n_children++; + MD_CHECK(md_push_container(ctx, &container)); + continue; + } + } + + /* Check whether we are table continuation. */ + if(pivot_line->type == MD_LINE_TABLE && md_is_table_row(ctx, off, &off) && + n_parents == ctx->n_containers) + { + line->type = MD_LINE_TABLE; + break; + } + + /* Check for ATX header. */ + if(line->indent < ctx->code_indent_offset && CH(off) == _T('#')) { + unsigned level; + + if(md_is_atxheader_line(ctx, off, &line->beg, &off, &level)) { + line->type = MD_LINE_ATXHEADER; + line->data = level; + break; + } + } + + /* Check whether we are starting code fence. */ + if(CH(off) == _T('`') || CH(off) == _T('~')) { + if(md_is_opening_code_fence(ctx, off, &off)) { + line->type = MD_LINE_FENCEDCODE; + line->data = 1; + break; + } + } + + /* Check for start of raw HTML block. */ + if(CH(off) == _T('<') && !(ctx->parser.flags & MD_FLAG_NOHTMLBLOCKS)) + { + ctx->html_block_type = md_is_html_block_start_condition(ctx, off); + + /* HTML block type 7 cannot interrupt paragraph. */ + if(ctx->html_block_type == 7 && pivot_line->type == MD_LINE_TEXT) + ctx->html_block_type = 0; + + if(ctx->html_block_type > 0) { + /* The line itself also may immediately close the block. */ + if(md_is_html_block_end_condition(ctx, off, &off) == ctx->html_block_type) { + /* Make sure this is the last line of the block. */ + ctx->html_block_type = 0; + } + + line->type = MD_LINE_HTML; + break; + } + } + + /* Check for table underline. */ + if((ctx->parser.flags & MD_FLAG_TABLES) && pivot_line->type == MD_LINE_TEXT && + (CH(off) == _T('|') || CH(off) == _T('-') || CH(off) == _T(':')) && + n_parents == ctx->n_containers) + { + unsigned col_count; + + if(ctx->current_block != NULL && ctx->current_block->n_lines == 1 && + md_is_table_underline(ctx, off, &off, &col_count) && + md_is_table_row(ctx, pivot_line->beg, NULL)) + { + line->data = col_count; + line->type = MD_LINE_TABLEUNDERLINE; + break; + } + } + + /* By default, we are normal text line. */ + line->type = MD_LINE_TEXT; + if(pivot_line->type == MD_LINE_TEXT && n_brothers + n_children == 0) { + /* Lazy continuation. */ + n_parents = ctx->n_containers; + } + + /* Check for task mark. */ + if((ctx->parser.flags & MD_FLAG_TASKLISTS) && n_brothers + n_children > 0 && + ISANYOF_(ctx->containers[ctx->n_containers-1].ch, _T("-+*.)"))) + { + OFF tmp = off; + + while(tmp < ctx->size && tmp < off + 3 && ISBLANK(tmp)) + tmp++; + if(tmp + 2 < ctx->size && CH(tmp) == _T('[') && + ISANYOF(tmp+1, _T("xX ")) && CH(tmp+2) == _T(']') && + (tmp + 3 == ctx->size || ISBLANK(tmp+3) || ISNEWLINE(tmp+3))) + { + MD_CONTAINER* task_container = (n_children > 0 ? &ctx->containers[ctx->n_containers-1] : &container); + task_container->is_task = TRUE; + task_container->task_mark_off = tmp + 1; + off = tmp + 3; + while(ISWHITESPACE(off)) + off++; + line->beg = off; + } + } + + break; + } + + /* Scan for end of the line. + * + * Note this is quite a bottleneck of the parsing as we here iterate almost + * over compete document. + */ +#if defined __linux__ && !defined MD4C_USE_UTF16 + /* Recent glibc versions have superbly optimized strcspn(), even using + * vectorization if available. */ + if(ctx->doc_ends_with_newline && off < ctx->size) { + while(TRUE) { + off += (OFF) strcspn(STR(off), "\r\n"); + + /* strcspn() can stop on zero terminator; but that can appear + * anywhere in the Markfown input... */ + if(CH(off) == _T('\0')) + off++; + else + break; + } + } else +#endif + { + /* Optimization: Use some loop unrolling. */ + while(off + 3 < ctx->size && !ISNEWLINE(off+0) && !ISNEWLINE(off+1) + && !ISNEWLINE(off+2) && !ISNEWLINE(off+3)) + off += 4; + while(off < ctx->size && !ISNEWLINE(off)) + off++; + } + + /* Set end of the line. */ + line->end = off; + + /* But for ATX header, we should exclude the optional trailing mark. */ + if(line->type == MD_LINE_ATXHEADER) { + OFF tmp = line->end; + while(tmp > line->beg && CH(tmp-1) == _T(' ')) + tmp--; + while(tmp > line->beg && CH(tmp-1) == _T('#')) + tmp--; + if(tmp == line->beg || CH(tmp-1) == _T(' ') || (ctx->parser.flags & MD_FLAG_PERMISSIVEATXHEADERS)) + line->end = tmp; + } + + /* Trim trailing spaces. */ + if(line->type != MD_LINE_INDENTEDCODE && line->type != MD_LINE_FENCEDCODE) { + while(line->end > line->beg && CH(line->end-1) == _T(' ')) + line->end--; + } + + /* Eat also the new line. */ + if(off < ctx->size && CH(off) == _T('\r')) + off++; + if(off < ctx->size && CH(off) == _T('\n')) + off++; + + *p_end = off; + + /* If we belong to a list after seeing a blank line, the list is loose. */ + if(prev_line_has_list_loosening_effect && line->type != MD_LINE_BLANK && n_parents + n_brothers > 0) { + MD_CONTAINER* c = &ctx->containers[n_parents + n_brothers - 1]; + if(c->ch != _T('>')) { + MD_BLOCK* block = (MD_BLOCK*) (((char*)ctx->block_bytes) + c->block_byte_off); + block->flags |= MD_BLOCK_LOOSE_LIST; + } + } + + /* Leave any containers we are not part of anymore. */ + if(n_children == 0 && n_parents + n_brothers < ctx->n_containers) + MD_CHECK(md_leave_child_containers(ctx, n_parents + n_brothers)); + + /* Enter any container we found a mark for. */ + if(n_brothers > 0) { + MD_ASSERT(n_brothers == 1); + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_LI, + ctx->containers[n_parents].task_mark_off, + (ctx->containers[n_parents].is_task ? CH(ctx->containers[n_parents].task_mark_off) : 0), + MD_BLOCK_CONTAINER_CLOSER)); + MD_CHECK(md_push_container_bytes(ctx, MD_BLOCK_LI, + container.task_mark_off, + (container.is_task ? CH(container.task_mark_off) : 0), + MD_BLOCK_CONTAINER_OPENER)); + ctx->containers[n_parents].is_task = container.is_task; + ctx->containers[n_parents].task_mark_off = container.task_mark_off; + } + + if(n_children > 0) + MD_CHECK(md_enter_child_containers(ctx, n_children, line->data)); + +abort: + return ret; +} + +static int +md_process_line(MD_CTX* ctx, const MD_LINE_ANALYSIS** p_pivot_line, MD_LINE_ANALYSIS* line) +{ + const MD_LINE_ANALYSIS* pivot_line = *p_pivot_line; + int ret = 0; + + /* Blank line ends current leaf block. */ + if(line->type == MD_LINE_BLANK) { + MD_CHECK(md_end_current_block(ctx)); + *p_pivot_line = &md_dummy_blank_line; + return 0; + } + + /* Some line types form block on their own. */ + if(line->type == MD_LINE_HR || line->type == MD_LINE_ATXHEADER) { + MD_CHECK(md_end_current_block(ctx)); + + /* Add our single-line block. */ + MD_CHECK(md_start_new_block(ctx, line)); + MD_CHECK(md_add_line_into_current_block(ctx, line)); + MD_CHECK(md_end_current_block(ctx)); + *p_pivot_line = &md_dummy_blank_line; + return 0; + } + + /* MD_LINE_SETEXTUNDERLINE changes meaning of the current block and ends it. */ + if(line->type == MD_LINE_SETEXTUNDERLINE) { + MD_ASSERT(ctx->current_block != NULL); + ctx->current_block->type = MD_BLOCK_H; + ctx->current_block->data = line->data; + ctx->current_block->flags |= MD_BLOCK_SETEXT_HEADER; + MD_CHECK(md_add_line_into_current_block(ctx, line)); + MD_CHECK(md_end_current_block(ctx)); + if(ctx->current_block == NULL) { + *p_pivot_line = &md_dummy_blank_line; + } else { + /* This happens if we have consumed all the body as link ref. defs. + * and downgraded the underline into start of a new paragraph block. */ + line->type = MD_LINE_TEXT; + *p_pivot_line = line; + } + return 0; + } + + /* MD_LINE_TABLEUNDERLINE changes meaning of the current block. */ + if(line->type == MD_LINE_TABLEUNDERLINE) { + MD_ASSERT(ctx->current_block != NULL); + MD_ASSERT(ctx->current_block->n_lines == 1); + ctx->current_block->type = MD_BLOCK_TABLE; + ctx->current_block->data = line->data; + MD_ASSERT(pivot_line != &md_dummy_blank_line); + ((MD_LINE_ANALYSIS*)pivot_line)->type = MD_LINE_TABLE; + MD_CHECK(md_add_line_into_current_block(ctx, line)); + return 0; + } + + /* The current block also ends if the line has different type. */ + if(line->type != pivot_line->type) + MD_CHECK(md_end_current_block(ctx)); + + /* The current line may start a new block. */ + if(ctx->current_block == NULL) { + MD_CHECK(md_start_new_block(ctx, line)); + *p_pivot_line = line; + } + + /* In all other cases the line is just a continuation of the current block. */ + MD_CHECK(md_add_line_into_current_block(ctx, line)); + +abort: + return ret; +} + +static int +md_process_doc(MD_CTX *ctx) +{ + const MD_LINE_ANALYSIS* pivot_line = &md_dummy_blank_line; + MD_LINE_ANALYSIS line_buf[2]; + MD_LINE_ANALYSIS* line = &line_buf[0]; + OFF off = 0; + int ret = 0; + + MD_ENTER_BLOCK(MD_BLOCK_DOC, NULL); + + while(off < ctx->size) { + if(line == pivot_line) + line = (line == &line_buf[0] ? &line_buf[1] : &line_buf[0]); + + MD_CHECK(md_analyze_line(ctx, off, &off, pivot_line, line)); + MD_CHECK(md_process_line(ctx, &pivot_line, line)); + } + + md_end_current_block(ctx); + + MD_CHECK(md_build_ref_def_hashtable(ctx)); + + /* Process all blocks. */ + MD_CHECK(md_leave_child_containers(ctx, 0)); + MD_CHECK(md_process_all_blocks(ctx)); + + MD_LEAVE_BLOCK(MD_BLOCK_DOC, NULL); + +abort: + +#if 0 + /* Output some memory consumption statistics. */ + { + char buffer[256]; + sprintf(buffer, "Alloced %u bytes for block buffer.", + (unsigned)(ctx->alloc_block_bytes)); + MD_LOG(buffer); + + sprintf(buffer, "Alloced %u bytes for containers buffer.", + (unsigned)(ctx->alloc_containers * sizeof(MD_CONTAINER))); + MD_LOG(buffer); + + sprintf(buffer, "Alloced %u bytes for marks buffer.", + (unsigned)(ctx->alloc_marks * sizeof(MD_MARK))); + MD_LOG(buffer); + + sprintf(buffer, "Alloced %u bytes for aux. buffer.", + (unsigned)(ctx->alloc_buffer * sizeof(MD_CHAR))); + MD_LOG(buffer); + } +#endif + + return ret; +} + + +/******************** + *** Public API *** + ********************/ + +int +md_parse(const MD_CHAR* text, MD_SIZE size, const MD_PARSER* parser, void* userdata) +{ + MD_CTX ctx; + int i; + int ret; + + if(parser->abi_version != 0) { + if(parser->debug_log != NULL) + parser->debug_log("Unsupported abi_version.", userdata); + return -1; + } + + /* Setup context structure. */ + memset(&ctx, 0, sizeof(MD_CTX)); + ctx.text = text; + ctx.size = size; + memcpy(&ctx.parser, parser, sizeof(MD_PARSER)); + ctx.userdata = userdata; + ctx.code_indent_offset = (ctx.parser.flags & MD_FLAG_NOINDENTEDCODEBLOCKS) ? (OFF)(-1) : 4; + md_build_mark_char_map(&ctx); + ctx.doc_ends_with_newline = (size > 0 && ISNEWLINE_(text[size-1])); + + /* Reset all unresolved opener mark chains. */ + for(i = 0; i < (int) SIZEOF_ARRAY(ctx.mark_chains); i++) { + ctx.mark_chains[i].head = -1; + ctx.mark_chains[i].tail = -1; + } + ctx.unresolved_link_head = -1; + ctx.unresolved_link_tail = -1; + + /* All the work. */ + ret = md_process_doc(&ctx); + + /* Clean-up. */ + md_free_ref_defs(&ctx); + md_free_ref_def_hashtable(&ctx); + free(ctx.buffer); + free(ctx.marks); + free(ctx.block_bytes); + free(ctx.containers); + + return ret; +} diff --git a/src/3rdparty/md4c/md4c.h b/src/3rdparty/md4c/md4c.h new file mode 100644 index 0000000000..dcdadad88d --- /dev/null +++ b/src/3rdparty/md4c/md4c.h @@ -0,0 +1,362 @@ +/* + * MD4C: Markdown parser for C + * (http://github.com/mity/md4c) + * + * Copyright (c) 2016-2019 Martin Mitas + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + */ + +#ifndef MD4C_MARKDOWN_H +#define MD4C_MARKDOWN_H + +#ifdef __cplusplus + extern "C" { +#endif + +#if defined MD4C_USE_UTF16 + /* Magic to support UTF-16. Not that in order to use it, you have to define + * the macro MD4C_USE_UTF16 both when building MD4C as well as when + * including this header in your code. */ + #ifdef _WIN32 + #include <windows.h> + typedef WCHAR MD_CHAR; + #else + #error MD4C_USE_UTF16 is only supported on Windows. + #endif +#else + typedef char MD_CHAR; +#endif + +typedef unsigned MD_SIZE; +typedef unsigned MD_OFFSET; + + +/* Block represents a part of document hierarchy structure like a paragraph + * or list item. + */ +typedef enum MD_BLOCKTYPE { + /* <body>...</body> */ + MD_BLOCK_DOC = 0, + + /* <blockquote>...</blockquote> */ + MD_BLOCK_QUOTE, + + /* <ul>...</ul> + * Detail: Structure MD_BLOCK_UL_DETAIL. */ + MD_BLOCK_UL, + + /* <ol>...</ol> + * Detail: Structure MD_BLOCK_OL_DETAIL. */ + MD_BLOCK_OL, + + /* <li>...</li> + * Detail: Structure MD_BLOCK_LI_DETAIL. */ + MD_BLOCK_LI, + + /* <hr> */ + MD_BLOCK_HR, + + /* <h1>...</h1> (for levels up to 6) + * Detail: Structure MD_BLOCK_H_DETAIL. */ + MD_BLOCK_H, + + /* <pre><code>...</code></pre> + * Note the text lines within code blocks are terminated with '\n' + * instead of explicit MD_TEXT_BR. */ + MD_BLOCK_CODE, + + /* Raw HTML block. This itself does not correspond to any particular HTML + * tag. The contents of it _is_ raw HTML source intended to be put + * in verbatim form to the HTML output. */ + MD_BLOCK_HTML, + + /* <p>...</p> */ + MD_BLOCK_P, + + /* <table>...</table> and its contents. + * Detail: Structure MD_BLOCK_TD_DETAIL (used with MD_BLOCK_TH and MD_BLOCK_TD) + * Note all of these are used only if extension MD_FLAG_TABLES is enabled. */ + MD_BLOCK_TABLE, + MD_BLOCK_THEAD, + MD_BLOCK_TBODY, + MD_BLOCK_TR, + MD_BLOCK_TH, + MD_BLOCK_TD +} MD_BLOCKTYPE; + +/* Span represents an in-line piece of a document which should be rendered with + * the same font, color and other attributes. A sequence of spans forms a block + * like paragraph or list item. */ +typedef enum MD_SPANTYPE { + /* <em>...</em> */ + MD_SPAN_EM, + + /* <strong>...</strong> */ + MD_SPAN_STRONG, + + /* <a href="xxx">...</a> + * Detail: Structure MD_SPAN_A_DETAIL. */ + MD_SPAN_A, + + /* <img src="xxx">...</a> + * Detail: Structure MD_SPAN_IMG_DETAIL. + * Note: Image text can contain nested spans and even nested images. + * If rendered into ALT attribute of HTML <IMG> tag, it's responsibility + * of the renderer to deal with it. + */ + MD_SPAN_IMG, + + /* <code>...</code> */ + MD_SPAN_CODE, + + /* <del>...</del> + * Note: Recognized only when MD_FLAG_STRIKETHROUGH is enabled. + */ + MD_SPAN_DEL +} MD_SPANTYPE; + +/* Text is the actual textual contents of span. */ +typedef enum MD_TEXTTYPE { + /* Normal text. */ + MD_TEXT_NORMAL = 0, + + /* NULL character. CommonMark requires replacing NULL character with + * the replacement char U+FFFD, so this allows caller to do that easily. */ + MD_TEXT_NULLCHAR, + + /* Line breaks. + * Note these are not sent from blocks with verbatim output (MD_BLOCK_CODE + * or MD_BLOCK_HTML). In such cases, '\n' is part of the text itself. */ + MD_TEXT_BR, /* <br> (hard break) */ + MD_TEXT_SOFTBR, /* '\n' in source text where it is not semantically meaningful (soft break) */ + + /* Entity. + * (a) Named entity, e.g. + * (Note MD4C does not have a list of known entities. + * Anything matching the regexp /&[A-Za-z][A-Za-z0-9]{1,47};/ is + * treated as a named entity.) + * (b) Numerical entity, e.g. Ӓ + * (c) Hexadecimal entity, e.g. ካ + * + * As MD4C is mostly encoding agnostic, application gets the verbatim + * entity text into the MD_RENDERER::text_callback(). */ + MD_TEXT_ENTITY, + + /* Text in a code block (inside MD_BLOCK_CODE) or inlined code (`code`). + * If it is inside MD_BLOCK_CODE, it includes spaces for indentation and + * '\n' for new lines. MD_TEXT_BR and MD_TEXT_SOFTBR are not sent for this + * kind of text. */ + MD_TEXT_CODE, + + /* Text is a raw HTML. If it is contents of a raw HTML block (i.e. not + * an inline raw HTML), then MD_TEXT_BR and MD_TEXT_SOFTBR are not used. + * The text contains verbatim '\n' for the new lines. */ + MD_TEXT_HTML +} MD_TEXTTYPE; + + +/* Alignment enumeration. */ +typedef enum MD_ALIGN { + MD_ALIGN_DEFAULT = 0, /* When unspecified. */ + MD_ALIGN_LEFT, + MD_ALIGN_CENTER, + MD_ALIGN_RIGHT +} MD_ALIGN; + + +/* String attribute. + * + * This wraps strings which are outside of a normal text flow and which are + * propagated within various detailed structures, but which still may contain + * string portions of different types like e.g. entities. + * + * So, for example, lets consider an image has a title attribute string + * set to "foo " bar". (Note the string size is 14.) + * + * Then the attribute MD_SPAN_IMG_DETAIL::title shall provide the following: + * -- [0]: "foo " (substr_types[0] == MD_TEXT_NORMAL; substr_offsets[0] == 0) + * -- [1]: """ (substr_types[1] == MD_TEXT_ENTITY; substr_offsets[1] == 4) + * -- [2]: " bar" (substr_types[2] == MD_TEXT_NORMAL; substr_offsets[2] == 10) + * -- [3]: (n/a) (n/a ; substr_offsets[3] == 14) + * + * Note that these conditions are guaranteed: + * -- substr_offsets[0] == 0 + * -- substr_offsets[LAST+1] == size + * -- Only MD_TEXT_NORMAL, MD_TEXT_ENTITY, MD_TEXT_NULLCHAR substrings can appear. + */ +typedef struct MD_ATTRIBUTE { + const MD_CHAR* text; + MD_SIZE size; + const MD_TEXTTYPE* substr_types; + const MD_OFFSET* substr_offsets; +} MD_ATTRIBUTE; + + +/* Detailed info for MD_BLOCK_UL. */ +typedef struct MD_BLOCK_UL_DETAIL { + int is_tight; /* Non-zero if tight list, zero if loose. */ + MD_CHAR mark; /* Item bullet character in MarkDown source of the list, e.g. '-', '+', '*'. */ +} MD_BLOCK_UL_DETAIL; + +/* Detailed info for MD_BLOCK_OL. */ +typedef struct MD_BLOCK_OL_DETAIL { + unsigned start; /* Start index of the ordered list. */ + int is_tight; /* Non-zero if tight list, zero if loose. */ + MD_CHAR mark_delimiter; /* Character delimiting the item marks in MarkDown source, e.g. '.' or ')' */ +} MD_BLOCK_OL_DETAIL; + +/* Detailed info for MD_BLOCK_LI. */ +typedef struct MD_BLOCK_LI_DETAIL { + int is_task; /* Can be non-zero only with MD_FLAG_TASKLISTS */ + MD_CHAR task_mark; /* If is_task, then one of 'x', 'X' or ' '. Undefined otherwise. */ + MD_OFFSET task_mark_offset; /* If is_task, then offset in the input of the char between '[' and ']'. */ +} MD_BLOCK_LI_DETAIL; + +/* Detailed info for MD_BLOCK_H. */ +typedef struct MD_BLOCK_H_DETAIL { + unsigned level; /* Header level (1 - 6) */ +} MD_BLOCK_H_DETAIL; + +/* Detailed info for MD_BLOCK_CODE. */ +typedef struct MD_BLOCK_CODE_DETAIL { + MD_ATTRIBUTE info; + MD_ATTRIBUTE lang; + MD_CHAR fence_char; /* The character used for fenced code block; or zero for indented code block. */ +} MD_BLOCK_CODE_DETAIL; + +/* Detailed info for MD_BLOCK_TH and MD_BLOCK_TD. */ +typedef struct MD_BLOCK_TD_DETAIL { + MD_ALIGN align; +} MD_BLOCK_TD_DETAIL; + +/* Detailed info for MD_SPAN_A. */ +typedef struct MD_SPAN_A_DETAIL { + MD_ATTRIBUTE href; + MD_ATTRIBUTE title; +} MD_SPAN_A_DETAIL; + +/* Detailed info for MD_SPAN_IMG. */ +typedef struct MD_SPAN_IMG_DETAIL { + MD_ATTRIBUTE src; + MD_ATTRIBUTE title; +} MD_SPAN_IMG_DETAIL; + + +/* Flags specifying extensions/deviations from CommonMark specification. + * + * By default (when MD_RENDERER::flags == 0), we follow CommonMark specification. + * The following flags may allow some extensions or deviations from it. + */ +#define MD_FLAG_COLLAPSEWHITESPACE 0x0001 /* In MD_TEXT_NORMAL, collapse non-trivial whitespace into single ' ' */ +#define MD_FLAG_PERMISSIVEATXHEADERS 0x0002 /* Do not require space in ATX headers ( ###header ) */ +#define MD_FLAG_PERMISSIVEURLAUTOLINKS 0x0004 /* Recognize URLs as autolinks even without '<', '>' */ +#define MD_FLAG_PERMISSIVEEMAILAUTOLINKS 0x0008 /* Recognize e-mails as autolinks even without '<', '>' and 'mailto:' */ +#define MD_FLAG_NOINDENTEDCODEBLOCKS 0x0010 /* Disable indented code blocks. (Only fenced code works.) */ +#define MD_FLAG_NOHTMLBLOCKS 0x0020 /* Disable raw HTML blocks. */ +#define MD_FLAG_NOHTMLSPANS 0x0040 /* Disable raw HTML (inline). */ +#define MD_FLAG_TABLES 0x0100 /* Enable tables extension. */ +#define MD_FLAG_STRIKETHROUGH 0x0200 /* Enable strikethrough extension. */ +#define MD_FLAG_PERMISSIVEWWWAUTOLINKS 0x0400 /* Enable WWW autolinks (even without any scheme prefix, if they begin with 'www.') */ +#define MD_FLAG_TASKLISTS 0x0800 /* Enable task list extension. */ + +#define MD_FLAG_PERMISSIVEAUTOLINKS (MD_FLAG_PERMISSIVEEMAILAUTOLINKS | MD_FLAG_PERMISSIVEURLAUTOLINKS | MD_FLAG_PERMISSIVEWWWAUTOLINKS) +#define MD_FLAG_NOHTML (MD_FLAG_NOHTMLBLOCKS | MD_FLAG_NOHTMLSPANS) + +/* Convenient sets of flags corresponding to well-known Markdown dialects. + * + * Note we may only support subset of features of the referred dialect. + * The constant just enables those extensions which bring us as close as + * possible given what features we implement. + * + * ABI compatibility note: Meaning of these can change in time as new + * extensions, bringing the dialect closer to the original, are implemented. + */ +#define MD_DIALECT_COMMONMARK 0 +#define MD_DIALECT_GITHUB (MD_FLAG_PERMISSIVEAUTOLINKS | MD_FLAG_TABLES | MD_FLAG_STRIKETHROUGH | MD_FLAG_TASKLISTS) + +/* Renderer structure. + */ +typedef struct MD_PARSER { + /* Reserved. Set to zero. + */ + unsigned abi_version; + + /* Dialect options. Bitmask of MD_FLAG_xxxx values. + */ + unsigned flags; + + /* Caller-provided rendering callbacks. + * + * For some block/span types, more detailed information is provided in a + * type-specific structure pointed by the argument 'detail'. + * + * The last argument of all callbacks, 'userdata', is just propagated from + * md_parse() and is available for any use by the application. + * + * Note any strings provided to the callbacks as their arguments or as + * members of any detail structure are generally not zero-terminated. + * Application has take the respective size information into account. + * + * Callbacks may abort further parsing of the document by returning non-zero. + */ + int (*enter_block)(MD_BLOCKTYPE /*type*/, void* /*detail*/, void* /*userdata*/); + int (*leave_block)(MD_BLOCKTYPE /*type*/, void* /*detail*/, void* /*userdata*/); + + int (*enter_span)(MD_SPANTYPE /*type*/, void* /*detail*/, void* /*userdata*/); + int (*leave_span)(MD_SPANTYPE /*type*/, void* /*detail*/, void* /*userdata*/); + + int (*text)(MD_TEXTTYPE /*type*/, const MD_CHAR* /*text*/, MD_SIZE /*size*/, void* /*userdata*/); + + /* Debug callback. Optional (may be NULL). + * + * If provided and something goes wrong, this function gets called. + * This is intended for debugging and problem diagnosis for developers; + * it is not intended to provide any errors suitable for displaying to an + * end user. + */ + void (*debug_log)(const char* /*msg*/, void* /*userdata*/); + + /* Reserved. Set to NULL. + */ + void (*syntax)(void); +} MD_PARSER; + + +/* For backward compatibility. Do not use in new code. */ +typedef MD_PARSER MD_RENDERER; + + +/* Parse the Markdown document stored in the string 'text' of size 'size'. + * The renderer provides callbacks to be called during the parsing so the + * caller can render the document on the screen or convert the Markdown + * to another format. + * + * Zero is returned on success. If a runtime error occurs (e.g. a memory + * fails), -1 is returned. If the processing is aborted due any callback + * returning non-zero, md_parse() the return value of the callback is returned. + */ +int md_parse(const MD_CHAR* text, MD_SIZE size, const MD_PARSER* parser, void* userdata); + + +#ifdef __cplusplus + } /* extern "C" { */ +#endif + +#endif /* MD4C_MARKDOWN_H */ diff --git a/src/3rdparty/md4c/qt_attribution.json b/src/3rdparty/md4c/qt_attribution.json new file mode 100644 index 0000000000..fa0fd18853 --- /dev/null +++ b/src/3rdparty/md4c/qt_attribution.json @@ -0,0 +1,15 @@ +{ + "Id": "md4c", + "Name": "MD4C", + "QDocModule": "qtgui", + "QtUsage": "Optionally used in QTextDocument if configured with textmarkdownreader.", + + "Description": "A CommonMark-compliant Markdown parser.", + "Homepage": "https://github.com/mity/md4c", + "License": "MIT License", + "LicenseId": "MIT", + "LicenseFile": "LICENSE.md", + "Version": "0.3.3", + "DownloadLocation": "https://github.com/mity/md4c/releases/tag/release-0.3.3", + "Copyright": "Copyright © 2016-2019 Martin Mitáš" +} diff --git a/src/3rdparty/pcre2/patches/0001-fix-rtems-build-undefine-madvise.patch b/src/3rdparty/pcre2/patches/0001-fix-rtems-build-undefine-madvise.patch new file mode 100644 index 0000000000..074b39df85 --- /dev/null +++ b/src/3rdparty/pcre2/patches/0001-fix-rtems-build-undefine-madvise.patch @@ -0,0 +1,28 @@ +From ac10063196685fe6124055feb1275e13a78f562e Mon Sep 17 00:00:00 2001 +From: Mikhail Svetkin <mikhail.svetkin@qt.io> +Date: Tue, 20 Mar 2018 14:03:54 +0100 +Subject: [PATCH] rtems: Fix pcre2 build (madvise undefined) + +RTEMS does not have madvise. We can use only posix_madvise + +Change-Id: Ia18b7cd2d7f9db84331f7e2350d060b9e85b30c8 +--- + src/3rdparty/pcre2/src/sljit/sljitUtils.c | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/src/3rdparty/pcre2/src/sljit/sljitUtils.c b/src/3rdparty/pcre2/src/sljit/sljitUtils.c +index 5c2a838932..2ead044b1b 100644 +--- a/src/3rdparty/pcre2/src/sljit/sljitUtils.c ++++ b/src/3rdparty/pcre2/src/sljit/sljitUtils.c +@@ -315,7 +315,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_u8 *SLJIT_FUNC sljit_stack_resize(struct sljit_st + aligned_new_start = (sljit_uw)new_start & ~sljit_page_align; + aligned_old_start = ((sljit_uw)stack->start) & ~sljit_page_align; + /* If madvise is available, we release the unnecessary space. */ +-#if defined(MADV_DONTNEED) ++#if defined(MADV_DONTNEED) && !defined(__rtems__) + if (aligned_new_start > aligned_old_start) + madvise((void*)aligned_old_start, aligned_new_start - aligned_old_start, MADV_DONTNEED); + #elif defined(POSIX_MADV_DONTNEED) +-- +2.21.0 + diff --git a/src/3rdparty/pcre2/src/sljit/sljitUtils.c b/src/3rdparty/pcre2/src/sljit/sljitUtils.c index 5c2a838932..2ead044b1b 100644 --- a/src/3rdparty/pcre2/src/sljit/sljitUtils.c +++ b/src/3rdparty/pcre2/src/sljit/sljitUtils.c @@ -315,7 +315,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_u8 *SLJIT_FUNC sljit_stack_resize(struct sljit_st aligned_new_start = (sljit_uw)new_start & ~sljit_page_align; aligned_old_start = ((sljit_uw)stack->start) & ~sljit_page_align; /* If madvise is available, we release the unnecessary space. */ -#if defined(MADV_DONTNEED) +#if defined(MADV_DONTNEED) && !defined(__rtems__) if (aligned_new_start > aligned_old_start) madvise((void*)aligned_old_start, aligned_new_start - aligned_old_start, MADV_DONTNEED); #elif defined(POSIX_MADV_DONTNEED) diff --git a/src/3rdparty/sha3/brg_endian.h b/src/3rdparty/sha3/brg_endian.h index 09d2a8b6a9..9bb306e678 100644 --- a/src/3rdparty/sha3/brg_endian.h +++ b/src/3rdparty/sha3/brg_endian.h @@ -42,7 +42,7 @@ Changes for ARM 9/9/2010 [Downstream relative to Gladman's GitHub, upstream to Q #elif defined( __linux__ ) || defined( __GNUC__ ) || defined( __GNU_LIBRARY__ ) # if !defined( __MINGW32__ ) && !defined( _AIX ) && !defined(Q_OS_QNX) # include <endian.h> -# if !defined( __BEOS__ ) +# if !defined( __BEOS__ ) && !defined(Q_OS_RTEMS) # include <byteswap.h> # endif # endif diff --git a/src/3rdparty/sqlite.pri b/src/3rdparty/sqlite.pri index 068764c726..6405beb3d0 100644 --- a/src/3rdparty/sqlite.pri +++ b/src/3rdparty/sqlite.pri @@ -1,5 +1,6 @@ CONFIG(release, debug|release):DEFINES *= NDEBUG -DEFINES += SQLITE_ENABLE_COLUMN_METADATA SQLITE_OMIT_LOAD_EXTENSION SQLITE_OMIT_COMPLETE SQLITE_ENABLE_FTS3 SQLITE_ENABLE_FTS3_PARENTHESIS SQLITE_ENABLE_FTS5 SQLITE_ENABLE_RTREE SQLITE_ENABLE_JSON1 +QT_FOR_CONFIG += core-private +DEFINES += SQLITE_ENABLE_COLUMN_METADATA SQLITE_OMIT_COMPLETE SQLITE_ENABLE_FTS3 SQLITE_ENABLE_FTS3_PARENTHESIS SQLITE_ENABLE_FTS5 SQLITE_ENABLE_RTREE SQLITE_ENABLE_JSON1 !contains(CONFIG, largefile):DEFINES += SQLITE_DISABLE_LFS qtConfig(posix_fallocate): DEFINES += HAVE_POSIX_FALLOCATE=1 winrt { @@ -8,6 +9,7 @@ winrt { } qnx: DEFINES += _QNX_SOURCE !win32:!winrt:!winphone: DEFINES += HAVE_USLEEP=1 +qtConfig(dlopen): QMAKE_USE += libdl integrity: QMAKE_CFLAGS += -include qplatformdefs.h INCLUDEPATH += $$PWD/sqlite SOURCES += $$PWD/sqlite/sqlite3.c |