From c142f1124815a766455ca326074057e86e7501b8 Mon Sep 17 00:00:00 2001 From: Magnus Norddahl Date: Wed, 12 Apr 2023 01:23:06 +0200 Subject: [PATCH] Upgrade to latest version of zvulkan and hook up minAccelerationStructureScratchOffsetAlignment --- src/lightmap/gpuraytracer.cpp | 26 +- src/lightmap/gpuraytracer.h | 2 +- thirdparty/ZVulkan/CMakeLists.txt | 3 - .../zvulkan/vk_mem_alloc/vk_mem_alloc.h | 24036 ++++++++-------- .../ZVulkan/include/zvulkan/vulkanbuilders.h | 168 +- .../include/zvulkan/vulkancompatibledevice.h | 5 +- .../ZVulkan/include/zvulkan/vulkandevice.h | 4 +- .../ZVulkan/include/zvulkan/vulkaninstance.h | 43 +- .../ZVulkan/include/zvulkan/vulkanobjects.h | 48 +- .../ZVulkan/include/zvulkan/vulkansurface.h | 46 +- .../ZVulkan/include/zvulkan/vulkanswapchain.h | 67 +- thirdparty/ZVulkan/src/vulkanbuilders.cpp | 541 +- thirdparty/ZVulkan/src/vulkandevice.cpp | 20 +- thirdparty/ZVulkan/src/vulkaninstance.cpp | 55 +- thirdparty/ZVulkan/src/vulkansurface.cpp | 44 +- thirdparty/ZVulkan/src/vulkanswapchain.cpp | 605 +- 16 files changed, 13053 insertions(+), 12660 deletions(-) diff --git a/src/lightmap/gpuraytracer.cpp b/src/lightmap/gpuraytracer.cpp index 0163554..5493b32 100644 --- a/src/lightmap/gpuraytracer.cpp +++ b/src/lightmap/gpuraytracer.cpp @@ -80,9 +80,16 @@ GPURaytracer::GPURaytracer() if(!rdoc_api) LoadRenderDoc(); - auto instance = std::make_shared(VKDebug); - device = std::make_unique(instance, nullptr, VulkanCompatibleDevice::SelectDevice(instance, nullptr, 0)); + auto instance = VulkanInstanceBuilder() + .DebugLayer(VKDebug) + .Create(); + + device = VulkanDeviceBuilder() + .OptionalRayQuery() + .Create(instance); + useRayQuery = !NoRtx && device->PhysicalDevice.Features.RayQuery.rayQuery; + PrintVulkanInfo(); } @@ -609,6 +616,7 @@ void GPURaytracer::CreateBottomLevelAccelerationStructure() blScratchBuffer = BufferBuilder() .Usage(VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) .Size(sizeInfo.buildScratchSize) + .MinAlignment(device->PhysicalDevice.Properties.AccelerationStructure.minAccelerationStructureScratchOffsetAlignment) .DebugName("blScratchBuffer") .Create(device.get()); @@ -691,6 +699,7 @@ void GPURaytracer::CreateTopLevelAccelerationStructure() tlScratchBuffer = BufferBuilder() .Usage(VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) .Size(sizeInfo.buildScratchSize) + .MinAlignment(device->PhysicalDevice.Properties.AccelerationStructure.minAccelerationStructureScratchOffsetAlignment) .DebugName("tlScratchBuffer") .Create(device.get()); @@ -717,17 +726,20 @@ void GPURaytracer::CreateShaders() traceprefix += "#line 1\r\n"; vertShader = ShaderBuilder() - .VertexShader(prefix + glsl_vert) + .Type(ShaderType::Vertex) + .AddSource("vert.glsl", prefix + glsl_vert) .DebugName("vertShader") .Create("vertShader", device.get()); fragShader = ShaderBuilder() - .FragmentShader(traceprefix + glsl_frag) + .Type(ShaderType::Fragment) + .AddSource("frag.glsl", traceprefix + glsl_frag) .DebugName("fragShader") .Create("fragShader", device.get()); fragResolveShader = ShaderBuilder() - .FragmentShader(prefix + glsl_frag_resolve) + .Type(ShaderType::Fragment) + .AddSource("frag_resolve.glsl", prefix + glsl_frag_resolve) .DebugName("fragResolveShader") .Create("fragResolveShader", device.get()); } @@ -964,7 +976,7 @@ LightmapImage GPURaytracer::CreateImage(int width, int height) void GPURaytracer::CreateUniformBuffer() { - VkDeviceSize align = device->PhysicalDevice.Properties.limits.minUniformBufferOffsetAlignment; + VkDeviceSize align = device->PhysicalDevice.Properties.Properties.limits.minUniformBufferOffsetAlignment; uniformStructStride = (sizeof(Uniforms) + align - 1) / align * align; uniformBuffer = BufferBuilder() @@ -1047,7 +1059,7 @@ std::vector GPURaytracer::CreateCollisionNodes() void GPURaytracer::PrintVulkanInfo() { - const auto& props = device->PhysicalDevice.Properties; + const auto& props = device->PhysicalDevice.Properties.Properties; std::string deviceType; switch (props.deviceType) diff --git a/src/lightmap/gpuraytracer.h b/src/lightmap/gpuraytracer.h index ec9b6ce..d536e9e 100644 --- a/src/lightmap/gpuraytracer.h +++ b/src/lightmap/gpuraytracer.h @@ -154,7 +154,7 @@ private: int uniformStructs = 256; VkDeviceSize uniformStructStride = sizeof(Uniforms); - std::unique_ptr device; + std::shared_ptr device; bool useRayQuery = true; diff --git a/thirdparty/ZVulkan/CMakeLists.txt b/thirdparty/ZVulkan/CMakeLists.txt index 7c00ce0..8a99d6a 100644 --- a/thirdparty/ZVulkan/CMakeLists.txt +++ b/thirdparty/ZVulkan/CMakeLists.txt @@ -3,7 +3,6 @@ project(zvulkan) set(ZVULKAN_SOURCES src/vulkanbuilders.cpp - src/vulkancompatibledevice.cpp src/vulkandevice.cpp src/vulkaninstance.cpp src/vulkansurface.cpp @@ -171,8 +170,6 @@ source_group("include" REGULAR_EXPRESSION "${CMAKE_CURRENT_SOURCE_DIR}/include/z source_group("include\\vk_mem_alloc" REGULAR_EXPRESSION "${CMAKE_CURRENT_SOURCE_DIR}/include/zvulkan/vk_mem_alloc/.+") source_group("include\\volk" REGULAR_EXPRESSION "${CMAKE_CURRENT_SOURCE_DIR}/include/zvulkan/volk/.+") -source_group("example" REGULAR_EXPRESSION "${CMAKE_CURRENT_SOURCE_DIR}/example/.+") - include_directories(include include/zvulkan src) if(WIN32) diff --git a/thirdparty/ZVulkan/include/zvulkan/vk_mem_alloc/vk_mem_alloc.h b/thirdparty/ZVulkan/include/zvulkan/vk_mem_alloc/vk_mem_alloc.h index fd44722..7b04e54 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vk_mem_alloc/vk_mem_alloc.h +++ b/thirdparty/ZVulkan/include/zvulkan/vk_mem_alloc/vk_mem_alloc.h @@ -1,5 +1,5 @@ // -// Copyright (c) 2017-2021 Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2017-2022 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 @@ -25,12 +25,12 @@ /** \mainpage Vulkan Memory Allocator -Version 3.0.0-development (2021-06-21) +Version 3.0.1 (2022-05-26) -Copyright (c) 2017-2021 Advanced Micro Devices, Inc. All rights reserved. \n +Copyright (c) 2017-2022 Advanced Micro Devices, Inc. All rights reserved. \n License: MIT -Documentation of all members: vk_mem_alloc.h +API documentation divided into groups: [Modules](modules.html) \section main_table_of_contents Table of contents @@ -49,7 +49,6 @@ Documentation of all members: vk_mem_alloc.h - [Mapping functions](@ref memory_mapping_mapping_functions) - [Persistently mapped memory](@ref memory_mapping_persistently_mapped_memory) - [Cache flush and invalidate](@ref memory_mapping_cache_control) - - [Finding out if memory is mappable](@ref memory_mapping_finding_if_memory_mappable) - \subpage staying_within_budget - [Querying for budget](@ref staying_within_budget_querying_for_budget) - [Controlling memory usage](@ref staying_within_budget_controlling_memory_usage) @@ -61,59 +60,89 @@ Documentation of all members: vk_mem_alloc.h - [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 virtual_allocator - \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 opengl_interop - \subpage usage_patterns - - [Common mistakes](@ref usage_patterns_common_mistakes) - - [Simple patterns](@ref usage_patterns_simple) - - [Advanced patterns](@ref usage_patterns_advanced) + - [GPU-only resource](@ref usage_patterns_gpu_only) + - [Staging copy for upload](@ref usage_patterns_staging_copy_upload) + - [Readback](@ref usage_patterns_readback) + - [Advanced data uploading](@ref usage_patterns_advanced_data_uploading) + - [Other use cases](@ref usage_patterns_other_use_cases) - \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 enabling_buffer_device_address - - \subpage vk_amd_device_coherent_memory +- Extension support + - \subpage vk_khr_dedicated_allocation + - \subpage enabling_buffer_device_address + - \subpage vk_ext_memory_priority + - \subpage vk_amd_device_coherent_memory - \subpage general_considerations - [Thread safety](@ref general_considerations_thread_safety) + - [Versioning and compatibility](@ref general_considerations_versioning_and_compatibility) - [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) +- [**Product page on GPUOpen**](https://gpuopen.com/gaming-product/vulkan-memory-allocator/) +- [**Source repository on GitHub**](https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator) + +\defgroup group_init Library initialization + +\brief API elements related to the initialization and management of the entire library, especially #VmaAllocator object. + +\defgroup group_alloc Memory allocation + +\brief API elements related to the allocation, deallocation, and management of Vulkan memory, buffers, images. +Most basic ones being: vmaCreateBuffer(), vmaCreateImage(). + +\defgroup group_virtual Virtual allocator + +\brief API elements related to the mechanism of \ref virtual_allocator - using the core allocation algorithm +for user-defined purpose without allocating any real GPU memory. + +\defgroup group_stats Statistics + +\brief API elements that query current status of the allocator, from memory usage, budget, to full dump of the internal state in JSON format. +See documentation chapter: \ref statistics. */ + #ifdef __cplusplus extern "C" { #endif -/* -Define this macro to 0/1 to disable/enable support for recording functionality, -available through VmaAllocatorCreateInfo::pRecordSettings. -*/ -#ifndef VMA_RECORDING_ENABLED - #define VMA_RECORDING_ENABLED 0 +#ifndef VULKAN_H_ + #include +#endif + +// Define this macro to declare maximum supported Vulkan version in format AAABBBCCC, +// where AAA = major, BBB = minor, CCC = patch. +// If you want to use version > 1.0, it still needs to be enabled via VmaAllocatorCreateInfo::vulkanApiVersion. +#if !defined(VMA_VULKAN_VERSION) + #if defined(VK_VERSION_1_3) + #define VMA_VULKAN_VERSION 1003000 + #elif defined(VK_VERSION_1_2) + #define VMA_VULKAN_VERSION 1002000 + #elif defined(VK_VERSION_1_1) + #define VMA_VULKAN_VERSION 1001000 + #else + #define VMA_VULKAN_VERSION 1000000 + #endif #endif #if defined(__ANDROID__) && defined(VK_NO_PROTOTYPES) && VMA_STATIC_VULKAN_FUNCTIONS @@ -145,23 +174,6 @@ available through VmaAllocatorCreateInfo::pRecordSettings. #endif // #if VMA_VULKAN_VERSION >= 1001000 #endif // #if defined(__ANDROID__) && VMA_STATIC_VULKAN_FUNCTIONS && VK_NO_PROTOTYPES -#ifndef VULKAN_H_ - #include -#endif - -// Define this macro to declare maximum supported Vulkan version in format AAABBBCCC, -// where AAA = major, BBB = minor, CCC = patch. -// If you want to use version > 1.0, it still needs to be enabled via VmaAllocatorCreateInfo::vulkanApiVersion. -#if !defined(VMA_VULKAN_VERSION) - #if defined(VK_VERSION_1_2) - #define VMA_VULKAN_VERSION 1002000 - #elif defined(VK_VERSION_1_1) - #define VMA_VULKAN_VERSION 1001000 - #else - #define VMA_VULKAN_VERSION 1000000 - #endif -#endif - #if !defined(VMA_DEDICATED_ALLOCATION) #if VK_KHR_get_memory_requirements2 && VK_KHR_dedicated_allocation #define VMA_DEDICATED_ALLOCATION 1 @@ -278,50 +290,29 @@ available through VmaAllocatorCreateInfo::pRecordSettings. #endif #endif -/** \struct VmaAllocator -\brief Represents main object of this library initialized. +#ifndef VMA_STATS_STRING_ENABLED + #define VMA_STATS_STRING_ENABLED 1 +#endif -Fill structure #VmaAllocatorCreateInfo and call function vmaCreateAllocator() to create it. -Call function vmaDestroyAllocator() to destroy it. +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// +// +// INTERFACE +// +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// -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. +// Sections for managing code placement in file, only for development purposes e.g. for convenient folding inside an IDE. +#ifndef _VMA_ENUM_DECLARATIONS + +/** +\addtogroup group_init +@{ */ -VK_DEFINE_HANDLE(VmaAllocator) - -/// Callback function called after successful vkAllocateMemory. -typedef void (VKAPI_PTR *PFN_vmaAllocateDeviceMemoryFunction)( - VmaAllocator VMA_NOT_NULL allocator, - uint32_t memoryType, - VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory, - VkDeviceSize size, - void* VMA_NULLABLE pUserData); -/// Callback function called before vkFreeMemory. -typedef void (VKAPI_PTR *PFN_vmaFreeDeviceMemoryFunction)( - VmaAllocator VMA_NOT_NULL allocator, - uint32_t memoryType, - VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory, - VkDeviceSize size, - void* VMA_NULLABLE pUserData); - -/** \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 VMA_NULLABLE pfnAllocate; - /// Optional, can be null. - PFN_vmaFreeDeviceMemoryFunction VMA_NULLABLE pfnFree; - /// Optional, can be null. - void* VMA_NULLABLE pUserData; -} VmaDeviceMemoryCallbacks; /// Flags for created #VmaAllocator. -typedef enum VmaAllocatorCreateFlagBits { +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. @@ -435,13 +426,536 @@ typedef enum VmaAllocatorCreateFlagBits { VMA_ALLOCATOR_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF } VmaAllocatorCreateFlagBits; +/// See #VmaAllocatorCreateFlagBits. typedef VkFlags VmaAllocatorCreateFlags; +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + +/// \brief Intended usage of the allocated memory. +typedef enum VmaMemoryUsage +{ + /** No intended memory usage specified. + Use other members of VmaAllocationCreateInfo to specify your requirements. + */ + VMA_MEMORY_USAGE_UNKNOWN = 0, + /** + \deprecated Obsolete, preserved for backward compatibility. + Prefers `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. + */ + VMA_MEMORY_USAGE_GPU_ONLY = 1, + /** + \deprecated Obsolete, preserved for backward compatibility. + Guarantees `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` and `VK_MEMORY_PROPERTY_HOST_COHERENT_BIT`. + */ + VMA_MEMORY_USAGE_CPU_ONLY = 2, + /** + \deprecated Obsolete, preserved for backward compatibility. + Guarantees `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT`, prefers `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. + */ + VMA_MEMORY_USAGE_CPU_TO_GPU = 3, + /** + \deprecated Obsolete, preserved for backward compatibility. + Guarantees `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT`, prefers `VK_MEMORY_PROPERTY_HOST_CACHED_BIT`. + */ + VMA_MEMORY_USAGE_GPU_TO_CPU = 4, + /** + \deprecated Obsolete, preserved for backward compatibility. + Prefers not `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. + */ + VMA_MEMORY_USAGE_CPU_COPY = 5, + /** + Lazily allocated GPU memory having `VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT`. + Exists mostly on mobile platforms. Using it on desktop PC or other GPUs with no such memory type present will fail the allocation. + + Usage: Memory for transient attachment images (color attachments, depth attachments etc.), created with `VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT`. + + Allocations with this usage are always created as dedicated - it implies #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. + */ + VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED = 6, + /** + Selects best memory type automatically. + This flag is recommended for most common use cases. + + When using this flag, if you want to map the allocation (using vmaMapMemory() or #VMA_ALLOCATION_CREATE_MAPPED_BIT), + you must pass one of the flags: #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT + in VmaAllocationCreateInfo::flags. + + It can be used only with functions that let the library know `VkBufferCreateInfo` or `VkImageCreateInfo`, e.g. + vmaCreateBuffer(), vmaCreateImage(), vmaFindMemoryTypeIndexForBufferInfo(), vmaFindMemoryTypeIndexForImageInfo() + and not with generic memory allocation functions. + */ + VMA_MEMORY_USAGE_AUTO = 7, + /** + Selects best memory type automatically with preference for GPU (device) memory. + + When using this flag, if you want to map the allocation (using vmaMapMemory() or #VMA_ALLOCATION_CREATE_MAPPED_BIT), + you must pass one of the flags: #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT + in VmaAllocationCreateInfo::flags. + + It can be used only with functions that let the library know `VkBufferCreateInfo` or `VkImageCreateInfo`, e.g. + vmaCreateBuffer(), vmaCreateImage(), vmaFindMemoryTypeIndexForBufferInfo(), vmaFindMemoryTypeIndexForImageInfo() + and not with generic memory allocation functions. + */ + VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE = 8, + /** + Selects best memory type automatically with preference for CPU (host) memory. + + When using this flag, if you want to map the allocation (using vmaMapMemory() or #VMA_ALLOCATION_CREATE_MAPPED_BIT), + you must pass one of the flags: #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT + in VmaAllocationCreateInfo::flags. + + It can be used only with functions that let the library know `VkBufferCreateInfo` or `VkImageCreateInfo`, e.g. + vmaCreateBuffer(), vmaCreateImage(), vmaFindMemoryTypeIndexForBufferInfo(), vmaFindMemoryTypeIndexForImageInfo() + and not with generic memory allocation functions. + */ + VMA_MEMORY_USAGE_AUTO_PREFER_HOST = 9, + + 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. + */ + 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. + */ + 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. + + It is 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). + */ + VMA_ALLOCATION_CREATE_MAPPED_BIT = 0x00000004, + /** \deprecated Preserved for backward compatibility. Consider using vmaSetAllocationName() instead. + + 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 `pName`. 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, + /** Create both buffer/image and allocation, but don't bind them together. + It is useful when you want to bind yourself to do some more advanced binding, e.g. using some extensions. + The flag is meaningful only with functions that bind by default: vmaCreateBuffer(), vmaCreateImage(). + Otherwise it is ignored. + + If you want to make sure the new buffer/image is not tied to the new memory allocation + through `VkMemoryDedicatedAllocateInfoKHR` structure in case the allocation ends up in its own memory block, + use also flag #VMA_ALLOCATION_CREATE_CAN_ALIAS_BIT. + */ + VMA_ALLOCATION_CREATE_DONT_BIND_BIT = 0x00000080, + /** Create allocation only if additional device memory required for it, if any, won't exceed + memory budget. Otherwise return `VK_ERROR_OUT_OF_DEVICE_MEMORY`. + */ + VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT = 0x00000100, + /** \brief Set this flag if the allocated memory will have aliasing resources. + + Usage of this flag prevents supplying `VkMemoryDedicatedAllocateInfoKHR` when #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT is specified. + Otherwise created dedicated memory will not be suitable for aliasing resources, resulting in Vulkan Validation Layer errors. + */ + VMA_ALLOCATION_CREATE_CAN_ALIAS_BIT = 0x00000200, + /** + Requests possibility to map the allocation (using vmaMapMemory() or #VMA_ALLOCATION_CREATE_MAPPED_BIT). + + - If you use #VMA_MEMORY_USAGE_AUTO or other `VMA_MEMORY_USAGE_AUTO*` value, + you must use this flag to be able to map the allocation. Otherwise, mapping is incorrect. + - If you use other value of #VmaMemoryUsage, this flag is ignored and mapping is always possible in memory types that are `HOST_VISIBLE`. + This includes allocations created in \ref custom_memory_pools. + + Declares that mapped memory will only be written sequentially, e.g. using `memcpy()` or a loop writing number-by-number, + never read or accessed randomly, so a memory type can be selected that is uncached and write-combined. + + \warning Violating this declaration may work correctly, but will likely be very slow. + Watch out for implicit reads introduced by doing e.g. `pMappedData[i] += x;` + Better prepare your data in a local variable and `memcpy()` it to the mapped pointer all at once. + */ + VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT = 0x00000400, + /** + Requests possibility to map the allocation (using vmaMapMemory() or #VMA_ALLOCATION_CREATE_MAPPED_BIT). + + - If you use #VMA_MEMORY_USAGE_AUTO or other `VMA_MEMORY_USAGE_AUTO*` value, + you must use this flag to be able to map the allocation. Otherwise, mapping is incorrect. + - If you use other value of #VmaMemoryUsage, this flag is ignored and mapping is always possible in memory types that are `HOST_VISIBLE`. + This includes allocations created in \ref custom_memory_pools. + + Declares that mapped memory can be read, written, and accessed in random order, + so a `HOST_CACHED` memory type is required. + */ + VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT = 0x00000800, + /** + Together with #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT, + it says that despite request for host access, a not-`HOST_VISIBLE` memory type can be selected + if it may improve performance. + + By using this flag, you declare that you will check if the allocation ended up in a `HOST_VISIBLE` memory type + (e.g. using vmaGetAllocationMemoryProperties()) and if not, you will create some "staging" buffer and + issue an explicit transfer to write/read your data. + To prepare for this possibility, don't forget to add appropriate flags like + `VK_BUFFER_USAGE_TRANSFER_DST_BIT`, `VK_BUFFER_USAGE_TRANSFER_SRC_BIT` to the parameters of created buffer or image. + */ + VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT = 0x00001000, + /** Allocation strategy that chooses smallest possible free range for the allocation + to minimize memory usage and fragmentation, possibly at the expense of allocation time. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT = 0x00010000, + /** Allocation strategy that chooses first suitable free range for the allocation - + not necessarily in terms of the smallest offset but the one that is easiest and fastest to find + to minimize allocation time, possibly at the expense of allocation quality. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT = 0x00020000, + /** Allocation strategy that chooses always the lowest offset in available space. + This is not the most efficient strategy but achieves highly packed data. + Used internally by defragmentation, not recomended in typical usage. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT = 0x00040000, + /** Alias to #VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT. + */ + VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT = VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT, + /** Alias to #VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT. + */ + VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT = VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT, + /** A bit mask to extract only `STRATEGY` bits from entire set of flags. + */ + VMA_ALLOCATION_CREATE_STRATEGY_MASK = + VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT | + VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT | + VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT, + + VMA_ALLOCATION_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaAllocationCreateFlagBits; +/// See #VmaAllocationCreateFlagBits. +typedef VkFlags VmaAllocationCreateFlags; + +/// 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. + */ + VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT = 0x00000004, + + /** 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_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaPoolCreateFlagBits; +/// Flags to be passed as VmaPoolCreateInfo::flags. See #VmaPoolCreateFlagBits. +typedef VkFlags VmaPoolCreateFlags; + +/// Flags to be passed as VmaDefragmentationInfo::flags. +typedef enum VmaDefragmentationFlagBits +{ + /* \brief Use simple but fast algorithm for defragmentation. + May not achieve best results but will require least time to compute and least allocations to copy. + */ + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT = 0x1, + /* \brief Default defragmentation algorithm, applied also when no `ALGORITHM` flag is specified. + Offers a balance between defragmentation quality and the amount of allocations and bytes that need to be moved. + */ + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT = 0x2, + /* \brief Perform full defragmentation of memory. + Can result in notably more time to compute and allocations to copy, but will achieve best memory packing. + */ + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FULL_BIT = 0x4, + /** \brief Use the most roboust algorithm at the cost of time to compute and number of copies to make. + Only available when bufferImageGranularity is greater than 1, since it aims to reduce + alignment issues between different types of resources. + Otherwise falls back to same behavior as #VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FULL_BIT. + */ + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT = 0x8, + + /// A bit mask to extract only `ALGORITHM` bits from entire set of flags. + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_MASK = + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT | + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT | + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FULL_BIT | + VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT, + + VMA_DEFRAGMENTATION_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaDefragmentationFlagBits; +/// See #VmaDefragmentationFlagBits. +typedef VkFlags VmaDefragmentationFlags; + +/// Operation performed on single defragmentation move. See structure #VmaDefragmentationMove. +typedef enum VmaDefragmentationMoveOperation +{ + /// Buffer/image has been recreated at `dstTmpAllocation`, data has been copied, old buffer/image has been destroyed. `srcAllocation` should be changed to point to the new place. This is the default value set by vmaBeginDefragmentationPass(). + VMA_DEFRAGMENTATION_MOVE_OPERATION_COPY = 0, + /// Set this value if you cannot move the allocation. New place reserved at `dstTmpAllocation` will be freed. `srcAllocation` will remain unchanged. + VMA_DEFRAGMENTATION_MOVE_OPERATION_IGNORE = 1, + /// Set this value if you decide to abandon the allocation and you destroyed the buffer/image. New place reserved at `dstTmpAllocation` will be freed, along with `srcAllocation`, which will be destroyed. + VMA_DEFRAGMENTATION_MOVE_OPERATION_DESTROY = 2, +} VmaDefragmentationMoveOperation; + +/** @} */ + +/** +\addtogroup group_virtual +@{ +*/ + +/// Flags to be passed as VmaVirtualBlockCreateInfo::flags. +typedef enum VmaVirtualBlockCreateFlagBits +{ + /** \brief Enables alternative, linear allocation algorithm in this virtual block. + + 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. + */ + VMA_VIRTUAL_BLOCK_CREATE_LINEAR_ALGORITHM_BIT = 0x00000001, + + /** \brief Bit mask to extract only `ALGORITHM` bits from entire set of flags. + */ + VMA_VIRTUAL_BLOCK_CREATE_ALGORITHM_MASK = + VMA_VIRTUAL_BLOCK_CREATE_LINEAR_ALGORITHM_BIT, + + VMA_VIRTUAL_BLOCK_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaVirtualBlockCreateFlagBits; +/// Flags to be passed as VmaVirtualBlockCreateInfo::flags. See #VmaVirtualBlockCreateFlagBits. +typedef VkFlags VmaVirtualBlockCreateFlags; + +/// Flags to be passed as VmaVirtualAllocationCreateInfo::flags. +typedef enum VmaVirtualAllocationCreateFlagBits +{ + /** \brief Allocation will be created from upper stack in a double stack pool. + + This flag is only allowed for virtual blocks created with #VMA_VIRTUAL_BLOCK_CREATE_LINEAR_ALGORITHM_BIT flag. + */ + VMA_VIRTUAL_ALLOCATION_CREATE_UPPER_ADDRESS_BIT = VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT, + /** \brief Allocation strategy that tries to minimize memory usage. + */ + VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT = VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT, + /** \brief Allocation strategy that tries to minimize allocation time. + */ + VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT = VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT, + /** Allocation strategy that chooses always the lowest offset in available space. + This is not the most efficient strategy but achieves highly packed data. + */ + VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT = VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT, + /** \brief A bit mask to extract only `STRATEGY` bits from entire set of flags. + + These strategy flags are binary compatible with equivalent flags in #VmaAllocationCreateFlagBits. + */ + VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MASK = VMA_ALLOCATION_CREATE_STRATEGY_MASK, + + VMA_VIRTUAL_ALLOCATION_CREATE_FLAG_BITS_MAX_ENUM = 0x7FFFFFFF +} VmaVirtualAllocationCreateFlagBits; +/// Flags to be passed as VmaVirtualAllocationCreateInfo::flags. See #VmaVirtualAllocationCreateFlagBits. +typedef VkFlags VmaVirtualAllocationCreateFlags; + +/** @} */ + +#endif // _VMA_ENUM_DECLARATIONS + +#ifndef _VMA_DATA_TYPES_DECLARATIONS + +/** +\addtogroup group_init +@{ */ + +/** \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) + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + +/** \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) + +/** \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. +*/ +VK_DEFINE_HANDLE(VmaAllocation) + +/** \struct VmaDefragmentationContext +\brief An opaque object that represents started defragmentation process. + +Fill structure #VmaDefragmentationInfo and call function vmaBeginDefragmentation() to create it. +Call function vmaEndDefragmentation() to destroy it. +*/ +VK_DEFINE_HANDLE(VmaDefragmentationContext) + +/** @} */ + +/** +\addtogroup group_virtual +@{ +*/ + +/** \struct VmaVirtualAllocation +\brief Represents single memory allocation done inside VmaVirtualBlock. + +Use it as a unique identifier to virtual allocation within the single block. + +Use value `VK_NULL_HANDLE` to represent a null/invalid allocation. +*/ +VK_DEFINE_NON_DISPATCHABLE_HANDLE(VmaVirtualAllocation); + +/** @} */ + +/** +\addtogroup group_virtual +@{ +*/ + +/** \struct VmaVirtualBlock +\brief Handle to a virtual block object that allows to use core allocation algorithm without allocating any real GPU memory. + +Fill in #VmaVirtualBlockCreateInfo structure and use vmaCreateVirtualBlock() to create it. Use vmaDestroyVirtualBlock() to destroy it. +For more information, see documentation chapter \ref virtual_allocator. + +This object is not thread-safe - should not be used from multiple threads simultaneously, must be synchronized externally. +*/ +VK_DEFINE_HANDLE(VmaVirtualBlock) + +/** @} */ + +/** +\addtogroup group_init +@{ +*/ + +/// Callback function called after successful vkAllocateMemory. +typedef void (VKAPI_PTR* PFN_vmaAllocateDeviceMemoryFunction)( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t memoryType, + VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory, + VkDeviceSize size, + void* VMA_NULLABLE pUserData); + +/// Callback function called before vkFreeMemory. +typedef void (VKAPI_PTR* PFN_vmaFreeDeviceMemoryFunction)( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t memoryType, + VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory, + VkDeviceSize size, + void* VMA_NULLABLE pUserData); + +/** \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 VMA_NULLABLE pfnAllocate; + /// Optional, can be null. + PFN_vmaFreeDeviceMemoryFunction VMA_NULLABLE pfnFree; + /// Optional, can be null. + void* VMA_NULLABLE pUserData; +} VmaDeviceMemoryCallbacks; + /** \brief Pointers to some Vulkan functions - a subset used by the library. Used in VmaAllocatorCreateInfo::pVulkanFunctions. */ -typedef struct VmaVulkanFunctions { +typedef struct VmaVulkanFunctions +{ + /// Required when using VMA_DYNAMIC_VULKAN_FUNCTIONS. + PFN_vkGetInstanceProcAddr VMA_NULLABLE vkGetInstanceProcAddr; + /// Required when using VMA_DYNAMIC_VULKAN_FUNCTIONS. + PFN_vkGetDeviceProcAddr VMA_NULLABLE vkGetDeviceProcAddr; PFN_vkGetPhysicalDeviceProperties VMA_NULLABLE vkGetPhysicalDeviceProperties; PFN_vkGetPhysicalDeviceMemoryProperties VMA_NULLABLE vkGetPhysicalDeviceMemoryProperties; PFN_vkAllocateMemory VMA_NULLABLE vkAllocateMemory; @@ -460,46 +974,28 @@ typedef struct VmaVulkanFunctions { PFN_vkDestroyImage VMA_NULLABLE vkDestroyImage; PFN_vkCmdCopyBuffer VMA_NULLABLE vkCmdCopyBuffer; #if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + /// Fetch "vkGetBufferMemoryRequirements2" on Vulkan >= 1.1, fetch "vkGetBufferMemoryRequirements2KHR" when using VK_KHR_dedicated_allocation extension. PFN_vkGetBufferMemoryRequirements2KHR VMA_NULLABLE vkGetBufferMemoryRequirements2KHR; + /// Fetch "vkGetImageMemoryRequirements2" on Vulkan >= 1.1, fetch "vkGetImageMemoryRequirements2KHR" when using VK_KHR_dedicated_allocation extension. PFN_vkGetImageMemoryRequirements2KHR VMA_NULLABLE vkGetImageMemoryRequirements2KHR; #endif #if VMA_BIND_MEMORY2 || VMA_VULKAN_VERSION >= 1001000 + /// Fetch "vkBindBufferMemory2" on Vulkan >= 1.1, fetch "vkBindBufferMemory2KHR" when using VK_KHR_bind_memory2 extension. PFN_vkBindBufferMemory2KHR VMA_NULLABLE vkBindBufferMemory2KHR; + /// Fetch "vkBindImageMemory2" on Vulkan >= 1.1, fetch "vkBindImageMemory2KHR" when using VK_KHR_bind_memory2 extension. PFN_vkBindImageMemory2KHR VMA_NULLABLE vkBindImageMemory2KHR; #endif #if VMA_MEMORY_BUDGET || VMA_VULKAN_VERSION >= 1001000 PFN_vkGetPhysicalDeviceMemoryProperties2KHR VMA_NULLABLE vkGetPhysicalDeviceMemoryProperties2KHR; #endif +#if VMA_VULKAN_VERSION >= 1003000 + /// Fetch from "vkGetDeviceBufferMemoryRequirements" on Vulkan >= 1.3, but you can also fetch it from "vkGetDeviceBufferMemoryRequirementsKHR" if you enabled extension VK_KHR_maintenance4. + PFN_vkGetDeviceBufferMemoryRequirements VMA_NULLABLE vkGetDeviceBufferMemoryRequirements; + /// Fetch from "vkGetDeviceImageMemoryRequirements" on Vulkan >= 1.3, but you can also fetch it from "vkGetDeviceImageMemoryRequirementsKHR" if you enabled extension VK_KHR_maintenance4. + PFN_vkGetDeviceImageMemoryRequirements VMA_NULLABLE vkGetDeviceImageMemoryRequirements; +#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* VMA_NOT_NULL pFilePath; -} VmaRecordSettings; - /// Description of a Allocator to be created. typedef struct VmaAllocatorCreateInfo { @@ -520,20 +1016,6 @@ typedef struct VmaAllocatorCreateInfo /// Informative callbacks for `vkAllocateMemory`, `vkFreeMemory`. Optional. /** Optional, can be null. */ const VmaDeviceMemoryCallbacks* VMA_NULLABLE 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 @@ -565,13 +1047,6 @@ typedef struct VmaAllocatorCreateInfo For details see [Pointers to Vulkan functions](@ref config_Vulkan_functions). */ const VmaVulkanFunctions* VMA_NULLABLE 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* VMA_NULLABLE pRecordSettings; /** \brief Handle to Vulkan instance object. Starting from version 3.0.0 this member is no longer optional, it must be set! @@ -582,7 +1057,7 @@ typedef struct VmaAllocatorCreateInfo It must be a value in the format as created by macro `VK_MAKE_VERSION` or a constant like: `VK_API_VERSION_1_1`, `VK_API_VERSION_1_0`. The patch version number specified is ignored. Only the major and minor versions are considered. It must be less or equal (preferably equal) to value as passed to `vkCreateInstance` as `VkApplicationInfo::apiVersion`. - Only versions 1.0, 1.1, 1.2 are supported by the current implementation. + Only versions 1.0, 1.1, 1.2, 1.3 are supported by the current implementation. Leaving it initialized to zero is equivalent to `VK_API_VERSION_1_0`. */ uint32_t vulkanApiVersion; @@ -600,17 +1075,7 @@ typedef struct VmaAllocatorCreateInfo #endif // #if VMA_EXTERNAL_MEMORY } VmaAllocatorCreateInfo; -/// Creates Allocator object. -VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAllocator( - const VmaAllocatorCreateInfo* VMA_NOT_NULL pCreateInfo, - VmaAllocator VMA_NULLABLE * VMA_NOT_NULL pAllocator); - -/// Destroys allocator object. -VMA_CALL_PRE void VMA_CALL_POST vmaDestroyAllocator( - VmaAllocator VMA_NULLABLE allocator); - -/** \brief Information about existing #VmaAllocator object. -*/ +/// Information about existing #VmaAllocator object. typedef struct VmaAllocatorInfo { /** \brief Handle to Vulkan instance object. @@ -630,359 +1095,129 @@ typedef struct VmaAllocatorInfo VkDevice VMA_NOT_NULL device; } VmaAllocatorInfo; -/** \brief Returns information about existing #VmaAllocator object - handle to Vulkan device etc. - -It might be useful if you want to keep just the #VmaAllocator handle and fetch other required handles to -`VkPhysicalDevice`, `VkDevice` etc. every time using this function. -*/ -VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocatorInfo(VmaAllocator VMA_NOT_NULL allocator, VmaAllocatorInfo* VMA_NOT_NULL pAllocatorInfo); +/** @} */ /** -PhysicalDeviceProperties are fetched from physicalDevice by the allocator. -You can access it here, without fetching it again on your own. +\addtogroup group_stats +@{ */ -VMA_CALL_PRE void VMA_CALL_POST vmaGetPhysicalDeviceProperties( - VmaAllocator VMA_NOT_NULL allocator, - const VkPhysicalDeviceProperties* VMA_NULLABLE * VMA_NOT_NULL ppPhysicalDeviceProperties); -/** -PhysicalDeviceMemoryProperties are fetched from physicalDevice by the allocator. -You can access it here, without fetching it again on your own. +/** \brief Calculated statistics of memory usage e.g. in a specific memory type, heap, custom pool, or total. + +These are fast to calculate. +See functions: vmaGetHeapBudgets(), vmaGetPoolStatistics(). */ -VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryProperties( - VmaAllocator VMA_NOT_NULL allocator, - const VkPhysicalDeviceMemoryProperties* VMA_NULLABLE * VMA_NOT_NULL 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(). -*/ -VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryTypeProperties( - VmaAllocator VMA_NOT_NULL allocator, - uint32_t memoryTypeIndex, - VkMemoryPropertyFlags* VMA_NOT_NULL 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. -*/ -VMA_CALL_PRE void VMA_CALL_POST vmaSetCurrentFrameIndex( - VmaAllocator VMA_NOT_NULL allocator, - uint32_t frameIndex); - -/** \brief Calculated statistics of memory usage in entire allocator. -*/ -typedef struct VmaStatInfo +typedef struct VmaStatistics { - /// Number of `VkDeviceMemory` Vulkan memory blocks allocated. + /** \brief Number of `VkDeviceMemory` objects - Vulkan memory blocks allocated. + */ uint32_t blockCount; - /// Number of #VmaAllocation allocation objects allocated. + /** \brief Number of #VmaAllocation objects allocated. + + Dedicated allocations have their own blocks, so each one adds 1 to `allocationCount` as well as `blockCount`. + */ uint32_t allocationCount; + /** \brief Number of bytes allocated in `VkDeviceMemory` blocks. + + \note To avoid confusion, please be aware that what Vulkan calls an "allocation" - a whole `VkDeviceMemory` object + (e.g. as in `VkPhysicalDeviceLimits::maxMemoryAllocationCount`) is called a "block" in VMA, while VMA calls + "allocation" a #VmaAllocation object that represents a memory region sub-allocated from such block, usually for a single buffer or image. + */ + VkDeviceSize blockBytes; + /** \brief Total number of bytes occupied by all #VmaAllocation objects. + + Always less or equal than `blockBytes`. + Difference `(blockBytes - allocationBytes)` is the amount of memory allocated from Vulkan + but unused by any #VmaAllocation. + */ + VkDeviceSize allocationBytes; +} VmaStatistics; + +/** \brief More detailed statistics than #VmaStatistics. + +These are slower to calculate. Use for debugging purposes. +See functions: vmaCalculateStatistics(), vmaCalculatePoolStatistics(). + +Previous version of the statistics API provided averages, but they have been removed +because they can be easily calculated as: + +\code +VkDeviceSize allocationSizeAvg = detailedStats.statistics.allocationBytes / detailedStats.statistics.allocationCount; +VkDeviceSize unusedBytes = detailedStats.statistics.blockBytes - detailedStats.statistics.allocationBytes; +VkDeviceSize unusedRangeSizeAvg = unusedBytes / detailedStats.unusedRangeCount; +\endcode +*/ +typedef struct VmaDetailedStatistics +{ + /// Basic statistics. + VmaStatistics statistics; /// 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; + /// Smallest allocation size. `VK_WHOLE_SIZE` if there are 0 allocations. + VkDeviceSize allocationSizeMin; + /// Largest allocation size. 0 if there are 0 allocations. + VkDeviceSize allocationSizeMax; + /// Smallest empty range size. `VK_WHOLE_SIZE` if there are 0 empty ranges. + VkDeviceSize unusedRangeSizeMin; + /// Largest empty range size. 0 if there are 0 empty ranges. + VkDeviceSize unusedRangeSizeMax; +} VmaDetailedStatistics; -/// 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; +/** \brief General statistics from current state of the Allocator - +total memory usage across all memory heaps and types. -/** \brief Retrieves statistics from current state of the Allocator. - -This function is called "calculate" not "get" because it has to traverse all -internal data structures, so it may be quite slow. For faster but more brief statistics -suitable to be called every frame or every allocation, use vmaGetBudget(). - -Note that when using allocator from multiple threads, returned information may immediately -become outdated. +These are slower to calculate. Use for debugging purposes. +See function vmaCalculateStatistics(). */ -VMA_CALL_PRE void VMA_CALL_POST vmaCalculateStats( - VmaAllocator VMA_NOT_NULL allocator, - VmaStats* VMA_NOT_NULL pStats); +typedef struct VmaTotalStatistics +{ + VmaDetailedStatistics memoryType[VK_MAX_MEMORY_TYPES]; + VmaDetailedStatistics memoryHeap[VK_MAX_MEMORY_HEAPS]; + VmaDetailedStatistics total; +} VmaTotalStatistics; -/** \brief Statistics of current memory usage and available budget, in bytes, for specific memory heap. +/** \brief Statistics of current memory usage and available budget for a specific memory heap. + +These are fast to calculate. +See function vmaGetHeapBudgets(). */ typedef struct VmaBudget { - /** \brief Sum size of all `VkDeviceMemory` blocks allocated from particular heap, in bytes. + /** \brief Statistics fetched from the library. */ - VkDeviceSize blockBytes; - - /** \brief Sum size of all allocations created in particular heap, in bytes. - - Usually less or equal than `blockBytes`. - Difference `blockBytes - allocationBytes` is the amount of memory allocated but unused - - available for new allocations or wasted due to fragmentation. - - It might be greater than `blockBytes` if there are some allocations in lost state, as they account - to this value as well. - */ - VkDeviceSize allocationBytes; - + VmaStatistics statistics; /** \brief Estimated current memory usage of the program, in bytes. - Fetched from system using `VK_EXT_memory_budget` extension if enabled. + Fetched from system using VK_EXT_memory_budget extension if enabled. - It might be different than `blockBytes` (usually higher) due to additional implicit objects + It might be different than `statistics.blockBytes` (usually higher) due to additional implicit objects also occupying the memory, like swapchain, pipelines, descriptor heaps, command buffers, or `VkDeviceMemory` blocks allocated outside of this library, if any. */ VkDeviceSize usage; - /** \brief Estimated amount of memory available to the program, in bytes. - Fetched from system using `VK_EXT_memory_budget` extension if enabled. + Fetched from system using VK_EXT_memory_budget extension if enabled. It might be different (most probably smaller) than `VkMemoryHeap::size[heapIndex]` due to factors - external to the program, like other programs also consuming system resources. + external to the program, decided by the operating system. Difference `budget - usage` is the amount of additional memory that can probably be allocated without problems. Exceeding the budget may result in various problems. */ VkDeviceSize budget; } VmaBudget; -/** \brief Retrieves information about current memory budget for all memory heaps. +/** @} */ -\param allocator -\param[out] pBudget Must point to array with number of elements at least equal to number of memory heaps in physical device used. - -This function is called "get" not "calculate" because it is very fast, suitable to be called -every frame or every allocation. For more detailed statistics use vmaCalculateStats(). - -Note that when using allocator from multiple threads, returned information may immediately -become outdated. -*/ -VMA_CALL_PRE void VMA_CALL_POST vmaGetBudget( - VmaAllocator VMA_NOT_NULL allocator, - VmaBudget* VMA_NOT_NULL pBudget); - -#ifndef VMA_STATS_STRING_ENABLED -#define VMA_STATS_STRING_ENABLED 1 -#endif - -#if VMA_STATS_STRING_ENABLED - -/// Builds and returns statistics as string in JSON format. /** -@param allocator -@param[out] ppStatsString Must be freed using vmaFreeStatsString() function. -@param detailedMap +\addtogroup group_alloc +@{ */ -VMA_CALL_PRE void VMA_CALL_POST vmaBuildStatsString( - VmaAllocator VMA_NOT_NULL allocator, - char* VMA_NULLABLE * VMA_NOT_NULL ppStatsString, - VkBool32 detailedMap); -VMA_CALL_PRE void VMA_CALL_POST vmaFreeStatsString( - VmaAllocator VMA_NOT_NULL allocator, - char* VMA_NULLABLE pStatsString); +/** \brief Parameters of new #VmaAllocation. -#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). +To be used with functions like vmaCreateBuffer(), vmaCreateImage(), and many others. */ -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 (with LINEAR layout), 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, - /** CPU memory - memory that is preferably not `DEVICE_LOCAL`, but also not guaranteed to be `HOST_VISIBLE`. - - Usage: Staging copy of resources moved from GPU memory to CPU memory as part - of custom paging/residency mechanism, to be moved back to GPU memory when needed. - */ - VMA_MEMORY_USAGE_CPU_COPY = 5, - /** Lazily allocated GPU memory having `VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT`. - Exists mostly on mobile platforms. Using it on desktop PC or other GPUs with no such memory type present will fail the allocation. - - Usage: Memory for transient attachment images (color attachments, depth attachments etc.), created with `VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT`. - - Allocations with this usage are always created as dedicated - it implies #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. - */ - VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED = 6, - - 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. - - 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. - - It is 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, - /** Create both buffer/image and allocation, but don't bind them together. - It is useful when you want to bind yourself to do some more advanced binding, e.g. using some extensions. - The flag is meaningful only with functions that bind by default: vmaCreateBuffer(), vmaCreateImage(). - Otherwise it is ignored. - */ - VMA_ALLOCATION_CREATE_DONT_BIND_BIT = 0x00000080, - /** Create allocation only if additional device memory required for it, if any, won't exceed - memory budget. Otherwise return `VK_ERROR_OUT_OF_DEVICE_MEMORY`. - */ - VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT = 0x00000100, - - /** 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. @@ -1033,6 +1268,394 @@ typedef struct VmaAllocationCreateInfo float priority; } VmaAllocationCreateInfo; +/// 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. + In this case, the pool will also support dedicated allocations. + */ + 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 A floating-point value between 0 and 1, indicating the priority of the allocations in this pool relative to other memory allocations. + + It is used only when #VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT flag was used during creation of the #VmaAllocator object. + Otherwise, this variable is ignored. + */ + float priority; + /** \brief Additional minimum alignment to be used for all allocations created from this pool. Can be 0. + + Leave 0 (default) not to impose any additional alignment. If not 0, it must be a power of two. + It can be useful in cases where alignment returned by Vulkan by functions like `vkGetBufferMemoryRequirements` is not enough, + e.g. when doing interop with OpenGL. + */ + VkDeviceSize minAllocationAlignment; + /** \brief Additional `pNext` chain to be attached to `VkMemoryAllocateInfo` used for every allocation made by this pool. Optional. + + Optional, can be null. If not null, it must point to a `pNext` chain of structures that can be attached to `VkMemoryAllocateInfo`. + It can be useful for special needs such as adding `VkExportMemoryAllocateInfoKHR`. + Structures pointed by this member must remain alive and unchanged for the whole lifetime of the custom pool. + + Please note that some structures, e.g. `VkMemoryPriorityAllocateInfoEXT`, `VkMemoryDedicatedAllocateInfoKHR`, + can be attached automatically by this library when using other, more convenient of its features. + */ + void* VMA_NULLABLE pMemoryAllocateNext; +} VmaPoolCreateInfo; + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + +/// 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 the allocation is moved during \ref defragmentation. + */ + VkDeviceMemory VMA_NULLABLE_NON_DISPATCHABLE deviceMemory; + /** \brief Offset in `VkDeviceMemory` object to the beginning of this allocation, in bytes. `(deviceMemory, offset)` pair is unique to this allocation. + + You usually don't need to use this offset. If you create a buffer or an image together with the allocation using e.g. function + vmaCreateBuffer(), vmaCreateImage(), functions that operate on these resources refer to the beginning of the buffer or image, + not entire device memory block. Functions like vmaMapMemory(), vmaBindBufferMemory() also refer to the beginning of the allocation + and apply this offset automatically. + + It can change after the allocation is moved during \ref defragmentation. + */ + VkDeviceSize offset; + /** \brief Size of this allocation, in bytes. + + It never changes. + + \note Allocation size returned in this variable may be greater than the size + requested for the resource e.g. as `VkBufferCreateInfo::size`. Whole size of the + allocation is accessible for operations on memory e.g. using a pointer after + mapping with vmaMapMemory(), but operations on the resource e.g. using + `vkCmdCopyBuffer` must be limited to the size of the resource. + */ + 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 is null. + + It can change after call to vmaMapMemory(), vmaUnmapMemory(). + It can also change after the allocation is moved during \ref defragmentation. + */ + void* VMA_NULLABLE 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* VMA_NULLABLE pUserData; + /** \brief Custom allocation name that was set with vmaSetAllocationName(). + + It can change after call to vmaSetAllocationName() for this allocation. + + Another way to set custom name is to pass it in VmaAllocationCreateInfo::pUserData with + additional flag #VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT set [DEPRECATED]. + */ + const char* VMA_NULLABLE pName; +} VmaAllocationInfo; + +/** \brief Parameters for defragmentation. + +To be used with function vmaBeginDefragmentation(). +*/ +typedef struct VmaDefragmentationInfo +{ + /// \brief Use combination of #VmaDefragmentationFlagBits. + VmaDefragmentationFlags flags; + /** \brief Custom pool to be defragmented. + + If null then default pools will undergo defragmentation process. + */ + VmaPool VMA_NULLABLE pool; + /** \brief Maximum numbers of bytes that can be copied during single pass, while moving allocations to different places. + + `0` means no limit. + */ + VkDeviceSize maxBytesPerPass; + /** \brief Maximum number of allocations that can be moved during single pass to a different place. + + `0` means no limit. + */ + uint32_t maxAllocationsPerPass; +} VmaDefragmentationInfo; + +/// Single move of an allocation to be done for defragmentation. +typedef struct VmaDefragmentationMove +{ + /// Operation to be performed on the allocation by vmaEndDefragmentationPass(). Default value is #VMA_DEFRAGMENTATION_MOVE_OPERATION_COPY. You can modify it. + VmaDefragmentationMoveOperation operation; + /// Allocation that should be moved. + VmaAllocation VMA_NOT_NULL srcAllocation; + /** \brief Temporary allocation pointing to destination memory that will replace `srcAllocation`. + + \warning Do not store this allocation in your data structures! It exists only temporarily, for the duration of the defragmentation pass, + to be used for binding new buffer/image to the destination memory using e.g. vmaBindBufferMemory(). + vmaEndDefragmentationPass() will destroy it and make `srcAllocation` point to this memory. + */ + VmaAllocation VMA_NOT_NULL dstTmpAllocation; +} VmaDefragmentationMove; + +/** \brief Parameters for incremental defragmentation steps. + +To be used with function vmaBeginDefragmentationPass(). +*/ +typedef struct VmaDefragmentationPassMoveInfo +{ + /// Number of elements in the `pMoves` array. + uint32_t moveCount; + /** \brief Array of moves to be performed by the user in the current defragmentation pass. + + Pointer to an array of `moveCount` elements, owned by VMA, created in vmaBeginDefragmentationPass(), destroyed in vmaEndDefragmentationPass(). + + For each element, you should: + + 1. Create a new buffer/image in the place pointed by VmaDefragmentationMove::dstMemory + VmaDefragmentationMove::dstOffset. + 2. Copy data from the VmaDefragmentationMove::srcAllocation e.g. using `vkCmdCopyBuffer`, `vkCmdCopyImage`. + 3. Make sure these commands finished executing on the GPU. + 4. Destroy the old buffer/image. + + Only then you can finish defragmentation pass by calling vmaEndDefragmentationPass(). + After this call, the allocation will point to the new place in memory. + + Alternatively, if you cannot move specific allocation, you can set VmaDefragmentationMove::operation to #VMA_DEFRAGMENTATION_MOVE_OPERATION_IGNORE. + + Alternatively, if you decide you want to completely remove the allocation: + + 1. Destroy its buffer/image. + 2. Set VmaDefragmentationMove::operation to #VMA_DEFRAGMENTATION_MOVE_OPERATION_DESTROY. + + Then, after vmaEndDefragmentationPass() the allocation will be freed. + */ + VmaDefragmentationMove* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(moveCount) pMoves; +} VmaDefragmentationPassMoveInfo; + +/// Statistics returned for defragmentation process in function vmaEndDefragmentation(). +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; + +/** @} */ + +/** +\addtogroup group_virtual +@{ +*/ + +/// Parameters of created #VmaVirtualBlock object to be passed to vmaCreateVirtualBlock(). +typedef struct VmaVirtualBlockCreateInfo +{ + /** \brief Total size of the virtual block. + + Sizes can be expressed in bytes or any units you want as long as you are consistent in using them. + For example, if you allocate from some array of structures, 1 can mean single instance of entire structure. + */ + VkDeviceSize size; + + /** \brief Use combination of #VmaVirtualBlockCreateFlagBits. + */ + VmaVirtualBlockCreateFlags flags; + + /** \brief Custom CPU memory allocation callbacks. Optional. + + Optional, can be null. When specified, they will be used for all CPU-side memory allocations. + */ + const VkAllocationCallbacks* VMA_NULLABLE pAllocationCallbacks; +} VmaVirtualBlockCreateInfo; + +/// Parameters of created virtual allocation to be passed to vmaVirtualAllocate(). +typedef struct VmaVirtualAllocationCreateInfo +{ + /** \brief Size of the allocation. + + Cannot be zero. + */ + VkDeviceSize size; + /** \brief Required alignment of the allocation. Optional. + + Must be power of two. Special value 0 has the same meaning as 1 - means no special alignment is required, so allocation can start at any offset. + */ + VkDeviceSize alignment; + /** \brief Use combination of #VmaVirtualAllocationCreateFlagBits. + */ + VmaVirtualAllocationCreateFlags flags; + /** \brief Custom pointer to be associated with the allocation. Optional. + + It can be any value and can be used for user-defined purposes. It can be fetched or changed later. + */ + void* VMA_NULLABLE pUserData; +} VmaVirtualAllocationCreateInfo; + +/// Parameters of an existing virtual allocation, returned by vmaGetVirtualAllocationInfo(). +typedef struct VmaVirtualAllocationInfo +{ + /** \brief Offset of the allocation. + + Offset at which the allocation was made. + */ + VkDeviceSize offset; + /** \brief Size of the allocation. + + Same value as passed in VmaVirtualAllocationCreateInfo::size. + */ + VkDeviceSize size; + /** \brief Custom pointer associated with the allocation. + + Same value as passed in VmaVirtualAllocationCreateInfo::pUserData or to vmaSetVirtualAllocationUserData(). + */ + void* VMA_NULLABLE pUserData; +} VmaVirtualAllocationInfo; + +/** @} */ + +#endif // _VMA_DATA_TYPES_DECLARATIONS + +#ifndef _VMA_FUNCTION_HEADERS + +/** +\addtogroup group_init +@{ +*/ + +/// Creates #VmaAllocator object. +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAllocator( + const VmaAllocatorCreateInfo* VMA_NOT_NULL pCreateInfo, + VmaAllocator VMA_NULLABLE* VMA_NOT_NULL pAllocator); + +/// Destroys allocator object. +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyAllocator( + VmaAllocator VMA_NULLABLE allocator); + +/** \brief Returns information about existing #VmaAllocator object - handle to Vulkan device etc. + +It might be useful if you want to keep just the #VmaAllocator handle and fetch other required handles to +`VkPhysicalDevice`, `VkDevice` etc. every time using this function. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocatorInfo( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocatorInfo* VMA_NOT_NULL pAllocatorInfo); + +/** +PhysicalDeviceProperties are fetched from physicalDevice by the allocator. +You can access it here, without fetching it again on your own. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetPhysicalDeviceProperties( + VmaAllocator VMA_NOT_NULL allocator, + const VkPhysicalDeviceProperties* VMA_NULLABLE* VMA_NOT_NULL ppPhysicalDeviceProperties); + +/** +PhysicalDeviceMemoryProperties are fetched from physicalDevice by the allocator. +You can access it here, without fetching it again on your own. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryProperties( + VmaAllocator VMA_NOT_NULL allocator, + const VkPhysicalDeviceMemoryProperties* VMA_NULLABLE* VMA_NOT_NULL 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(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryTypeProperties( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t memoryTypeIndex, + VkMemoryPropertyFlags* VMA_NOT_NULL pFlags); + +/** \brief Sets index of the current frame. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaSetCurrentFrameIndex( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t frameIndex); + +/** @} */ + +/** +\addtogroup group_stats +@{ +*/ + +/** \brief Retrieves statistics from current state of the Allocator. + +This function is called "calculate" not "get" because it has to traverse all +internal data structures, so it may be quite slow. Use it for debugging purposes. +For faster but more brief statistics suitable to be called every frame or every allocation, +use vmaGetHeapBudgets(). + +Note that when using allocator from multiple threads, returned information may immediately +become outdated. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaCalculateStatistics( + VmaAllocator VMA_NOT_NULL allocator, + VmaTotalStatistics* VMA_NOT_NULL pStats); + +/** \brief Retrieves information about current memory usage and budget for all memory heaps. + +\param allocator +\param[out] pBudgets Must point to array with number of elements at least equal to number of memory heaps in physical device used. + +This function is called "get" not "calculate" because it is very fast, suitable to be called +every frame or every allocation. For more detailed statistics use vmaCalculateStatistics(). + +Note that when using allocator from multiple threads, returned information may immediately +become outdated. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetHeapBudgets( + VmaAllocator VMA_NOT_NULL allocator, + VmaBudget* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL("VkPhysicalDeviceMemoryProperties::memoryHeapCount") pBudgets); + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + /** \brief Helps to find memoryTypeIndex, given memoryTypeBits and VmaAllocationCreateInfo. @@ -1060,12 +1683,6 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndex( 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` */ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( VmaAllocator VMA_NOT_NULL allocator, @@ -1078,12 +1695,6 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( 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` */ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForImageInfo( VmaAllocator VMA_NOT_NULL allocator, @@ -1091,172 +1702,16 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForImageInfo( const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, uint32_t* VMA_NOT_NULL 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; - /** \brief A floating-point value between 0 and 1, indicating the priority of the allocations in this pool relative to other memory allocations. - - It is used only when #VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT flag was used during creation of the #VmaAllocator object. - Otherwise, this variable is ignored. - */ - float priority; - /** \brief Additional minimum alignment to be used for all allocations created from this pool. Can be 0. - - Leave 0 (default) not to impose any additional alignment. If not 0, it must be a power of two. - It can be useful in cases where alignment returned by Vulkan by functions like `vkGetBufferMemoryRequirements` is not enough, - e.g. when doing interop with OpenGL. - */ - VkDeviceSize minAllocationAlignment; - /** \brief Additional `pNext` chain to be attached to `VkMemoryAllocateInfo` used for every allocation made by this pool. Optional. - - Optional, can be null. If not null, it must point to a `pNext` chain of structures that can be attached to `VkMemoryAllocateInfo`. - It can be useful for special needs such as adding `VkExportMemoryAllocateInfoKHR`. - Structures pointed by this member must remain alive and unchanged for the whole lifetime of the custom pool. - - Please note that some structures, e.g. `VkMemoryPriorityAllocateInfoEXT`, `VkMemoryDedicatedAllocateInfoKHR`, - can be attached automatically by this library when using other, more convenient of its features. - */ - void* VMA_NULLABLE pMemoryAllocateNext; -} 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. +\param allocator Allocator object. +\param pCreateInfo Parameters of pool to create. +\param[out] pPool Handle to created pool. */ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreatePool( VmaAllocator VMA_NOT_NULL allocator, const VmaPoolCreateInfo* VMA_NOT_NULL pCreateInfo, - VmaPool VMA_NULLABLE * VMA_NOT_NULL pPool); + VmaPool VMA_NULLABLE* VMA_NOT_NULL pPool); /** \brief Destroys #VmaPool object and frees Vulkan device memory. */ @@ -1264,27 +1719,41 @@ VMA_CALL_PRE void VMA_CALL_POST vmaDestroyPool( VmaAllocator VMA_NOT_NULL allocator, VmaPool VMA_NULLABLE pool); +/** @} */ + +/** +\addtogroup group_stats +@{ +*/ + /** \brief Retrieves statistics of existing #VmaPool object. -@param allocator Allocator object. -@param pool Pool object. -@param[out] pPoolStats Statistics of specified pool. +\param allocator Allocator object. +\param pool Pool object. +\param[out] pPoolStats Statistics of specified pool. */ -VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStats( +VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStatistics( VmaAllocator VMA_NOT_NULL allocator, VmaPool VMA_NOT_NULL pool, - VmaPoolStats* VMA_NOT_NULL pPoolStats); + VmaStatistics* VMA_NOT_NULL pPoolStats); -/** \brief Marks all allocations in given pool as lost if they are not used in current frame or VmaPoolCreateInfo::frameInUseCount back from now. +/** \brief Retrieves detailed statistics of existing #VmaPool object. -@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. +\param allocator Allocator object. +\param pool Pool object. +\param[out] pPoolStats Statistics of specified pool. */ -VMA_CALL_PRE void VMA_CALL_POST vmaMakePoolAllocationsLost( +VMA_CALL_PRE void VMA_CALL_POST vmaCalculatePoolStatistics( VmaAllocator VMA_NOT_NULL allocator, VmaPool VMA_NOT_NULL pool, - size_t* VMA_NULLABLE pLostAllocationCount); + VmaDetailedStatistics* VMA_NOT_NULL pPoolStats); + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ /** \brief Checks magic number in margins around all allocations in given memory pool in search for corruptions. @@ -1300,7 +1769,9 @@ Possible return values: `VMA_ASSERT` is also fired in that case. - Other value: Error returned by Vulkan, e.g. memory mapping failure. */ -VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckPoolCorruption(VmaAllocator VMA_NOT_NULL allocator, VmaPool VMA_NOT_NULL pool); +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckPoolCorruption( + VmaAllocator VMA_NOT_NULL allocator, + VmaPool VMA_NOT_NULL pool); /** \brief Retrieves name of a custom pool. @@ -1311,7 +1782,7 @@ destroyed or its name is changed using vmaSetPoolName(). VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolName( VmaAllocator VMA_NOT_NULL allocator, VmaPool VMA_NOT_NULL pool, - const char* VMA_NULLABLE * VMA_NOT_NULL ppName); + const char* VMA_NULLABLE* VMA_NOT_NULL ppName); /** \brief Sets name of a custom pool. @@ -1323,93 +1794,13 @@ VMA_CALL_PRE void VMA_CALL_POST vmaSetPoolName( VmaPool VMA_NOT_NULL pool, const char* VMA_NULLABLE pName); -/** \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 VMA_NULLABLE_NON_DISPATCHABLE deviceMemory; - /** \brief Offset in `VkDeviceMemory` object to the beginning of this allocation, in bytes. `(deviceMemory, offset)` pair is unique to this allocation. - - You usually don't need to use this offset. If you create a buffer or an image together with the allocation using e.g. function - vmaCreateBuffer(), vmaCreateImage(), functions that operate on these resources refer to the beginning of the buffer or image, - not entire device memory block. Functions like vmaMapMemory(), vmaBindBufferMemory() also refer to the beginning of the allocation - and apply this offset automatically. - - 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. - - \note Allocation size returned in this variable may be greater than the size - requested for the resource e.g. as `VkBufferCreateInfo::size`. Whole size of the - allocation is accessible for operations on memory e.g. using a pointer after - mapping with vmaMapMemory(), but operations on the resource e.g. using - `vkCmdCopyBuffer` must be limited to the size of the resource. - */ - 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 is 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* VMA_NULLABLE 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* VMA_NULLABLE pUserData; -} VmaAllocationInfo; - /** \brief General purpose memory allocation. -@param allocator -@param pVkMemoryRequirements -@param pCreateInfo -@param[out] pAllocation Handle to allocated memory. -@param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). +\param allocator +\param pVkMemoryRequirements +\param pCreateInfo +\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(). @@ -1420,17 +1811,17 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemory( VmaAllocator VMA_NOT_NULL allocator, const VkMemoryRequirements* VMA_NOT_NULL pVkMemoryRequirements, const VmaAllocationCreateInfo* VMA_NOT_NULL pCreateInfo, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE 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. +\param allocator Allocator object. +\param pVkMemoryRequirements Memory requirements for each allocation. +\param pCreateInfo Creation parameters for each allocation. +\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(). @@ -1447,31 +1838,49 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryPages( const VkMemoryRequirements* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pVkMemoryRequirements, const VmaAllocationCreateInfo* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pCreateInfo, size_t allocationCount, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pAllocations, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pAllocations, VmaAllocationInfo* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) pAllocationInfo); -/** -@param allocator -@param buffer -@param pCreateInfo -@param[out] pAllocation Handle to allocated memory. -@param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). +/** \brief Allocates memory suitable for given `VkBuffer`. -You should free the memory using vmaFreeMemory(). +\param allocator +\param buffer +\param pCreateInfo +\param[out] pAllocation Handle to allocated memory. +\param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). + +It only creates #VmaAllocation. To bind the memory to the buffer, use vmaBindBufferMemory(). + +This is a special-purpose function. In most cases you should use vmaCreateBuffer(). + +You must free the allocation using vmaFreeMemory() when no longer needed. */ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForBuffer( VmaAllocator VMA_NOT_NULL allocator, VkBuffer VMA_NOT_NULL_NON_DISPATCHABLE buffer, const VmaAllocationCreateInfo* VMA_NOT_NULL pCreateInfo, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); -/// Function similar to vmaAllocateMemoryForBuffer(). +/** \brief Allocates memory suitable for given `VkImage`. + +\param allocator +\param image +\param pCreateInfo +\param[out] pAllocation Handle to allocated memory. +\param[out] pAllocationInfo Optional. Information about allocated memory. It can be later fetched using function vmaGetAllocationInfo(). + +It only creates #VmaAllocation. To bind the memory to the buffer, use vmaBindImageMemory(). + +This is a special-purpose function. In most cases you should use vmaCreateImage(). + +You must free the allocation using vmaFreeMemory() when no longer needed. +*/ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForImage( VmaAllocator VMA_NOT_NULL allocator, VkImage VMA_NOT_NULL_NON_DISPATCHABLE image, const VmaAllocationCreateInfo* VMA_NOT_NULL pCreateInfo, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); /** \brief Frees memory previously allocated using vmaAllocateMemory(), vmaAllocateMemoryForBuffer(), or vmaAllocateMemoryForImage(). @@ -1495,58 +1904,27 @@ Passing `VK_NULL_HANDLE` as elements of `pAllocations` array is valid. Such entr VMA_CALL_PRE void VMA_CALL_POST vmaFreeMemoryPages( VmaAllocator VMA_NOT_NULL allocator, size_t allocationCount, - const VmaAllocation VMA_NULLABLE * VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pAllocations); + const VmaAllocation VMA_NULLABLE* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pAllocations); -/** \brief Returns current information about specified allocation and atomically marks it as used in current frame. +/** \brief Returns current information about specified allocation. Current paramteres 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. +Although this function doesn't lock any mutex, so it should be quite efficient, +you should 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). */ VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationInfo( VmaAllocator VMA_NOT_NULL allocator, VmaAllocation VMA_NOT_NULL allocation, VmaAllocationInfo* VMA_NOT_NULL 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 is 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`. -*/ -VMA_CALL_PRE VkBool32 VMA_CALL_POST vmaTouchAllocation( - VmaAllocator VMA_NOT_NULL allocator, - VmaAllocation VMA_NOT_NULL 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. +The value of pointer `pUserData` is 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. */ VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationUserData( @@ -1554,27 +1932,40 @@ VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationUserData( VmaAllocation VMA_NOT_NULL allocation, void* VMA_NULLABLE pUserData); -/** \brief Creates new allocation that is in lost state from the beginning. +/** \brief Sets pName in given allocation to new value. -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. +`pName` 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 `pName`. String +passed as pName 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 +`pName` is freed from memory. */ -VMA_CALL_PRE void VMA_CALL_POST vmaCreateLostAllocation( +VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationName( VmaAllocator VMA_NOT_NULL allocator, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation); + VmaAllocation VMA_NOT_NULL allocation, + const char* VMA_NULLABLE pName); + +/** +\brief Given an allocation, returns Property Flags of its memory type. + +This is just a convenience function. Same information can be obtained using +vmaGetAllocationInfo() + vmaGetMemoryProperties(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationMemoryProperties( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkMemoryPropertyFlags* VMA_NOT_NULL pFlags); /** \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 offsetted to the beginning of region assigned to this particular -allocation. + +\warning +If the allocation is part of a bigger `VkDeviceMemory` block, returned pointer is +correctly offsetted to the beginning of region assigned to this particular allocation. +Unlike the result of `vkMapMemory`, it points to the allocation, not to the beginning of the whole block. +You should not add VmaAllocationInfo::offset to it! Mapping is internally reference-counted and synchronized, so despite raw Vulkan function `vkMapMemory()` cannot be used to map same block of `VkDeviceMemory` @@ -1598,10 +1989,6 @@ vmaMapMemory(). You must not call vmaUnmapMemory() additional time to free the 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. - This function doesn't automatically flush or invalidate caches. If the allocation is made from a memory types that is not `HOST_COHERENT`, you also need to use vmaInvalidateAllocation() / vmaFlushAllocation(), as required by Vulkan specification. @@ -1609,7 +1996,7 @@ you also need to use vmaInvalidateAllocation() / vmaFlushAllocation(), as requir VMA_CALL_PRE VkResult VMA_CALL_POST vmaMapMemory( VmaAllocator VMA_NOT_NULL allocator, VmaAllocation VMA_NOT_NULL allocation, - void* VMA_NULLABLE * VMA_NOT_NULL ppData); + void* VMA_NULLABLE* VMA_NOT_NULL ppData); /** \brief Unmaps memory represented by given allocation, mapped previously using vmaMapMemory(). @@ -1694,7 +2081,7 @@ called, otherwise `VK_SUCCESS`. VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocations( VmaAllocator VMA_NOT_NULL allocator, uint32_t allocationCount, - const VmaAllocation VMA_NOT_NULL * VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) allocations, + const VmaAllocation VMA_NOT_NULL* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) allocations, const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) offsets, const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) sizes); @@ -1715,14 +2102,14 @@ called, otherwise `VK_SUCCESS`. VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocations( VmaAllocator VMA_NOT_NULL allocator, uint32_t allocationCount, - const VmaAllocation VMA_NOT_NULL * VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) allocations, + const VmaAllocation VMA_NOT_NULL* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) allocations, const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) offsets, const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) sizes); /** \brief Checks magic number in margins around all allocations in given memory types (in both default and custom pools) in search for corruptions. -@param allocator -@param memoryTypeBits Bit mask, where each bit set means that a memory type with that index should be checked. +\param allocator +\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 @@ -1736,245 +2123,77 @@ Possible return values: `VMA_ASSERT` is also fired in that case. - Other value: Error returned by Vulkan, e.g. memory mapping failure. */ -VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption(VmaAllocator VMA_NOT_NULL 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_INCREMENTAL = 0x1, - 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. - */ - const VmaAllocation VMA_NOT_NULL * VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) 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* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) 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. - */ - const VmaPool VMA_NOT_NULL * VMA_NULLABLE VMA_LEN_IF_NOT_NULL(poolCount) 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 VMA_NULLABLE commandBuffer; -} VmaDefragmentationInfo2; - -typedef struct VmaDefragmentationPassMoveInfo { - VmaAllocation VMA_NOT_NULL allocation; - VkDeviceMemory VMA_NOT_NULL_NON_DISPATCHABLE memory; - VkDeviceSize offset; -} VmaDefragmentationPassMoveInfo; - -/** \brief Parameters for incremental defragmentation steps. - -To be used with function vmaBeginDefragmentationPass(). -*/ -typedef struct VmaDefragmentationPassInfo { - uint32_t moveCount; - VmaDefragmentationPassMoveInfo* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(moveCount) pMoves; -} VmaDefragmentationPassInfo; - -/** \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; +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t memoryTypeBits); /** \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. +\param allocator Allocator object. +\param pInfo Structure filled with parameters of defragmentation. +\param[out] pContext Context object that must be passed to vmaEndDefragmentation() to finish defragmentation. +\returns +- `VK_SUCCESS` if defragmentation can begin. +- `VK_ERROR_FEATURE_NOT_PRESENT` if defragmentation is not supported. -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(). - -For more information and important limitations regarding defragmentation, see documentation chapter: +For more information about defragmentation, see documentation chapter: [Defragmentation](@ref defragmentation). */ -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragmentationBegin( +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentation( VmaAllocator VMA_NOT_NULL allocator, - const VmaDefragmentationInfo2* VMA_NOT_NULL pInfo, - VmaDefragmentationStats* VMA_NULLABLE pStats, - VmaDefragmentationContext VMA_NULLABLE * VMA_NOT_NULL pContext); + const VmaDefragmentationInfo* VMA_NOT_NULL pInfo, + VmaDefragmentationContext VMA_NULLABLE* VMA_NOT_NULL 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. -*/ -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragmentationEnd( - VmaAllocator VMA_NOT_NULL allocator, - VmaDefragmentationContext VMA_NULLABLE context); +\param allocator Allocator object. +\param context Context object that has been created by vmaBeginDefragmentation(). +\param[out] pStats Optional stats for the defragmentation. Can be null. +Use this function to finish defragmentation started by vmaBeginDefragmentation(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaEndDefragmentation( + VmaAllocator VMA_NOT_NULL allocator, + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationStats* VMA_NULLABLE pStats); + +/** \brief Starts single defragmentation pass. + +\param allocator Allocator object. +\param context Context object that has been created by vmaBeginDefragmentation(). +\param[out] pPassInfo Computed informations for current pass. +\returns +- `VK_SUCCESS` if no more moves are possible. Then you can omit call to vmaEndDefragmentationPass() and simply end whole defragmentation. +- `VK_INCOMPLETE` if there are pending moves returned in `pPassInfo`. You need to perform them, call vmaEndDefragmentationPass(), + and then preferably try another pass with vmaBeginDefragmentationPass(). +*/ VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentationPass( VmaAllocator VMA_NOT_NULL allocator, - VmaDefragmentationContext VMA_NULLABLE context, - VmaDefragmentationPassInfo* VMA_NOT_NULL pInfo -); + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo); + +/** \brief Ends single defragmentation pass. + +\param allocator Allocator object. +\param context Context object that has been created by vmaBeginDefragmentation(). +\param pPassInfo Computed informations for current pass filled by vmaBeginDefragmentationPass() and possibly modified by you. + +Returns `VK_SUCCESS` if no more moves are possible or `VK_INCOMPLETE` if more defragmentations are possible. + +Ends incremental defragmentation pass and commits all defragmentation moves from `pPassInfo`. +After this call: + +- Allocations at `pPassInfo[i].srcAllocation` that had `pPassInfo[i].operation ==` #VMA_DEFRAGMENTATION_MOVE_OPERATION_COPY + (which is the default) will be pointing to the new destination place. +- Allocation at `pPassInfo[i].srcAllocation` that had `pPassInfo[i].operation ==` #VMA_DEFRAGMENTATION_MOVE_OPERATION_DESTROY + will be freed. + +If no more moves are possible you can end whole defragmentation. +*/ VMA_CALL_PRE VkResult VMA_CALL_POST vmaEndDefragmentationPass( VmaAllocator VMA_NOT_NULL allocator, - VmaDefragmentationContext VMA_NULLABLE context -); - -/** \brief Deprecated. Compacts memory by moving allocations. - -@param allocator -@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. -*/ -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragment( - VmaAllocator VMA_NOT_NULL allocator, - const VmaAllocation VMA_NOT_NULL * VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(allocationCount) pAllocations, - size_t allocationCount, - VkBool32* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) pAllocationsChanged, - const VmaDefragmentationInfo* VMA_NULLABLE pDefragmentationInfo, - VmaDefragmentationStats* VMA_NULLABLE pDefragmentationStats); + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo); /** \brief Binds buffer to allocation. @@ -1995,11 +2214,11 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory( /** \brief Binds buffer to allocation with additional parameters. -@param allocator -@param allocation -@param allocationLocalOffset Additional offset to be added while binding, relative to the beginning of the `allocation`. Normally it should be 0. -@param buffer -@param pNext A chain of structures to be attached to `VkBindBufferMemoryInfoKHR` structure used internally. Normally it should be null. +\param allocator +\param allocation +\param allocationLocalOffset Additional offset to be added while binding, relative to the beginning of the `allocation`. Normally it should be 0. +\param buffer +\param pNext A chain of structures to be attached to `VkBindBufferMemoryInfoKHR` structure used internally. Normally it should be null. This function is similar to vmaBindBufferMemory(), but it provides additional parameters. @@ -2032,11 +2251,11 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory( /** \brief Binds image to allocation with additional parameters. -@param allocator -@param allocation -@param allocationLocalOffset Additional offset to be added while binding, relative to the beginning of the `allocation`. Normally it should be 0. -@param image -@param pNext A chain of structures to be attached to `VkBindImageMemoryInfoKHR` structure used internally. Normally it should be null. +\param allocator +\param allocation +\param allocationLocalOffset Additional offset to be added while binding, relative to the beginning of the `allocation`. Normally it should be 0. +\param image +\param pNext A chain of structures to be attached to `VkBindImageMemoryInfoKHR` structure used internally. Normally it should be null. This function is similar to vmaBindImageMemory(), but it provides additional parameters. @@ -2050,13 +2269,14 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory2( VkImage VMA_NOT_NULL_NON_DISPATCHABLE image, const void* VMA_NULLABLE pNext); -/** -@param allocator -@param pBufferCreateInfo -@param pAllocationCreateInfo -@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(). +/** \brief Creates a new `VkBuffer`, allocates and binds memory for it. + +\param allocator +\param pBufferCreateInfo +\param pAllocationCreateInfo +\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: @@ -2065,7 +2285,7 @@ This function automatically: -# 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. +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 @@ -2074,8 +2294,8 @@ 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 +and if dedicated allocation is possible +(#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. @@ -2087,8 +2307,8 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBuffer( VmaAllocator VMA_NOT_NULL allocator, const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, - VkBuffer VMA_NULLABLE_NON_DISPATCHABLE * VMA_NOT_NULL pBuffer, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pBuffer, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); /** \brief Creates a buffer with additional minimum alignment. @@ -2102,10 +2322,35 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBufferWithAlignment( const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, VkDeviceSize minAlignment, - VkBuffer VMA_NULLABLE_NON_DISPATCHABLE * VMA_NOT_NULL pBuffer, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pBuffer, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); +/** \brief Creates a new `VkBuffer`, binds already created memory for it. + +\param allocator +\param allocation Allocation that provides memory to be used for binding new buffer to it. +\param pBufferCreateInfo +\param[out] pBuffer Buffer that was created. + +This function automatically: + +-# Creates buffer. +-# Binds the buffer with the supplied memory. + +If any of these operations fail, buffer is not created, +returned value is negative error code and `*pBuffer` is null. + +If the function succeeded, you must destroy the buffer when you +no longer need it using `vkDestroyBuffer()`. If you want to also destroy the corresponding +allocation you can use convenience function vmaDestroyBuffer(). +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingBuffer( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pBuffer); + /** \brief Destroys Vulkan buffer and frees allocated memory. This is just a convenience function equivalent to: @@ -2127,10 +2372,17 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateImage( VmaAllocator VMA_NOT_NULL allocator, const VkImageCreateInfo* VMA_NOT_NULL pImageCreateInfo, const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, - VkImage VMA_NULLABLE_NON_DISPATCHABLE * VMA_NOT_NULL pImage, - VmaAllocation VMA_NULLABLE * VMA_NOT_NULL pAllocation, + VkImage VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pImage, + VmaAllocation VMA_NULLABLE* VMA_NOT_NULL pAllocation, VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); +/// Function similar to vmaCreateAliasingBuffer(). +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingImage( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const VkImageCreateInfo* VMA_NOT_NULL pImageCreateInfo, + VkImage VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pImage); + /** \brief Destroys Vulkan image and frees allocated memory. This is just a convenience function equivalent to: @@ -2147,12 +2399,164 @@ VMA_CALL_PRE void VMA_CALL_POST vmaDestroyImage( VkImage VMA_NULLABLE_NON_DISPATCHABLE image, VmaAllocation VMA_NULLABLE allocation); +/** @} */ + +/** +\addtogroup group_virtual +@{ +*/ + +/** \brief Creates new #VmaVirtualBlock object. + +\param pCreateInfo Parameters for creation. +\param[out] pVirtualBlock Returned virtual block object or `VMA_NULL` if creation failed. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateVirtualBlock( + const VmaVirtualBlockCreateInfo* VMA_NOT_NULL pCreateInfo, + VmaVirtualBlock VMA_NULLABLE* VMA_NOT_NULL pVirtualBlock); + +/** \brief Destroys #VmaVirtualBlock object. + +Please note that you should consciously handle virtual allocations that could remain unfreed in the block. +You should either free them individually using vmaVirtualFree() or call vmaClearVirtualBlock() +if you are sure this is what you want. If you do neither, an assert is called. + +If you keep pointers to some additional metadata associated with your virtual allocations in their `pUserData`, +don't forget to free them. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyVirtualBlock( + VmaVirtualBlock VMA_NULLABLE virtualBlock); + +/** \brief Returns true of the #VmaVirtualBlock is empty - contains 0 virtual allocations and has all its space available for new allocations. +*/ +VMA_CALL_PRE VkBool32 VMA_CALL_POST vmaIsVirtualBlockEmpty( + VmaVirtualBlock VMA_NOT_NULL virtualBlock); + +/** \brief Returns information about a specific virtual allocation within a virtual block, like its size and `pUserData` pointer. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetVirtualAllocationInfo( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaVirtualAllocation VMA_NOT_NULL_NON_DISPATCHABLE allocation, VmaVirtualAllocationInfo* VMA_NOT_NULL pVirtualAllocInfo); + +/** \brief Allocates new virtual allocation inside given #VmaVirtualBlock. + +If the allocation fails due to not enough free space available, `VK_ERROR_OUT_OF_DEVICE_MEMORY` is returned +(despite the function doesn't ever allocate actual GPU memory). +`pAllocation` is then set to `VK_NULL_HANDLE` and `pOffset`, if not null, it set to `UINT64_MAX`. + +\param virtualBlock Virtual block +\param pCreateInfo Parameters for the allocation +\param[out] pAllocation Returned handle of the new allocation +\param[out] pOffset Returned offset of the new allocation. Optional, can be null. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaVirtualAllocate( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + const VmaVirtualAllocationCreateInfo* VMA_NOT_NULL pCreateInfo, + VmaVirtualAllocation VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pAllocation, + VkDeviceSize* VMA_NULLABLE pOffset); + +/** \brief Frees virtual allocation inside given #VmaVirtualBlock. + +It is correct to call this function with `allocation == VK_NULL_HANDLE` - it does nothing. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaVirtualFree( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaVirtualAllocation VMA_NULLABLE_NON_DISPATCHABLE allocation); + +/** \brief Frees all virtual allocations inside given #VmaVirtualBlock. + +You must either call this function or free each virtual allocation individually with vmaVirtualFree() +before destroying a virtual block. Otherwise, an assert is called. + +If you keep pointer to some additional metadata associated with your virtual allocation in its `pUserData`, +don't forget to free it as well. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaClearVirtualBlock( + VmaVirtualBlock VMA_NOT_NULL virtualBlock); + +/** \brief Changes custom pointer associated with given virtual allocation. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaSetVirtualAllocationUserData( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaVirtualAllocation VMA_NOT_NULL_NON_DISPATCHABLE allocation, + void* VMA_NULLABLE pUserData); + +/** \brief Calculates and returns statistics about virtual allocations and memory usage in given #VmaVirtualBlock. + +This function is fast to call. For more detailed statistics, see vmaCalculateVirtualBlockStatistics(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetVirtualBlockStatistics( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaStatistics* VMA_NOT_NULL pStats); + +/** \brief Calculates and returns detailed statistics about virtual allocations and memory usage in given #VmaVirtualBlock. + +This function is slow to call. Use for debugging purposes. +For less detailed statistics, see vmaGetVirtualBlockStatistics(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaCalculateVirtualBlockStatistics( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaDetailedStatistics* VMA_NOT_NULL pStats); + +/** @} */ + +#if VMA_STATS_STRING_ENABLED +/** +\addtogroup group_stats +@{ +*/ + +/** \brief Builds and returns a null-terminated string in JSON format with information about given #VmaVirtualBlock. +\param virtualBlock Virtual block. +\param[out] ppStatsString Returned string. +\param detailedMap Pass `VK_FALSE` to only obtain statistics as returned by vmaCalculateVirtualBlockStatistics(). Pass `VK_TRUE` to also obtain full list of allocations and free spaces. + +Returned string must be freed using vmaFreeVirtualBlockStatsString(). +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaBuildVirtualBlockStatsString( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + char* VMA_NULLABLE* VMA_NOT_NULL ppStatsString, + VkBool32 detailedMap); + +/// Frees a string returned by vmaBuildVirtualBlockStatsString(). +VMA_CALL_PRE void VMA_CALL_POST vmaFreeVirtualBlockStatsString( + VmaVirtualBlock VMA_NOT_NULL virtualBlock, + char* VMA_NULLABLE pStatsString); + +/** \brief Builds and returns statistics as a null-terminated string in JSON format. +\param allocator +\param[out] ppStatsString Must be freed using vmaFreeStatsString() function. +\param detailedMap +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaBuildStatsString( + VmaAllocator VMA_NOT_NULL allocator, + char* VMA_NULLABLE* VMA_NOT_NULL ppStatsString, + VkBool32 detailedMap); + +VMA_CALL_PRE void VMA_CALL_POST vmaFreeStatsString( + VmaAllocator VMA_NOT_NULL allocator, + char* VMA_NULLABLE pStatsString); + +/** @} */ + +#endif // VMA_STATS_STRING_ENABLED + +#endif // _VMA_FUNCTION_HEADERS + #ifdef __cplusplus } #endif #endif // AMD_VULKAN_MEMORY_ALLOCATOR_H +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// +// +// IMPLEMENTATION +// +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// + // For Visual Studio IntelliSense. #if defined(__cplusplus) && defined(__INTELLISENSE__) #define VMA_IMPLEMENTATION @@ -2165,15 +2569,13 @@ VMA_CALL_PRE void VMA_CALL_POST vmaDestroyImage( #include #include #include +#include -#if VMA_RECORDING_ENABLED - #include - #if defined(_WIN32) - #include - #else - #include - #include - #endif +#ifdef _MSC_VER + #include // For functions like __popcnt, _BitScanForward etc. +#endif +#if __cplusplus >= 202002L || _MSVC_LANG >= 202002L // C++20 + #include // For std::popcount #endif /******************************************************************************* @@ -2182,6 +2584,7 @@ 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. */ +#ifndef _VMA_CONFIGURATION /* Define this macro to 1 to make the library fetch pointers to Vulkan functions @@ -2197,29 +2600,14 @@ internally, like: Define this macro to 1 to make the library fetch pointers to Vulkan functions internally, like: - vulkanFunctions.vkAllocateMemory = (PFN_vkAllocateMemory)vkGetDeviceProcAddr(m_hDevice, vkAllocateMemory); + vulkanFunctions.vkAllocateMemory = (PFN_vkAllocateMemory)vkGetDeviceProcAddr(device, "vkAllocateMemory"); + +To use this feature in new versions of VMA you now have to pass +VmaVulkanFunctions::vkGetInstanceProcAddr and vkGetDeviceProcAddr as +VmaAllocatorCreateInfo::pVulkanFunctions. Other members can be null. */ #if !defined(VMA_DYNAMIC_VULKAN_FUNCTIONS) #define VMA_DYNAMIC_VULKAN_FUNCTIONS 1 - #if defined(VK_NO_PROTOTYPES) - extern PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr; - extern PFN_vkGetDeviceProcAddr vkGetDeviceProcAddr; - #endif -#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 @@ -2228,7 +2616,6 @@ the containers. #define VMA_USE_STL_SHARED_MUTEX 1 // Visual studio defines __cplusplus properly only when passed additional parameter: /Zc:__cplusplus // Otherwise it is always 199711L, despite shared_mutex works since Visual Studio 2015 Update 2. - // See: https://blogs.msdn.microsoft.com/vcblog/2018/04/09/msvc-now-correctly-reports-__cplusplus/ #elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 190023918 && __cplusplus == 199711L && _MSVC_LANG >= 201703L #define VMA_USE_STL_SHARED_MUTEX 1 #else @@ -2237,28 +2624,33 @@ the containers. #endif /* -THESE INCLUDES ARE NOT ENABLED BY DEFAULT. -Library has its own container implementation. -*/ -#if VMA_USE_STL_VECTOR - #include -#endif +Define this macro to include custom header files without having to edit this file directly, e.g.: -#if VMA_USE_STL_UNORDERED_MAP - #include -#endif + // Inside of "my_vma_configuration_user_includes.h": -#if VMA_USE_STL_LIST - #include -#endif + #include "my_custom_assert.h" // for MY_CUSTOM_ASSERT + #include "my_custom_min.h" // for my_custom_min + #include + #include -/* -Following headers are used in this CONFIGURATION section only, so feel free to + // Inside a different file, which includes "vk_mem_alloc.h": + + #define VMA_CONFIGURATION_USER_INCLUDES_H "my_vma_configuration_user_includes.h" + #define VMA_ASSERT(expr) MY_CUSTOM_ASSERT(expr) + #define VMA_MIN(v1, v2) (my_custom_min(v1, v2)) + #include "vk_mem_alloc.h" + ... + +The following headers are used in this CONFIGURATION section only, so feel free to remove them if not needed. */ -#include // for assert -#include // for min, max -#include +#if !defined(VMA_CONFIGURATION_USER_INCLUDES_H) + #include // for assert + #include // for min, max + #include +#else + #include VMA_CONFIGURATION_USER_INCLUDES_H +#endif #ifndef VMA_NULL // Value used as null pointer. Define it to e.g.: nullptr, NULL, 0, (void*)0. @@ -2377,6 +2769,21 @@ static void vma_aligned_free(void* VMA_NULLABLE ptr) #endif #endif +#ifndef VMA_COUNT_BITS_SET + // Returns number of bits set to 1 in (v) + #define VMA_COUNT_BITS_SET(v) VmaCountBitsSet(v) +#endif + +#ifndef VMA_BITSCAN_LSB + // Scans integer for index of first nonzero value from the Least Significant Bit (LSB). If mask is 0 then returns UINT8_MAX + #define VMA_BITSCAN_LSB(mask) VmaBitScanLSB(mask) +#endif + +#ifndef VMA_BITSCAN_MSB + // Scans integer for index of first nonzero value from the Most Significant Bit (MSB). If mask is 0 then returns UINT8_MAX + #define VMA_BITSCAN_MSB(mask) VmaBitScanMSB(mask) +#endif + #ifndef VMA_MIN #define VMA_MIN(v1, v2) ((std::min)((v1), (v2))) #endif @@ -2520,7 +2927,7 @@ If providing your own implementation, you need to implement a subset of std::ato #ifndef VMA_DEBUG_MARGIN /** - Minimum margin before and after every allocation, in bytes. + Minimum margin after every allocation, in bytes. Set nonzero for debugging purposes only. */ #define VMA_DEBUG_MARGIN (0) @@ -2537,7 +2944,7 @@ If providing your own implementation, you need to implement a subset of std::ato #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 + enable writing magic value to the margin after every allocation and validating it, so that memory corruptions (out-of-bounds writes) are detected. */ #define VMA_DEBUG_DETECT_CORRUPTION (0) @@ -2577,6 +2984,17 @@ If providing your own implementation, you need to implement a subset of std::ato #define VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE (256ull * 1024 * 1024) #endif +/* +Mapping hysteresis is a logic that launches when vmaMapMemory/vmaUnmapMemory is called +or a persistently mapped allocation is created and destroyed several times in a row. +It keeps additional +1 mapping of a device memory block to prevent calling actual +vkMapMemory/vkUnmapMemory too many times, which may improve performance and help +tools like RenderDOc. +*/ +#ifndef VMA_MAPPING_HYSTERESIS_ENABLED + #define VMA_MAPPING_HYSTERESIS_ENABLED 1 +#endif + #ifndef VMA_CLASS_NO_COPY #define VMA_CLASS_NO_COPY(className) \ private: \ @@ -2584,38 +3002,281 @@ If providing your own implementation, you need to implement a subset of std::ato 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; +#define VMA_VALIDATE(cond) do { if(!(cond)) { \ + VMA_ASSERT(0 && "Validation failed: " #cond); \ + return false; \ + } } while(false) /******************************************************************************* END OF CONFIGURATION */ +#endif // _VMA_CONFIGURATION -// # Copy of some Vulkan definitions so we don't need to check their existence just to handle few constants. +static const uint8_t VMA_ALLOCATION_FILL_PATTERN_CREATED = 0xDC; +static const uint8_t VMA_ALLOCATION_FILL_PATTERN_DESTROYED = 0xEF; +// Decimal 2139416166, float NaN, little-endian binary 66 E6 84 7F. +static const uint32_t VMA_CORRUPTION_DETECTION_MAGIC_VALUE = 0x7F84E666; + +// Copy of some Vulkan definitions so we don't need to check their existence just to handle few constants. static const uint32_t VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY = 0x00000040; static const uint32_t VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY = 0x00000080; static const uint32_t VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_COPY = 0x00020000; - +static const uint32_t VK_IMAGE_CREATE_DISJOINT_BIT_COPY = 0x00000200; +static const int32_t VK_IMAGE_TILING_DRM_FORMAT_MODIFIER_EXT_COPY = 1000158000; static const uint32_t VMA_ALLOCATION_INTERNAL_STRATEGY_MIN_OFFSET = 0x10000000u; +static const uint32_t VMA_ALLOCATION_TRY_COUNT = 32; +static const uint32_t VMA_VENDOR_ID_AMD = 4098; -static VkAllocationCallbacks VmaEmptyAllocationCallbacks = { - VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL }; +// This one is tricky. Vulkan specification defines this code as available since +// Vulkan 1.0, but doesn't actually define it in Vulkan SDK earlier than 1.2.131. +// See pull request #207. +#define VK_ERROR_UNKNOWN_COPY ((VkResult)-13) -// Returns number of bits set to 1 in (v). + +#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", +}; +#endif + +static VkAllocationCallbacks VmaEmptyAllocationCallbacks = + { VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL, VMA_NULL }; + + +#ifndef _VMA_ENUM_DECLARATIONS + +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 +}; + +enum VMA_CACHE_OPERATION +{ + VMA_CACHE_FLUSH, + VMA_CACHE_INVALIDATE +}; + +enum class VmaAllocationRequestType +{ + Normal, + TLSF, + // Used by "Linear" algorithm. + UpperAddress, + EndOf1st, + EndOf2nd, +}; + +#endif // _VMA_ENUM_DECLARATIONS + +#ifndef _VMA_FORWARD_DECLARATIONS +// Opaque handle used by allocation algorithms to identify single allocation in any conforming way. +VK_DEFINE_NON_DISPATCHABLE_HANDLE(VmaAllocHandle); + +struct VmaMutexLock; +struct VmaMutexLockRead; +struct VmaMutexLockWrite; + +template +struct AtomicTransactionalIncrement; + +template +struct VmaStlAllocator; + +template +class VmaVector; + +template +class VmaSmallVector; + +template +class VmaPoolAllocator; + +template +struct VmaListItem; + +template +class VmaRawList; + +template +class VmaList; + +template +class VmaIntrusiveLinkedList; + +// Unused in this version +#if 0 +template +struct VmaPair; +template +struct VmaPairFirstLess; + +template +class VmaMap; +#endif + +#if VMA_STATS_STRING_ENABLED +class VmaStringBuilder; +class VmaJsonWriter; +#endif + +class VmaDeviceMemoryBlock; + +struct VmaDedicatedAllocationListItemTraits; +class VmaDedicatedAllocationList; + +struct VmaSuballocation; +struct VmaSuballocationOffsetLess; +struct VmaSuballocationOffsetGreater; +struct VmaSuballocationItemSizeLess; + +typedef VmaList> VmaSuballocationList; + +struct VmaAllocationRequest; + +class VmaBlockMetadata; +class VmaBlockMetadata_Linear; +class VmaBlockMetadata_TLSF; + +class VmaBlockVector; + +struct VmaPoolListItemTraits; + +struct VmaCurrentBudgetData; + +class VmaAllocationObjectAllocator; + +#endif // _VMA_FORWARD_DECLARATIONS + + +#ifndef _VMA_FUNCTIONS + +/* +Returns number of bits set to 1 in (v). + +On specific platforms and compilers you can use instrinsics like: + +Visual Studio: + return __popcnt(v); +GCC, Clang: + return static_cast(__builtin_popcount(v)); + +Define macro VMA_COUNT_BITS_SET to provide your optimized implementation. +But you need to check in runtime whether user's CPU supports these, as some old processors don't. +*/ static inline uint32_t VmaCountBitsSet(uint32_t v) { +#if __cplusplus >= 202002L || _MSVC_LANG >= 202002L // C++20 + return std::popcount(v); +#else 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 >> 2) & 0x33333333) + (c & 0x33333333); + c = ((c >> 4) + c) & 0x0F0F0F0F; + c = ((c >> 8) + c) & 0x00FF00FF; c = ((c >> 16) + c) & 0x0000FFFF; return c; +#endif +} + +static inline uint8_t VmaBitScanLSB(uint64_t mask) +{ +#if defined(_MSC_VER) && defined(_WIN64) + unsigned long pos; + if (_BitScanForward64(&pos, mask)) + return static_cast(pos); + return UINT8_MAX; +#elif defined __GNUC__ || defined __clang__ + return static_cast(__builtin_ffsll(mask)) - 1U; +#else + uint8_t pos = 0; + uint64_t bit = 1; + do + { + if (mask & bit) + return pos; + bit <<= 1; + } while (pos++ < 63); + return UINT8_MAX; +#endif +} + +static inline uint8_t VmaBitScanLSB(uint32_t mask) +{ +#ifdef _MSC_VER + unsigned long pos; + if (_BitScanForward(&pos, mask)) + return static_cast(pos); + return UINT8_MAX; +#elif defined __GNUC__ || defined __clang__ + return static_cast(__builtin_ffs(mask)) - 1U; +#else + uint8_t pos = 0; + uint32_t bit = 1; + do + { + if (mask & bit) + return pos; + bit <<= 1; + } while (pos++ < 31); + return UINT8_MAX; +#endif +} + +static inline uint8_t VmaBitScanMSB(uint64_t mask) +{ +#if defined(_MSC_VER) && defined(_WIN64) + unsigned long pos; + if (_BitScanReverse64(&pos, mask)) + return static_cast(pos); +#elif defined __GNUC__ || defined __clang__ + if (mask) + return 63 - static_cast(__builtin_clzll(mask)); +#else + uint8_t pos = 63; + uint64_t bit = 1ULL << 63; + do + { + if (mask & bit) + return pos; + bit >>= 1; + } while (pos-- > 0); +#endif + return UINT8_MAX; +} + +static inline uint8_t VmaBitScanMSB(uint32_t mask) +{ +#ifdef _MSC_VER + unsigned long pos; + if (_BitScanReverse(&pos, mask)) + return static_cast(pos); +#elif defined __GNUC__ || defined __clang__ + if (mask) + return 31 - static_cast(__builtin_clz(mask)); +#else + uint8_t pos = 31; + uint32_t bit = 1UL << 31; + do + { + if (mask & bit) + return pos; + bit >>= 1; + } while (pos-- > 0); +#endif + return UINT8_MAX; } /* @@ -2626,7 +3287,7 @@ For 0 returns true. template inline bool VmaIsPow2(T x) { - return (x & (x-1)) == 0; + return (x & (x - 1)) == 0; } // Aligns given value up to nearest multiply of align value. For example: VmaAlignUp(11, 8) = 16. @@ -2637,6 +3298,7 @@ static inline T VmaAlignUp(T val, T alignment) VMA_HEAVY_ASSERT(VmaIsPow2(alignment)); return (val + alignment - 1) & ~(alignment - 1); } + // 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 @@ -2653,6 +3315,13 @@ static inline T VmaRoundDiv(T x, T y) return (x + (y / (T)2)) / y; } +// Divide by 'y' and round up to nearest integer. +template +static inline T VmaDivideRoundingUp(T x, T y) +{ + return (x + y - (T)1) / y; +} + // Returns smallest power of 2 greater or equal to v. static inline uint32_t VmaNextPow2(uint32_t v) { @@ -2665,6 +3334,7 @@ static inline uint32_t VmaNextPow2(uint32_t v) v++; return v; } + static inline uint64_t VmaNextPow2(uint64_t v) { v--; @@ -2689,6 +3359,7 @@ static inline uint32_t VmaPrevPow2(uint32_t v) v = v ^ (v >> 1); return v; } + static inline uint64_t VmaPrevPow2(uint64_t v) { v |= v >> 1; @@ -2706,66 +3377,6 @@ static inline bool VmaStrIsEmpty(const char* pStr) return pStr == VMA_NULL || *pStr == '\0'; } -#if VMA_STATS_STRING_ENABLED - -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 ""; - } -} - -#endif // #if VMA_STATS_STRING_ENABLED - -#ifndef VMA_SORT - -template -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 -void VmaQuickSort(Iterator beg, Iterator end, Compare cmp) -{ - if(beg < end) - { - Iterator it = VmaQuickSortPartition(beg, end, cmp); - VmaQuickSort(beg, it, cmp); - VmaQuickSort(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. @@ -2787,17 +3398,6 @@ static inline bool VmaBlocksOnSamePage( 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 @@ -2808,12 +3408,12 @@ static inline bool VmaIsBufferImageGranularityConflict( VmaSuballocationType suballocType1, VmaSuballocationType suballocType2) { - if(suballocType1 > suballocType2) + if (suballocType1 > suballocType2) { VMA_SWAP(suballocType1, suballocType2); } - switch(suballocType1) + switch (suballocType1) { case VMA_SUBALLOCATION_TYPE_FREE: return false; @@ -2844,7 +3444,7 @@ static void VmaWriteMagicValue(void* pData, VkDeviceSize offset) #if VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_DETECT_CORRUPTION 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; } @@ -2858,9 +3458,9 @@ static bool VmaValidateMagicValue(const void* pData, VkDeviceSize offset) #if VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_DETECT_CORRUPTION 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) + if (*pSrc != VMA_CORRUPTION_DETECTION_MAGIC_VALUE) { return false; } @@ -2881,55 +3481,6 @@ static void VmaFillGpuDefragmentationBufferCreateInfo(VkBufferCreateInfo& outBuf outBufCreateInfo.size = (VkDeviceSize)VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE; // Example size. } -// 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 = true) : - 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 @@ -2941,13 +3492,13 @@ Returned value is the found element, if present in the collection or place where new element with value (key) should be inserted. */ template -static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT &key, const CmpLess& cmp) +static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp) { size_t down = 0, up = (end - beg); - while(down < up) + while (down < up) { const size_t mid = down + (up - down) / 2; // Overflow-safe midpoint calculation - if(cmp(*(beg+mid), key)) + if (cmp(*(beg + mid), key)) { down = mid + 1; } @@ -2964,7 +3515,7 @@ IterT VmaBinaryFindSorted(const IterT& beg, const IterT& end, const KeyT& value, { IterT it = VmaBinaryFindFirstNotLess( beg, end, value, cmp); - if(it == end || + if (it == end || (!cmp(*it, value) && !cmp(value, *it))) { return it; @@ -2980,16 +3531,16 @@ T must be pointer type, e.g. VmaAllocation, VmaPool. template static bool VmaValidatePointerArray(uint32_t count, const T* arr) { - for(uint32_t i = 0; i < count; ++i) + for (uint32_t i = 0; i < count; ++i) { const T iPtr = arr[i]; - if(iPtr == VMA_NULL) + if (iPtr == VMA_NULL) { return false; } - for(uint32_t j = i + 1; j < count; ++j) + for (uint32_t j = i + 1; j < count; ++j) { - if(iPtr == arr[j]) + if (iPtr == arr[j]) { return false; } @@ -3005,13 +3556,161 @@ static inline void VmaPnextChainPushFront(MainT* mainStruct, NewT* newStruct) mainStruct->pNext = newStruct; } +// This is the main algorithm that guides the selection of a memory type best for an allocation - +// converts usage to required/preferred/not preferred flags. +static bool FindMemoryPreferences( + bool isIntegratedGPU, + const VmaAllocationCreateInfo& allocCreateInfo, + VkFlags bufImgUsage, // VkBufferCreateInfo::usage or VkImageCreateInfo::usage. UINT32_MAX if unknown. + VkMemoryPropertyFlags& outRequiredFlags, + VkMemoryPropertyFlags& outPreferredFlags, + VkMemoryPropertyFlags& outNotPreferredFlags) +{ + outRequiredFlags = allocCreateInfo.requiredFlags; + outPreferredFlags = allocCreateInfo.preferredFlags; + outNotPreferredFlags = 0; + + switch(allocCreateInfo.usage) + { + case VMA_MEMORY_USAGE_UNKNOWN: + break; + case VMA_MEMORY_USAGE_GPU_ONLY: + if(!isIntegratedGPU || (outPreferredFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + break; + case VMA_MEMORY_USAGE_CPU_ONLY: + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; + break; + case VMA_MEMORY_USAGE_CPU_TO_GPU: + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + if(!isIntegratedGPU || (outPreferredFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + break; + case VMA_MEMORY_USAGE_GPU_TO_CPU: + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + outPreferredFlags |= VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + break; + case VMA_MEMORY_USAGE_CPU_COPY: + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + break; + case VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED: + outRequiredFlags |= VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT; + break; + case VMA_MEMORY_USAGE_AUTO: + case VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE: + case VMA_MEMORY_USAGE_AUTO_PREFER_HOST: + { + if(bufImgUsage == UINT32_MAX) + { + VMA_ASSERT(0 && "VMA_MEMORY_USAGE_AUTO* values can only be used with functions like vmaCreateBuffer, vmaCreateImage so that the details of the created resource are known."); + return false; + } + // This relies on values of VK_IMAGE_USAGE_TRANSFER* being the same VK_BUFFER_IMAGE_TRANSFER*. + const bool deviceAccess = (bufImgUsage & ~(VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT)) != 0; + const bool hostAccessSequentialWrite = (allocCreateInfo.flags & VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT) != 0; + const bool hostAccessRandom = (allocCreateInfo.flags & VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT) != 0; + const bool hostAccessAllowTransferInstead = (allocCreateInfo.flags & VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT) != 0; + const bool preferDevice = allocCreateInfo.usage == VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE; + const bool preferHost = allocCreateInfo.usage == VMA_MEMORY_USAGE_AUTO_PREFER_HOST; + + // CPU random access - e.g. a buffer written to or transferred from GPU to read back on CPU. + if(hostAccessRandom) + { + if(!isIntegratedGPU && deviceAccess && hostAccessAllowTransferInstead && !preferHost) + { + // Nice if it will end up in HOST_VISIBLE, but more importantly prefer DEVICE_LOCAL. + // Omitting HOST_VISIBLE here is intentional. + // In case there is DEVICE_LOCAL | HOST_VISIBLE | HOST_CACHED, it will pick that one. + // Otherwise, this will give same weight to DEVICE_LOCAL as HOST_VISIBLE | HOST_CACHED and select the former if occurs first on the list. + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + } + else + { + // Always CPU memory, cached. + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + } + } + // CPU sequential write - may be CPU or host-visible GPU memory, uncached and write-combined. + else if(hostAccessSequentialWrite) + { + // Want uncached and write-combined. + outNotPreferredFlags |= VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + + if(!isIntegratedGPU && deviceAccess && hostAccessAllowTransferInstead && !preferHost) + { + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + } + else + { + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + // Direct GPU access, CPU sequential write (e.g. a dynamic uniform buffer updated every frame) + if(deviceAccess) + { + // Could go to CPU memory or GPU BAR/unified. Up to the user to decide. If no preference, choose GPU memory. + if(preferHost) + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + else + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + // GPU no direct access, CPU sequential write (e.g. an upload buffer to be transferred to the GPU) + else + { + // Could go to CPU memory or GPU BAR/unified. Up to the user to decide. If no preference, choose CPU memory. + if(preferDevice) + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + else + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + } + } + // No CPU access + else + { + // GPU access, no CPU access (e.g. a color attachment image) - prefer GPU memory + if(deviceAccess) + { + // ...unless there is a clear preference from the user not to do so. + if(preferHost) + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + else + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + // No direct GPU access, no CPU access, just transfers. + // It may be staging copy intended for e.g. preserving image for next frame (then better GPU memory) or + // a "swap file" copy to free some GPU memory (then better CPU memory). + // Up to the user to decide. If no preferece, assume the former and choose GPU memory. + if(preferHost) + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + else + outPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + } + break; + } + default: + VMA_ASSERT(0); + } + + // Avoid DEVICE_COHERENT unless explicitly requested. + if(((allocCreateInfo.requiredFlags | allocCreateInfo.preferredFlags) & + (VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY | VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY)) == 0) + { + outNotPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY; + } + + return true; +} + //////////////////////////////////////////////////////////////////////////////// // Memory allocation static void* VmaMalloc(const VkAllocationCallbacks* pAllocationCallbacks, size_t size, size_t alignment) { void* result = VMA_NULL; - if((pAllocationCallbacks != VMA_NULL) && + if ((pAllocationCallbacks != VMA_NULL) && (pAllocationCallbacks->pfnAllocation != VMA_NULL)) { result = (*pAllocationCallbacks->pfnAllocation)( @@ -3030,7 +3729,7 @@ static void* VmaMalloc(const VkAllocationCallbacks* pAllocationCallbacks, size_t static void VmaFree(const VkAllocationCallbacks* pAllocationCallbacks, void* ptr) { - if((pAllocationCallbacks != VMA_NULL) && + if ((pAllocationCallbacks != VMA_NULL) && (pAllocationCallbacks->pfnFree != VMA_NULL)) { (*pAllocationCallbacks->pfnFree)(pAllocationCallbacks->pUserData, ptr); @@ -3067,9 +3766,9 @@ static void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr template static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count) { - if(ptr != VMA_NULL) + if (ptr != VMA_NULL) { - for(size_t i = count; i--; ) + for (size_t i = count; i--; ) { ptr[i].~T(); } @@ -3079,317 +3778,39 @@ static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, static char* VmaCreateStringCopy(const VkAllocationCallbacks* allocs, const char* srcStr) { - if(srcStr != VMA_NULL) + if (srcStr != VMA_NULL) { const size_t len = strlen(srcStr); char* const result = vma_new_array(allocs, char, len + 1); memcpy(result, srcStr, len + 1); return result; } - else - { - return VMA_NULL; - } + return VMA_NULL; } +#if VMA_STATS_STRING_ENABLED +static char* VmaCreateStringCopy(const VkAllocationCallbacks* allocs, const char* srcStr, size_t strLen) +{ + if (srcStr != VMA_NULL) + { + char* const result = vma_new_array(allocs, char, strLen + 1); + memcpy(result, srcStr, strLen); + result[strLen] = '\0'; + return result; + } + return VMA_NULL; +} +#endif // VMA_STATS_STRING_ENABLED + static void VmaFreeString(const VkAllocationCallbacks* allocs, char* str) { - if(str != VMA_NULL) + if (str != VMA_NULL) { const size_t len = strlen(str); vma_delete_array(allocs, str, len + 1); } } -// STL-compatible allocator. -template -class VmaStlAllocator -{ -public: - const VkAllocationCallbacks* const m_pCallbacks; - typedef T value_type; - - VmaStlAllocator(const VkAllocationCallbacks* pCallbacks) : m_pCallbacks(pCallbacks) { } - template VmaStlAllocator(const VmaStlAllocator& src) : m_pCallbacks(src.m_pCallbacks) { } - - T* allocate(size_t n) { return VmaAllocateArray(m_pCallbacks, n); } - void deallocate(T* p, size_t n) { VmaFree(m_pCallbacks, p); } - - template - bool operator==(const VmaStlAllocator& rhs) const - { - return m_pCallbacks == rhs.m_pCallbacks; - } - template - bool operator!=(const VmaStlAllocator& rhs) const - { - return m_pCallbacks != rhs.m_pCallbacks; - } - - VmaStlAllocator& operator=(const VmaStlAllocator& x) = delete; - VmaStlAllocator(const VmaStlAllocator&) = default; -}; - -#if VMA_USE_STL_VECTOR - -#define VmaVector std::vector - -template -static void VmaVectorInsert(std::vector& vec, size_t index, const T& item) -{ - vec.insert(vec.begin() + index, item); -} - -template -static void VmaVectorRemove(std::vector& 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 -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(allocator.m_pCallbacks, count) : VMA_NULL), - m_Count(count), - m_Capacity(count) - { - } - - // This version of the constructor is here for compatibility with pre-C++14 std::vector. - // value is unused. - VmaVector(size_t count, const T& value, const AllocatorT& allocator) - : VmaVector(count, allocator) {} - - VmaVector(const VmaVector& src) : - m_Allocator(src.m_Allocator), - m_pArray(src.m_Count ? (T*)VmaAllocateArray(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& 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(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) - { - size_t newCapacity = m_Capacity; - if(newCount > m_Capacity) - { - newCapacity = VMA_MAX(newCount, VMA_MAX(m_Capacity * 3 / 2, (size_t)8)); - } - - if(newCapacity != m_Capacity) - { - T* const newArray = newCapacity ? VmaAllocateArray(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() - { - resize(0); - } - - void shrink_to_fit() - { - if(m_Capacity > m_Count) - { - T* newArray = VMA_NULL; - if(m_Count > 0) - { - newArray = VmaAllocateArray(m_Allocator.m_pCallbacks, m_Count); - memcpy(newArray, m_pArray, m_Count * sizeof(T)); - } - VmaFree(m_Allocator.m_pCallbacks, m_pArray); - m_Capacity = m_Count; - m_pArray = newArray; - } - } - - 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; - typedef const T* const_iterator; - - iterator begin() { return m_pArray; } - iterator end() { return m_pArray + m_Count; } - const_iterator cbegin() const { return m_pArray; } - const_iterator cend() const { return m_pArray + m_Count; } - const_iterator begin() const { return cbegin(); } - const_iterator end() const { return cend(); } - -private: - AllocatorT m_Allocator; - T* m_pArray; - size_t m_Count; - size_t m_Capacity; -}; - -template -static void VmaVectorInsert(VmaVector& vec, size_t index, const T& item) -{ - vec.insert(index, item); -} - -template -static void VmaVectorRemove(VmaVector& vec, size_t index) -{ - vec.remove(index); -} - -#endif // #if VMA_USE_STL_VECTOR - template size_t VmaVectorInsertSorted(VectorT& vector, const typename VectorT::value_type& value) { @@ -3411,7 +3832,7 @@ bool VmaVectorRemoveSorted(VectorT& vector, const typename VectorT::value_type& vector.end(), value, comparator); - if((it != vector.end()) && !comparator(*it, value) && !comparator(value, *it)) + if ((it != vector.end()) && !comparator(*it, value) && !comparator(value, *it)) { size_t indexToRemove = it - vector.begin(); VmaVectorRemove(vector, indexToRemove); @@ -3419,10 +3840,390 @@ bool VmaVectorRemoveSorted(VectorT& vector, const typename VectorT::value_type& } return false; } +#endif // _VMA_FUNCTIONS -//////////////////////////////////////////////////////////////////////////////// -// class VmaSmallVector +#ifndef _VMA_STATISTICS_FUNCTIONS +static void VmaClearStatistics(VmaStatistics& outStats) +{ + outStats.blockCount = 0; + outStats.allocationCount = 0; + outStats.blockBytes = 0; + outStats.allocationBytes = 0; +} + +static void VmaAddStatistics(VmaStatistics& inoutStats, const VmaStatistics& src) +{ + inoutStats.blockCount += src.blockCount; + inoutStats.allocationCount += src.allocationCount; + inoutStats.blockBytes += src.blockBytes; + inoutStats.allocationBytes += src.allocationBytes; +} + +static void VmaClearDetailedStatistics(VmaDetailedStatistics& outStats) +{ + VmaClearStatistics(outStats.statistics); + outStats.unusedRangeCount = 0; + outStats.allocationSizeMin = VK_WHOLE_SIZE; + outStats.allocationSizeMax = 0; + outStats.unusedRangeSizeMin = VK_WHOLE_SIZE; + outStats.unusedRangeSizeMax = 0; +} + +static void VmaAddDetailedStatisticsAllocation(VmaDetailedStatistics& inoutStats, VkDeviceSize size) +{ + inoutStats.statistics.allocationCount++; + inoutStats.statistics.allocationBytes += size; + inoutStats.allocationSizeMin = VMA_MIN(inoutStats.allocationSizeMin, size); + inoutStats.allocationSizeMax = VMA_MAX(inoutStats.allocationSizeMax, size); +} + +static void VmaAddDetailedStatisticsUnusedRange(VmaDetailedStatistics& inoutStats, VkDeviceSize size) +{ + inoutStats.unusedRangeCount++; + inoutStats.unusedRangeSizeMin = VMA_MIN(inoutStats.unusedRangeSizeMin, size); + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, size); +} + +static void VmaAddDetailedStatistics(VmaDetailedStatistics& inoutStats, const VmaDetailedStatistics& src) +{ + VmaAddStatistics(inoutStats.statistics, src.statistics); + inoutStats.unusedRangeCount += src.unusedRangeCount; + inoutStats.allocationSizeMin = VMA_MIN(inoutStats.allocationSizeMin, src.allocationSizeMin); + inoutStats.allocationSizeMax = VMA_MAX(inoutStats.allocationSizeMax, src.allocationSizeMax); + inoutStats.unusedRangeSizeMin = VMA_MIN(inoutStats.unusedRangeSizeMin, src.unusedRangeSizeMin); + inoutStats.unusedRangeSizeMax = VMA_MAX(inoutStats.unusedRangeSizeMax, src.unusedRangeSizeMax); +} + +#endif // _VMA_STATISTICS_FUNCTIONS + +#ifndef _VMA_MUTEX_LOCK +// 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 = true) : + 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 +#endif // _VMA_MUTEX_LOCK + +#ifndef _VMA_ATOMIC_TRANSACTIONAL_INCREMENT +// An object that increments given atomic but decrements it back in the destructor unless Commit() is called. +template +struct AtomicTransactionalIncrement +{ +public: + typedef std::atomic AtomicT; + + ~AtomicTransactionalIncrement() + { + if(m_Atomic) + --(*m_Atomic); + } + + void Commit() { m_Atomic = nullptr; } + T Increment(AtomicT* atomic) + { + m_Atomic = atomic; + return m_Atomic->fetch_add(1); + } + +private: + AtomicT* m_Atomic = nullptr; +}; +#endif // _VMA_ATOMIC_TRANSACTIONAL_INCREMENT + +#ifndef _VMA_STL_ALLOCATOR +// STL-compatible allocator. +template +struct VmaStlAllocator +{ + const VkAllocationCallbacks* const m_pCallbacks; + typedef T value_type; + + VmaStlAllocator(const VkAllocationCallbacks* pCallbacks) : m_pCallbacks(pCallbacks) {} + template + VmaStlAllocator(const VmaStlAllocator& src) : m_pCallbacks(src.m_pCallbacks) {} + VmaStlAllocator(const VmaStlAllocator&) = default; + VmaStlAllocator& operator=(const VmaStlAllocator&) = delete; + + T* allocate(size_t n) { return VmaAllocateArray(m_pCallbacks, n); } + void deallocate(T* p, size_t n) { VmaFree(m_pCallbacks, p); } + + template + bool operator==(const VmaStlAllocator& rhs) const + { + return m_pCallbacks == rhs.m_pCallbacks; + } + template + bool operator!=(const VmaStlAllocator& rhs) const + { + return m_pCallbacks != rhs.m_pCallbacks; + } +}; +#endif // _VMA_STL_ALLOCATOR + +#ifndef _VMA_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 +class VmaVector +{ +public: + typedef T value_type; + typedef T* iterator; + typedef const T* const_iterator; + + VmaVector(const AllocatorT& allocator); + VmaVector(size_t count, const AllocatorT& allocator); + // This version of the constructor is here for compatibility with pre-C++14 std::vector. + // value is unused. + VmaVector(size_t count, const T& value, const AllocatorT& allocator) : VmaVector(count, allocator) {} + VmaVector(const VmaVector& src); + VmaVector& operator=(const VmaVector& rhs); + ~VmaVector() { VmaFree(m_Allocator.m_pCallbacks, m_pArray); } + + bool empty() const { return m_Count == 0; } + size_t size() const { return m_Count; } + T* data() { return m_pArray; } + T& front() { 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* data() const { return m_pArray; } + const T& front() const { VMA_HEAVY_ASSERT(m_Count > 0); return m_pArray[0]; } + const T& back() const { VMA_HEAVY_ASSERT(m_Count > 0); return m_pArray[m_Count - 1]; } + + iterator begin() { return m_pArray; } + iterator end() { return m_pArray + m_Count; } + const_iterator cbegin() const { return m_pArray; } + const_iterator cend() const { return m_pArray + m_Count; } + const_iterator begin() const { return cbegin(); } + const_iterator end() const { return cend(); } + + void pop_front() { VMA_HEAVY_ASSERT(m_Count > 0); remove(0); } + void pop_back() { VMA_HEAVY_ASSERT(m_Count > 0); resize(size() - 1); } + void push_front(const T& src) { insert(0, src); } + + void push_back(const T& src); + void reserve(size_t newCapacity, bool freeMemory = false); + void resize(size_t newCount); + void clear() { resize(0); } + void shrink_to_fit(); + void insert(size_t index, const T& src); + void remove(size_t index); + + 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]; } + +private: + AllocatorT m_Allocator; + T* m_pArray; + size_t m_Count; + size_t m_Capacity; +}; + +#ifndef _VMA_VECTOR_FUNCTIONS +template +VmaVector::VmaVector(const AllocatorT& allocator) + : m_Allocator(allocator), + m_pArray(VMA_NULL), + m_Count(0), + m_Capacity(0) {} + +template +VmaVector::VmaVector(size_t count, const AllocatorT& allocator) + : m_Allocator(allocator), + m_pArray(count ? (T*)VmaAllocateArray(allocator.m_pCallbacks, count) : VMA_NULL), + m_Count(count), + m_Capacity(count) {} + +template +VmaVector::VmaVector(const VmaVector& src) + : m_Allocator(src.m_Allocator), + m_pArray(src.m_Count ? (T*)VmaAllocateArray(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)); + } +} + +template +VmaVector& VmaVector::operator=(const VmaVector& rhs) +{ + if (&rhs != this) + { + resize(rhs.m_Count); + if (m_Count != 0) + { + memcpy(m_pArray, rhs.m_pArray, m_Count * sizeof(T)); + } + } + return *this; +} + +template +void VmaVector::push_back(const T& src) +{ + const size_t newIndex = size(); + resize(newIndex + 1); + m_pArray[newIndex] = src; +} + +template +void VmaVector::reserve(size_t newCapacity, bool freeMemory) +{ + newCapacity = VMA_MAX(newCapacity, m_Count); + + if ((newCapacity < m_Capacity) && !freeMemory) + { + newCapacity = m_Capacity; + } + + if (newCapacity != m_Capacity) + { + T* const newArray = newCapacity ? VmaAllocateArray(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; + } +} + +template +void VmaVector::resize(size_t newCount) +{ + size_t newCapacity = m_Capacity; + if (newCount > m_Capacity) + { + newCapacity = VMA_MAX(newCount, VMA_MAX(m_Capacity * 3 / 2, (size_t)8)); + } + + if (newCapacity != m_Capacity) + { + T* const newArray = newCapacity ? VmaAllocateArray(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; +} + +template +void VmaVector::shrink_to_fit() +{ + if (m_Capacity > m_Count) + { + T* newArray = VMA_NULL; + if (m_Count > 0) + { + newArray = VmaAllocateArray(m_Allocator.m_pCallbacks, m_Count); + memcpy(newArray, m_pArray, m_Count * sizeof(T)); + } + VmaFree(m_Allocator.m_pCallbacks, m_pArray); + m_Capacity = m_Count; + m_pArray = newArray; + } +} + +template +void VmaVector::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; +} + +template +void VmaVector::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); +} +#endif // _VMA_VECTOR_FUNCTIONS + +template +static void VmaVectorInsert(VmaVector& vec, size_t index, const T& item) +{ + vec.insert(index, item); +} + +template +static void VmaVectorRemove(VmaVector& vec, size_t index) +{ + vec.remove(index); +} +#endif // _VMA_VECTOR + +#ifndef _VMA_SMALL_VECTOR /* This is a vector (a variable-sized array), optimized for the case when the array is small. @@ -3430,180 +4231,155 @@ It contains some number of elements in-place, which allows it to avoid heap allo when the actual number of elements is below that threshold. This allows normal "small" cases to be fast without losing generality for large inputs. */ - template class VmaSmallVector { public: typedef T value_type; + typedef T* iterator; - VmaSmallVector(const AllocatorT& allocator) : - m_Count(0), - m_DynamicArray(allocator) - { - } - VmaSmallVector(size_t count, const AllocatorT& allocator) : - m_Count(count), - m_DynamicArray(count > N ? count : 0, allocator) - { - } + VmaSmallVector(const AllocatorT& allocator); + VmaSmallVector(size_t count, const AllocatorT& allocator); template - VmaSmallVector(const VmaSmallVector& src) = delete; + VmaSmallVector(const VmaSmallVector&) = delete; template - VmaSmallVector& operator=(const VmaSmallVector& rhs) = delete; + VmaSmallVector& operator=(const VmaSmallVector&) = delete; + ~VmaSmallVector() = default; bool empty() const { return m_Count == 0; } size_t size() const { return m_Count; } T* data() { return m_Count > N ? m_DynamicArray.data() : m_StaticArray; } + T& front() { VMA_HEAVY_ASSERT(m_Count > 0); return data()[0]; } + T& back() { VMA_HEAVY_ASSERT(m_Count > 0); return data()[m_Count - 1]; } const T* data() const { return m_Count > N ? m_DynamicArray.data() : m_StaticArray; } - - T& operator[](size_t index) - { - VMA_HEAVY_ASSERT(index < m_Count); - return data()[index]; - } - const T& operator[](size_t index) const - { - VMA_HEAVY_ASSERT(index < m_Count); - return data()[index]; - } - - T& front() - { - VMA_HEAVY_ASSERT(m_Count > 0); - return data()[0]; - } - const T& front() const - { - VMA_HEAVY_ASSERT(m_Count > 0); - return data()[0]; - } - T& back() - { - VMA_HEAVY_ASSERT(m_Count > 0); - return data()[m_Count - 1]; - } - const T& back() const - { - VMA_HEAVY_ASSERT(m_Count > 0); - return data()[m_Count - 1]; - } - - void resize(size_t newCount, bool freeMemory = false) - { - if(newCount > N && m_Count > N) - { - // Any direction, staying in m_DynamicArray - m_DynamicArray.resize(newCount); - if(freeMemory) - { - m_DynamicArray.shrink_to_fit(); - } - } - else if(newCount > N && m_Count <= N) - { - // Growing, moving from m_StaticArray to m_DynamicArray - m_DynamicArray.resize(newCount); - if(m_Count > 0) - { - memcpy(m_DynamicArray.data(), m_StaticArray, m_Count * sizeof(T)); - } - } - else if(newCount <= N && m_Count > N) - { - // Shrinking, moving from m_DynamicArray to m_StaticArray - if(newCount > 0) - { - memcpy(m_StaticArray, m_DynamicArray.data(), newCount * sizeof(T)); - } - m_DynamicArray.resize(0); - if(freeMemory) - { - m_DynamicArray.shrink_to_fit(); - } - } - else - { - // Any direction, staying in m_StaticArray - nothing to do here - } - m_Count = newCount; - } - - void clear(bool freeMemory = false) - { - m_DynamicArray.clear(); - if(freeMemory) - { - m_DynamicArray.shrink_to_fit(); - } - m_Count = 0; - } - - void insert(size_t index, const T& src) - { - VMA_HEAVY_ASSERT(index <= m_Count); - const size_t oldCount = size(); - resize(oldCount + 1); - T* const dataPtr = data(); - if(index < oldCount) - { - // I know, this could be more optimal for case where memmove can be memcpy directly from m_StaticArray to m_DynamicArray. - memmove(dataPtr + (index + 1), dataPtr + index, (oldCount - index) * sizeof(T)); - } - dataPtr[index] = src; - } - - void remove(size_t index) - { - VMA_HEAVY_ASSERT(index < m_Count); - const size_t oldCount = size(); - if(index < oldCount - 1) - { - // I know, this could be more optimal for case where memmove can be memcpy directly from m_DynamicArray to m_StaticArray. - T* const dataPtr = data(); - memmove(dataPtr + index, dataPtr + (index + 1), (oldCount - index - 1) * sizeof(T)); - } - resize(oldCount - 1); - } - - void push_back(const T& src) - { - const size_t newIndex = size(); - resize(newIndex + 1); - data()[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; + const T& front() const { VMA_HEAVY_ASSERT(m_Count > 0); return data()[0]; } + const T& back() const { VMA_HEAVY_ASSERT(m_Count > 0); return data()[m_Count - 1]; } iterator begin() { return data(); } iterator end() { return data() + m_Count; } + void pop_front() { VMA_HEAVY_ASSERT(m_Count > 0); remove(0); } + void pop_back() { VMA_HEAVY_ASSERT(m_Count > 0); resize(size() - 1); } + void push_front(const T& src) { insert(0, src); } + + void push_back(const T& src); + void resize(size_t newCount, bool freeMemory = false); + void clear(bool freeMemory = false); + void insert(size_t index, const T& src); + void remove(size_t index); + + T& operator[](size_t index) { VMA_HEAVY_ASSERT(index < m_Count); return data()[index]; } + const T& operator[](size_t index) const { VMA_HEAVY_ASSERT(index < m_Count); return data()[index]; } + private: size_t m_Count; T m_StaticArray[N]; // Used when m_Size <= N VmaVector m_DynamicArray; // Used when m_Size > N }; -//////////////////////////////////////////////////////////////////////////////// -// class VmaPoolAllocator +#ifndef _VMA_SMALL_VECTOR_FUNCTIONS +template +VmaSmallVector::VmaSmallVector(const AllocatorT& allocator) + : m_Count(0), + m_DynamicArray(allocator) {} +template +VmaSmallVector::VmaSmallVector(size_t count, const AllocatorT& allocator) + : m_Count(count), + m_DynamicArray(count > N ? count : 0, allocator) {} + +template +void VmaSmallVector::push_back(const T& src) +{ + const size_t newIndex = size(); + resize(newIndex + 1); + data()[newIndex] = src; +} + +template +void VmaSmallVector::resize(size_t newCount, bool freeMemory) +{ + if (newCount > N && m_Count > N) + { + // Any direction, staying in m_DynamicArray + m_DynamicArray.resize(newCount); + if (freeMemory) + { + m_DynamicArray.shrink_to_fit(); + } + } + else if (newCount > N && m_Count <= N) + { + // Growing, moving from m_StaticArray to m_DynamicArray + m_DynamicArray.resize(newCount); + if (m_Count > 0) + { + memcpy(m_DynamicArray.data(), m_StaticArray, m_Count * sizeof(T)); + } + } + else if (newCount <= N && m_Count > N) + { + // Shrinking, moving from m_DynamicArray to m_StaticArray + if (newCount > 0) + { + memcpy(m_StaticArray, m_DynamicArray.data(), newCount * sizeof(T)); + } + m_DynamicArray.resize(0); + if (freeMemory) + { + m_DynamicArray.shrink_to_fit(); + } + } + else + { + // Any direction, staying in m_StaticArray - nothing to do here + } + m_Count = newCount; +} + +template +void VmaSmallVector::clear(bool freeMemory) +{ + m_DynamicArray.clear(); + if (freeMemory) + { + m_DynamicArray.shrink_to_fit(); + } + m_Count = 0; +} + +template +void VmaSmallVector::insert(size_t index, const T& src) +{ + VMA_HEAVY_ASSERT(index <= m_Count); + const size_t oldCount = size(); + resize(oldCount + 1); + T* const dataPtr = data(); + if (index < oldCount) + { + // I know, this could be more optimal for case where memmove can be memcpy directly from m_StaticArray to m_DynamicArray. + memmove(dataPtr + (index + 1), dataPtr + index, (oldCount - index) * sizeof(T)); + } + dataPtr[index] = src; +} + +template +void VmaSmallVector::remove(size_t index) +{ + VMA_HEAVY_ASSERT(index < m_Count); + const size_t oldCount = size(); + if (index < oldCount - 1) + { + // I know, this could be more optimal for case where memmove can be memcpy directly from m_DynamicArray to m_StaticArray. + T* const dataPtr = data(); + memmove(dataPtr + index, dataPtr + (index + 1), (oldCount - index - 1) * sizeof(T)); + } + resize(oldCount - 1); +} +#endif // _VMA_SMALL_VECTOR_FUNCTIONS +#endif // _VMA_SMALL_VECTOR + +#ifndef _VMA_POOL_ALLOCATOR /* 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 @@ -3625,7 +4401,6 @@ private: uint32_t NextFreeIndex; alignas(T) char Value[sizeof(T)]; }; - struct ItemBlock { Item* pItems; @@ -3635,14 +4410,15 @@ private: const VkAllocationCallbacks* m_pAllocationCallbacks; const uint32_t m_FirstBlockCapacity; - VmaVector< ItemBlock, VmaStlAllocator > m_ItemBlocks; + VmaVector> m_ItemBlocks; ItemBlock& CreateNewBlock(); }; +#ifndef _VMA_POOL_ALLOCATOR_FUNCTIONS template -VmaPoolAllocator::VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, uint32_t firstBlockCapacity) : - m_pAllocationCallbacks(pAllocationCallbacks), +VmaPoolAllocator::VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, uint32_t firstBlockCapacity) + : m_pAllocationCallbacks(pAllocationCallbacks), m_FirstBlockCapacity(firstBlockCapacity), m_ItemBlocks(VmaStlAllocator(pAllocationCallbacks)) { @@ -3652,7 +4428,7 @@ VmaPoolAllocator::VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCa template VmaPoolAllocator::~VmaPoolAllocator() { - for(size_t i = m_ItemBlocks.size(); i--; ) + for (size_t i = m_ItemBlocks.size(); i--;) vma_delete_array(m_pAllocationCallbacks, m_ItemBlocks[i].pItems, m_ItemBlocks[i].Capacity); m_ItemBlocks.clear(); } @@ -3660,11 +4436,11 @@ VmaPoolAllocator::~VmaPoolAllocator() template template T* VmaPoolAllocator::Alloc(Types&&... args) { - for(size_t i = m_ItemBlocks.size(); i--; ) + 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) + if (block.FirstFreeIndex != UINT32_MAX) { Item* const pItem = &block.pItems[block.FirstFreeIndex]; block.FirstFreeIndex = pItem->NextFreeIndex; @@ -3679,7 +4455,7 @@ template T* VmaPoolAllocator::Alloc(Types&&... args) Item* const pItem = &newBlock.pItems[0]; newBlock.FirstFreeIndex = pItem->NextFreeIndex; T* result = (T*)&pItem->Value; - new(result)T(std::forward(args)...); // Explicit constructor call. + new(result) T(std::forward(args)...); // Explicit constructor call. return result; } @@ -3687,7 +4463,7 @@ template void VmaPoolAllocator::Free(T* ptr) { // Search all memory blocks to find ptr. - for(size_t i = m_ItemBlocks.size(); i--; ) + for (size_t i = m_ItemBlocks.size(); i--; ) { ItemBlock& block = m_ItemBlocks[i]; @@ -3696,7 +4472,7 @@ void VmaPoolAllocator::Free(T* ptr) memcpy(&pItemPtr, &ptr, sizeof(pItemPtr)); // Check if pItemPtr is in address range of this block. - if((pItemPtr >= block.pItems) && (pItemPtr < block.pItems + block.Capacity)) + if ((pItemPtr >= block.pItems) && (pItemPtr < block.pItems + block.Capacity)) { ptr->~T(); // Explicit destructor call. const uint32_t index = static_cast(pItemPtr - block.pItems); @@ -3714,29 +4490,25 @@ typename VmaPoolAllocator::ItemBlock& VmaPoolAllocator::CreateNewBlock() const uint32_t newBlockCapacity = m_ItemBlocks.empty() ? m_FirstBlockCapacity : m_ItemBlocks.back().Capacity * 3 / 2; - const ItemBlock newBlock = { + const ItemBlock newBlock = + { vma_new_array(m_pAllocationCallbacks, Item, newBlockCapacity), newBlockCapacity, - 0 }; + 0 + }; m_ItemBlocks.push_back(newBlock); // Setup singly-linked list of all free items in this block. - for(uint32_t i = 0; i < newBlockCapacity - 1; ++i) + for (uint32_t i = 0; i < newBlockCapacity - 1; ++i) newBlock.pItems[i].NextFreeIndex = i + 1; newBlock.pItems[newBlockCapacity - 1].NextFreeIndex = UINT32_MAX; return m_ItemBlocks.back(); } +#endif // _VMA_POOL_ALLOCATOR_FUNCTIONS +#endif // _VMA_POOL_ALLOCATOR -//////////////////////////////////////////////////////////////////////////////// -// class VmaRawList, VmaList - -#if VMA_USE_STL_LIST - -#define VmaList std::list - -#else // #if VMA_USE_STL_LIST - +#ifndef _VMA_RAW_LIST template struct VmaListItem { @@ -3754,32 +4526,33 @@ public: typedef VmaListItem ItemType; VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks); - ~VmaRawList(); - void Clear(); + // Intentionally not calling Clear, because that would be unnecessary + // computations to return all items to m_ItemAllocator as free. + ~VmaRawList() = default; 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* Front() const { return m_pFront; } const ItemType* Back() const { return m_pBack; } - ItemType* PushBack(); ItemType* PushFront(); - ItemType* PushBack(const T& value); + ItemType* PushBack(); ItemType* PushFront(const T& value); - void PopBack(); + ItemType* PushBack(const T& value); void PopFront(); + void PopBack(); // 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 Clear(); void Remove(ItemType* pItem); private: @@ -3790,39 +4563,35 @@ private: size_t m_Count; }; +#ifndef _VMA_RAW_LIST_FUNCTIONS template -VmaRawList::VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks) : - m_pAllocationCallbacks(pAllocationCallbacks), +VmaRawList::VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks) + : m_pAllocationCallbacks(pAllocationCallbacks), m_ItemAllocator(pAllocationCallbacks, 128), m_pFront(VMA_NULL), m_pBack(VMA_NULL), - m_Count(0) -{ -} + m_Count(0) {} template -VmaRawList::~VmaRawList() +VmaListItem* VmaRawList::PushFront() { - // Intentionally not calling Clear, because that would be unnecessary - // computations to return all items to m_ItemAllocator as free. -} - -template -void VmaRawList::Clear() -{ - if(IsEmpty() == false) + ItemType* const pNewItem = m_ItemAllocator.Alloc(); + pNewItem->pPrev = VMA_NULL; + if (IsEmpty()) { - 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; + 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 @@ -3848,24 +4617,10 @@ VmaListItem* VmaRawList::PushBack() } template -VmaListItem* VmaRawList::PushFront() +VmaListItem* VmaRawList::PushFront(const T& value) { - 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; - } + ItemType* const pNewItem = PushFront(); + pNewItem->Value = value; return pNewItem; } @@ -3878,11 +4633,18 @@ VmaListItem* VmaRawList::PushBack(const T& value) } template -VmaListItem* VmaRawList::PushFront(const T& value) +void VmaRawList::PopFront() { - ItemType* const pNewItem = PushFront(); - pNewItem->Value = value; - return pNewItem; + 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 @@ -3901,18 +4663,21 @@ void VmaRawList::PopBack() } template -void VmaRawList::PopFront() +void VmaRawList::Clear() { - VMA_HEAVY_ASSERT(m_Count > 0); - ItemType* const pFrontItem = m_pFront; - ItemType* const pNextItem = pFrontItem->pNext; - if(pNextItem != VMA_NULL) + if (IsEmpty() == false) { - pNextItem->pPrev = VMA_NULL; + 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; } - m_pFront = pNextItem; - m_ItemAllocator.Free(pFrontItem); - --m_Count; } template @@ -4012,173 +4777,129 @@ VmaListItem* VmaRawList::InsertAfter(ItemType* pItem, const T& value) newItem->Value = value; return newItem; } +#endif // _VMA_RAW_LIST_FUNCTIONS +#endif // _VMA_RAW_LIST +#ifndef _VMA_LIST template class VmaList { VMA_CLASS_NO_COPY(VmaList) public: + class reverse_iterator; + class const_iterator; + class const_reverse_iterator; + class iterator { + friend class const_iterator; + friend class VmaList; public: - iterator() : - m_pList(VMA_NULL), - m_pItem(VMA_NULL) - { - } + iterator() : m_pList(VMA_NULL), m_pItem(VMA_NULL) {} + iterator(const reverse_iterator& src) : m_pList(src.m_pList), m_pItem(src.m_pItem) {} - 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; - } + 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; - } + 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; } - iterator operator++(int) - { - iterator result = *this; - ++*this; - return result; - } - iterator operator--(int) - { - iterator result = *this; - --*this; - return result; - } + 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; - } + iterator& operator++() { VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); m_pItem = m_pItem->pNext; return *this; } + iterator& operator--(); private: VmaRawList* m_pList; VmaListItem* m_pItem; - iterator(VmaRawList* pList, VmaListItem* pItem) : - m_pList(pList), - m_pItem(pItem) - { - } - - friend class VmaList; + iterator(VmaRawList* pList, VmaListItem* pItem) : m_pList(pList), m_pItem(pItem) {} }; - - class const_iterator + class reverse_iterator { + friend class const_reverse_iterator; + friend class VmaList; public: - const_iterator() : - m_pList(VMA_NULL), - m_pItem(VMA_NULL) - { - } + reverse_iterator() : m_pList(VMA_NULL), m_pItem(VMA_NULL) {} + reverse_iterator(const iterator& src) : m_pList(src.m_pList), m_pItem(src.m_pItem) {} - const_iterator(const iterator& src) : - m_pList(src.m_pList), - m_pItem(src.m_pItem) - { - } + 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; } - 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; - } + bool operator==(const reverse_iterator& rhs) const { VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); return m_pItem == rhs.m_pItem; } + bool operator!=(const reverse_iterator& rhs) const { VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); return m_pItem != rhs.m_pItem; } - 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; - } + reverse_iterator operator++(int) { reverse_iterator result = *this; ++* this; return result; } + reverse_iterator operator--(int) { reverse_iterator result = *this; --* this; return result; } - 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; - } + reverse_iterator& operator++() { VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); m_pItem = m_pItem->pPrev; return *this; } + reverse_iterator& operator--(); private: - const_iterator(const VmaRawList* pList, const VmaListItem* pItem) : - m_pList(pList), - m_pItem(pItem) - { - } + VmaRawList* m_pList; + VmaListItem* m_pItem; + reverse_iterator(VmaRawList* pList, VmaListItem* pItem) : m_pList(pList), m_pItem(pItem) {} + }; + class const_iterator + { + friend class VmaList; + 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_iterator(const reverse_iterator& src) : m_pList(src.m_pList), m_pItem(src.m_pItem) {} + + iterator drop_const() { return { const_cast*>(m_pList), const_cast*>(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; } + + 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; } + + const_iterator operator++(int) { const_iterator result = *this; ++* this; return result; } + const_iterator operator--(int) { const_iterator result = *this; --* this; return result; } + + const_iterator& operator++() { VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); m_pItem = m_pItem->pNext; return *this; } + const_iterator& operator--(); + + private: const VmaRawList* m_pList; const VmaListItem* m_pItem; + const_iterator(const VmaRawList* pList, const VmaListItem* pItem) : m_pList(pList), m_pItem(pItem) {} + }; + class const_reverse_iterator + { friend class VmaList; + public: + const_reverse_iterator() : m_pList(VMA_NULL), m_pItem(VMA_NULL) {} + const_reverse_iterator(const reverse_iterator& src) : m_pList(src.m_pList), m_pItem(src.m_pItem) {} + const_reverse_iterator(const iterator& src) : m_pList(src.m_pList), m_pItem(src.m_pItem) {} + + reverse_iterator drop_const() { return { const_cast*>(m_pList), const_cast*>(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; } + + bool operator==(const const_reverse_iterator& rhs) const { VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); return m_pItem == rhs.m_pItem; } + bool operator!=(const const_reverse_iterator& rhs) const { VMA_HEAVY_ASSERT(m_pList == rhs.m_pList); return m_pItem != rhs.m_pItem; } + + const_reverse_iterator operator++(int) { const_reverse_iterator result = *this; ++* this; return result; } + const_reverse_iterator operator--(int) { const_reverse_iterator result = *this; --* this; return result; } + + const_reverse_iterator& operator++() { VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); m_pItem = m_pItem->pPrev; return *this; } + const_reverse_iterator& operator--(); + + private: + const VmaRawList* m_pList; + const VmaListItem* m_pItem; + + const_reverse_iterator(const VmaRawList* pList, const VmaListItem* pItem) : m_pList(pList), m_pItem(pItem) {} }; - VmaList(const AllocatorT& allocator) : m_RawList(allocator.m_pCallbacks) { } + VmaList(const AllocatorT& allocator) : m_RawList(allocator.m_pCallbacks) {} bool empty() const { return m_RawList.IsEmpty(); } size_t size() const { return m_RawList.GetCount(); } @@ -4192,20 +4913,89 @@ public: const_iterator begin() const { return cbegin(); } const_iterator end() const { return cend(); } - void clear() { m_RawList.Clear(); } + reverse_iterator rbegin() { return reverse_iterator(&m_RawList, m_RawList.Back()); } + reverse_iterator rend() { return reverse_iterator(&m_RawList, VMA_NULL); } + + const_reverse_iterator crbegin() const { return const_reverse_iterator(&m_RawList, m_RawList.Back()); } + const_reverse_iterator crend() const { return const_reverse_iterator(&m_RawList, VMA_NULL); } + + const_reverse_iterator rbegin() const { return crbegin(); } + const_reverse_iterator rend() const { return crend(); } + 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)); } + void clear() { m_RawList.Clear(); } + void erase(iterator it) { m_RawList.Remove(it.m_pItem); } + private: VmaRawList m_RawList; }; -#endif // #if VMA_USE_STL_LIST +#ifndef _VMA_LIST_FUNCTIONS +template +typename VmaList::iterator& VmaList::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; +} -//////////////////////////////////////////////////////////////////////////////// -// class VmaIntrusiveLinkedList +template +typename VmaList::reverse_iterator& VmaList::reverse_iterator::operator--() +{ + if (m_pItem != VMA_NULL) + { + m_pItem = m_pItem->pNext; + } + else + { + VMA_HEAVY_ASSERT(!m_pList->IsEmpty()); + m_pItem = m_pList->Front(); + } + return *this; +} +template +typename VmaList::const_iterator& VmaList::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; +} + +template +typename VmaList::const_reverse_iterator& VmaList::const_reverse_iterator::operator--() +{ + if (m_pItem != VMA_NULL) + { + m_pItem = m_pItem->pNext; + } + else + { + VMA_HEAVY_ASSERT(!m_pList->IsEmpty()); + m_pItem = m_pList->Back(); + } + return *this; +} +#endif // _VMA_LIST_FUNCTIONS +#endif // _VMA_LIST + +#ifndef _VMA_INTRUSIVE_LINKED_LIST /* Expected interface of ItemTypeTraits: struct MyItemTypeTraits @@ -4224,251 +5014,249 @@ public: typedef typename ItemTypeTraits::ItemType ItemType; static ItemType* GetPrev(const ItemType* item) { return ItemTypeTraits::GetPrev(item); } static ItemType* GetNext(const ItemType* item) { return ItemTypeTraits::GetNext(item); } + // Movable, not copyable. - VmaIntrusiveLinkedList() { } - VmaIntrusiveLinkedList(const VmaIntrusiveLinkedList& src) = delete; - VmaIntrusiveLinkedList(VmaIntrusiveLinkedList&& src) : - m_Front(src.m_Front), m_Back(src.m_Back), m_Count(src.m_Count) - { - src.m_Front = src.m_Back = VMA_NULL; - src.m_Count = 0; - } - ~VmaIntrusiveLinkedList() - { - VMA_HEAVY_ASSERT(IsEmpty()); - } - VmaIntrusiveLinkedList& operator=(const VmaIntrusiveLinkedList& src) = delete; - VmaIntrusiveLinkedList& operator=(VmaIntrusiveLinkedList&& src) - { - if(&src != this) - { - VMA_HEAVY_ASSERT(IsEmpty()); - m_Front = src.m_Front; - m_Back = src.m_Back; - m_Count = src.m_Count; - src.m_Front = src.m_Back = VMA_NULL; - src.m_Count = 0; - } - return *this; - } - void RemoveAll() - { - if(!IsEmpty()) - { - ItemType* item = m_Back; - while(item != VMA_NULL) - { - ItemType* const prevItem = ItemTypeTraits::AccessPrev(item); - ItemTypeTraits::AccessPrev(item) = VMA_NULL; - ItemTypeTraits::AccessNext(item) = VMA_NULL; - item = prevItem; - } - m_Front = VMA_NULL; - m_Back = VMA_NULL; - m_Count = 0; - } - } + VmaIntrusiveLinkedList() = default; + VmaIntrusiveLinkedList(VmaIntrusiveLinkedList && src); + VmaIntrusiveLinkedList(const VmaIntrusiveLinkedList&) = delete; + VmaIntrusiveLinkedList& operator=(VmaIntrusiveLinkedList&& src); + VmaIntrusiveLinkedList& operator=(const VmaIntrusiveLinkedList&) = delete; + ~VmaIntrusiveLinkedList() { VMA_HEAVY_ASSERT(IsEmpty()); } + size_t GetCount() const { return m_Count; } bool IsEmpty() const { return m_Count == 0; } ItemType* Front() { return m_Front; } - const ItemType* Front() const { return m_Front; } ItemType* Back() { return m_Back; } + const ItemType* Front() const { return m_Front; } const ItemType* Back() const { return m_Back; } - void PushBack(ItemType* item) - { - VMA_HEAVY_ASSERT(ItemTypeTraits::GetPrev(item) == VMA_NULL && ItemTypeTraits::GetNext(item) == VMA_NULL); - if(IsEmpty()) - { - m_Front = item; - m_Back = item; - m_Count = 1; - } - else - { - ItemTypeTraits::AccessPrev(item) = m_Back; - ItemTypeTraits::AccessNext(m_Back) = item; - m_Back = item; - ++m_Count; - } - } - void PushFront(ItemType* item) - { - VMA_HEAVY_ASSERT(ItemTypeTraits::GetPrev(item) == VMA_NULL && ItemTypeTraits::GetNext(item) == VMA_NULL); - if(IsEmpty()) - { - m_Front = item; - m_Back = item; - m_Count = 1; - } - else - { - ItemTypeTraits::AccessNext(item) = m_Front; - ItemTypeTraits::AccessPrev(m_Front) = item; - m_Front = item; - ++m_Count; - } - } - ItemType* PopBack() - { - VMA_HEAVY_ASSERT(m_Count > 0); - ItemType* const backItem = m_Back; - ItemType* const prevItem = ItemTypeTraits::GetPrev(backItem); - if(prevItem != VMA_NULL) - { - ItemTypeTraits::AccessNext(prevItem) = VMA_NULL; - } - m_Back = prevItem; - --m_Count; - ItemTypeTraits::AccessPrev(backItem) = VMA_NULL; - ItemTypeTraits::AccessNext(backItem) = VMA_NULL; - return backItem; - } - ItemType* PopFront() - { - VMA_HEAVY_ASSERT(m_Count > 0); - ItemType* const frontItem = m_Front; - ItemType* const nextItem = ItemTypeTraits::GetNext(frontItem); - if(nextItem != VMA_NULL) - { - ItemTypeTraits::AccessPrev(nextItem) = VMA_NULL; - } - m_Front = nextItem; - --m_Count; - ItemTypeTraits::AccessPrev(frontItem) = VMA_NULL; - ItemTypeTraits::AccessNext(frontItem) = VMA_NULL; - return frontItem; - } + + void PushBack(ItemType* item); + void PushFront(ItemType* item); + ItemType* PopBack(); + ItemType* PopFront(); // MyItem can be null - it means PushBack. - void InsertBefore(ItemType* existingItem, ItemType* newItem) - { - VMA_HEAVY_ASSERT(newItem != VMA_NULL && ItemTypeTraits::GetPrev(newItem) == VMA_NULL && ItemTypeTraits::GetNext(newItem) == VMA_NULL); - if(existingItem != VMA_NULL) - { - ItemType* const prevItem = ItemTypeTraits::GetPrev(existingItem); - ItemTypeTraits::AccessPrev(newItem) = prevItem; - ItemTypeTraits::AccessNext(newItem) = existingItem; - ItemTypeTraits::AccessPrev(existingItem) = newItem; - if(prevItem != VMA_NULL) - { - ItemTypeTraits::AccessNext(prevItem) = newItem; - } - else - { - VMA_HEAVY_ASSERT(m_Front == existingItem); - m_Front = newItem; - } - ++m_Count; - } - else - PushBack(newItem); - } + void InsertBefore(ItemType* existingItem, ItemType* newItem); // MyItem can be null - it means PushFront. - void InsertAfter(ItemType* existingItem, ItemType* newItem) - { - VMA_HEAVY_ASSERT(newItem != VMA_NULL && ItemTypeTraits::GetPrev(newItem) == VMA_NULL && ItemTypeTraits::GetNext(newItem) == VMA_NULL); - if(existingItem != VMA_NULL) - { - ItemType* const nextItem = ItemTypeTraits::GetNext(existingItem); - ItemTypeTraits::AccessNext(newItem) = nextItem; - ItemTypeTraits::AccessPrev(newItem) = existingItem; - ItemTypeTraits::AccessNext(existingItem) = newItem; - if(nextItem != VMA_NULL) - { - ItemTypeTraits::AccessPrev(nextItem) = newItem; - } - else - { - VMA_HEAVY_ASSERT(m_Back == existingItem); - m_Back = newItem; - } - ++m_Count; - } - else - return PushFront(newItem); - } - void Remove(ItemType* item) - { - VMA_HEAVY_ASSERT(item != VMA_NULL && m_Count > 0); - if(ItemTypeTraits::GetPrev(item) != VMA_NULL) - { - ItemTypeTraits::AccessNext(ItemTypeTraits::AccessPrev(item)) = ItemTypeTraits::GetNext(item); - } - else - { - VMA_HEAVY_ASSERT(m_Front == item); - m_Front = ItemTypeTraits::GetNext(item); - } + void InsertAfter(ItemType* existingItem, ItemType* newItem); + void Remove(ItemType* item); + void RemoveAll(); - if(ItemTypeTraits::GetNext(item) != VMA_NULL) - { - ItemTypeTraits::AccessPrev(ItemTypeTraits::AccessNext(item)) = ItemTypeTraits::GetPrev(item); - } - else - { - VMA_HEAVY_ASSERT(m_Back == item); - m_Back = ItemTypeTraits::GetPrev(item); - } - ItemTypeTraits::AccessPrev(item) = VMA_NULL; - ItemTypeTraits::AccessNext(item) = VMA_NULL; - --m_Count; - } private: ItemType* m_Front = VMA_NULL; ItemType* m_Back = VMA_NULL; size_t m_Count = 0; }; -//////////////////////////////////////////////////////////////////////////////// -// class VmaMap +#ifndef _VMA_INTRUSIVE_LINKED_LIST_FUNCTIONS +template +VmaIntrusiveLinkedList::VmaIntrusiveLinkedList(VmaIntrusiveLinkedList&& src) + : m_Front(src.m_Front), m_Back(src.m_Back), m_Count(src.m_Count) +{ + src.m_Front = src.m_Back = VMA_NULL; + src.m_Count = 0; +} + +template +VmaIntrusiveLinkedList& VmaIntrusiveLinkedList::operator=(VmaIntrusiveLinkedList&& src) +{ + if (&src != this) + { + VMA_HEAVY_ASSERT(IsEmpty()); + m_Front = src.m_Front; + m_Back = src.m_Back; + m_Count = src.m_Count; + src.m_Front = src.m_Back = VMA_NULL; + src.m_Count = 0; + } + return *this; +} + +template +void VmaIntrusiveLinkedList::PushBack(ItemType* item) +{ + VMA_HEAVY_ASSERT(ItemTypeTraits::GetPrev(item) == VMA_NULL && ItemTypeTraits::GetNext(item) == VMA_NULL); + if (IsEmpty()) + { + m_Front = item; + m_Back = item; + m_Count = 1; + } + else + { + ItemTypeTraits::AccessPrev(item) = m_Back; + ItemTypeTraits::AccessNext(m_Back) = item; + m_Back = item; + ++m_Count; + } +} + +template +void VmaIntrusiveLinkedList::PushFront(ItemType* item) +{ + VMA_HEAVY_ASSERT(ItemTypeTraits::GetPrev(item) == VMA_NULL && ItemTypeTraits::GetNext(item) == VMA_NULL); + if (IsEmpty()) + { + m_Front = item; + m_Back = item; + m_Count = 1; + } + else + { + ItemTypeTraits::AccessNext(item) = m_Front; + ItemTypeTraits::AccessPrev(m_Front) = item; + m_Front = item; + ++m_Count; + } +} + +template +typename VmaIntrusiveLinkedList::ItemType* VmaIntrusiveLinkedList::PopBack() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const backItem = m_Back; + ItemType* const prevItem = ItemTypeTraits::GetPrev(backItem); + if (prevItem != VMA_NULL) + { + ItemTypeTraits::AccessNext(prevItem) = VMA_NULL; + } + m_Back = prevItem; + --m_Count; + ItemTypeTraits::AccessPrev(backItem) = VMA_NULL; + ItemTypeTraits::AccessNext(backItem) = VMA_NULL; + return backItem; +} + +template +typename VmaIntrusiveLinkedList::ItemType* VmaIntrusiveLinkedList::PopFront() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const frontItem = m_Front; + ItemType* const nextItem = ItemTypeTraits::GetNext(frontItem); + if (nextItem != VMA_NULL) + { + ItemTypeTraits::AccessPrev(nextItem) = VMA_NULL; + } + m_Front = nextItem; + --m_Count; + ItemTypeTraits::AccessPrev(frontItem) = VMA_NULL; + ItemTypeTraits::AccessNext(frontItem) = VMA_NULL; + return frontItem; +} + +template +void VmaIntrusiveLinkedList::InsertBefore(ItemType* existingItem, ItemType* newItem) +{ + VMA_HEAVY_ASSERT(newItem != VMA_NULL && ItemTypeTraits::GetPrev(newItem) == VMA_NULL && ItemTypeTraits::GetNext(newItem) == VMA_NULL); + if (existingItem != VMA_NULL) + { + ItemType* const prevItem = ItemTypeTraits::GetPrev(existingItem); + ItemTypeTraits::AccessPrev(newItem) = prevItem; + ItemTypeTraits::AccessNext(newItem) = existingItem; + ItemTypeTraits::AccessPrev(existingItem) = newItem; + if (prevItem != VMA_NULL) + { + ItemTypeTraits::AccessNext(prevItem) = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_Front == existingItem); + m_Front = newItem; + } + ++m_Count; + } + else + PushBack(newItem); +} + +template +void VmaIntrusiveLinkedList::InsertAfter(ItemType* existingItem, ItemType* newItem) +{ + VMA_HEAVY_ASSERT(newItem != VMA_NULL && ItemTypeTraits::GetPrev(newItem) == VMA_NULL && ItemTypeTraits::GetNext(newItem) == VMA_NULL); + if (existingItem != VMA_NULL) + { + ItemType* const nextItem = ItemTypeTraits::GetNext(existingItem); + ItemTypeTraits::AccessNext(newItem) = nextItem; + ItemTypeTraits::AccessPrev(newItem) = existingItem; + ItemTypeTraits::AccessNext(existingItem) = newItem; + if (nextItem != VMA_NULL) + { + ItemTypeTraits::AccessPrev(nextItem) = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_Back == existingItem); + m_Back = newItem; + } + ++m_Count; + } + else + return PushFront(newItem); +} + +template +void VmaIntrusiveLinkedList::Remove(ItemType* item) +{ + VMA_HEAVY_ASSERT(item != VMA_NULL && m_Count > 0); + if (ItemTypeTraits::GetPrev(item) != VMA_NULL) + { + ItemTypeTraits::AccessNext(ItemTypeTraits::AccessPrev(item)) = ItemTypeTraits::GetNext(item); + } + else + { + VMA_HEAVY_ASSERT(m_Front == item); + m_Front = ItemTypeTraits::GetNext(item); + } + + if (ItemTypeTraits::GetNext(item) != VMA_NULL) + { + ItemTypeTraits::AccessPrev(ItemTypeTraits::AccessNext(item)) = ItemTypeTraits::GetPrev(item); + } + else + { + VMA_HEAVY_ASSERT(m_Back == item); + m_Back = ItemTypeTraits::GetPrev(item); + } + ItemTypeTraits::AccessPrev(item) = VMA_NULL; + ItemTypeTraits::AccessNext(item) = VMA_NULL; + --m_Count; +} + +template +void VmaIntrusiveLinkedList::RemoveAll() +{ + if (!IsEmpty()) + { + ItemType* item = m_Back; + while (item != VMA_NULL) + { + ItemType* const prevItem = ItemTypeTraits::AccessPrev(item); + ItemTypeTraits::AccessPrev(item) = VMA_NULL; + ItemTypeTraits::AccessNext(item) = VMA_NULL; + item = prevItem; + } + m_Front = VMA_NULL; + m_Back = VMA_NULL; + m_Count = 0; + } +} +#endif // _VMA_INTRUSIVE_LINKED_LIST_FUNCTIONS +#endif // _VMA_INTRUSIVE_LINKED_LIST // 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, std::equal_to, VmaStlAllocator< std::pair > > - -#else // #if VMA_USE_STL_UNORDERED_MAP - +#ifndef _VMA_PAIR template struct VmaPair { T1 first; T2 second; - VmaPair() : first(), second() { } - VmaPair(const T1& firstSrc, const T2& secondSrc) : first(firstSrc), second(secondSrc) { } + 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 -class VmaMap -{ -public: - typedef VmaPair PairType; - typedef PairType* iterator; - - VmaMap(const VmaStlAllocator& 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 > m_Vector; -}; - -#define VMA_MAP_TYPE(KeyT, ValueT) VmaMap - template struct VmaPairFirstLess { @@ -4481,7 +5269,34 @@ struct VmaPairFirstLess return lhs.first < rhsFirst; } }; +#endif // _VMA_PAIR +#ifndef _VMA_MAP +/* Class compatible with subset of interface of std::unordered_map. +KeyT, ValueT must be POD because they will be stored in VmaVector. +*/ +template +class VmaMap +{ +public: + typedef VmaPair PairType; + typedef PairType* iterator; + + VmaMap(const VmaStlAllocator& allocator) : m_Vector(allocator) {} + + iterator begin() { return m_Vector.begin(); } + iterator end() { return m_Vector.end(); } + size_t size() { return m_Vector.size(); } + + void insert(const PairType& pair); + iterator find(const KeyT& key); + void erase(iterator it); + +private: + VmaVector< PairType, VmaStlAllocator> m_Vector; +}; + +#ifndef _VMA_MAP_FUNCTIONS template void VmaMap::insert(const PairType& pair) { @@ -4501,7 +5316,7 @@ VmaPair* VmaMap::find(const KeyT& key) m_Vector.data() + m_Vector.size(), key, VmaPairFirstLess()); - if((it != m_Vector.end()) && (it->first == key)) + if ((it != m_Vector.end()) && (it->first == key)) { return it; } @@ -4516,25 +5331,639 @@ void VmaMap::erase(iterator it) { VmaVectorRemove(m_Vector, it - m_Vector.begin()); } - -#endif // #if VMA_USE_STL_UNORDERED_MAP +#endif // _VMA_MAP_FUNCTIONS +#endif // _VMA_MAP #endif // #if 0 -//////////////////////////////////////////////////////////////////////////////// +#if !defined(_VMA_STRING_BUILDER) && VMA_STATS_STRING_ENABLED +class VmaStringBuilder +{ +public: + VmaStringBuilder(const VkAllocationCallbacks* allocationCallbacks) : m_Data(VmaStlAllocator(allocationCallbacks)) {} + ~VmaStringBuilder() = default; -class VmaDeviceMemoryBlock; + size_t GetLength() const { return m_Data.size(); } + const char* GetData() const { return m_Data.data(); } + void AddNewLine() { Add('\n'); } + void Add(char ch) { m_Data.push_back(ch); } -enum VMA_CACHE_OPERATION { VMA_CACHE_FLUSH, VMA_CACHE_INVALIDATE }; + void Add(const char* pStr); + void AddNumber(uint32_t num); + void AddNumber(uint64_t num); + void AddPointer(const void* ptr); +private: + VmaVector> m_Data; +}; + +#ifndef _VMA_STRING_BUILDER_FUNCTIONS +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]; + buf[10] = '\0'; + char* p = &buf[10]; + do + { + *--p = '0' + (num % 10); + num /= 10; + } while (num); + Add(p); +} + +void VmaStringBuilder::AddNumber(uint64_t num) +{ + char buf[21]; + buf[20] = '\0'; + char* p = &buf[20]; + do + { + *--p = '0' + (num % 10); + num /= 10; + } while (num); + Add(p); +} + +void VmaStringBuilder::AddPointer(const void* ptr) +{ + char buf[21]; + VmaPtrToStr(buf, sizeof(buf), ptr); + Add(buf); +} +#endif //_VMA_STRING_BUILDER_FUNCTIONS +#endif // _VMA_STRING_BUILDER + +#if !defined(_VMA_JSON_WRITER) && VMA_STATS_STRING_ENABLED +/* +Allows to conveniently build a correct JSON document to be written to the +VmaStringBuilder passed to the constructor. +*/ +class VmaJsonWriter +{ + VMA_CLASS_NO_COPY(VmaJsonWriter) +public: + // sb - string builder to write the document to. Must remain alive for the whole lifetime of this object. + VmaJsonWriter(const VkAllocationCallbacks* pAllocationCallbacks, VmaStringBuilder& sb); + ~VmaJsonWriter(); + + // Begins object by writing "{". + // Inside an object, you must call pairs of WriteString and a value, e.g.: + // j.BeginObject(true); j.WriteString("A"); j.WriteNumber(1); j.WriteString("B"); j.WriteNumber(2); j.EndObject(); + // Will write: { "A": 1, "B": 2 } + void BeginObject(bool singleLine = false); + // Ends object by writing "}". + void EndObject(); + + // Begins array by writing "[". + // Inside an array, you can write a sequence of any values. + void BeginArray(bool singleLine = false); + // Ends array by writing "[". + void EndArray(); + + // Writes a string value inside "". + // pStr can contain any ANSI characters, including '"', new line etc. - they will be properly escaped. + void WriteString(const char* pStr); + + // Begins writing a string value. + // Call BeginString, ContinueString, ContinueString, ..., EndString instead of + // WriteString to conveniently build the string content incrementally, made of + // parts including numbers. + void BeginString(const char* pStr = VMA_NULL); + // Posts next part of an open string. + void ContinueString(const char* pStr); + // Posts next part of an open string. The number is converted to decimal characters. + void ContinueString(uint32_t n); + void ContinueString(uint64_t n); + void ContinueString_Size(size_t n); + // Posts next part of an open string. Pointer value is converted to characters + // using "%p" formatting - shown as hexadecimal number, e.g.: 000000081276Ad00 + void ContinueString_Pointer(const void* ptr); + // Ends writing a string value by writing '"'. + void EndString(const char* pStr = VMA_NULL); + + // Writes a number value. + void WriteNumber(uint32_t n); + void WriteNumber(uint64_t n); + void WriteSize(size_t n); + // Writes a boolean value - false or true. + void WriteBool(bool b); + // Writes a null value. + void WriteNull(); + +private: + enum COLLECTION_TYPE + { + COLLECTION_TYPE_OBJECT, + COLLECTION_TYPE_ARRAY, + }; + struct StackItem + { + COLLECTION_TYPE type; + uint32_t valueCount; + bool singleLineMode; + }; + + static const char* const INDENT; + + VmaStringBuilder& m_SB; + VmaVector< StackItem, VmaStlAllocator > m_Stack; + bool m_InsideString; + + // Write size_t for less than 64bits + void WriteSize(size_t n, std::integral_constant) { m_SB.AddNumber(static_cast(n)); } + // Write size_t for 64bits + void WriteSize(size_t n, std::integral_constant) { m_SB.AddNumber(static_cast(n)); } + + void BeginValue(bool isString); + void WriteIndent(bool oneLess = false); +}; +const char* const VmaJsonWriter::INDENT = " "; + +#ifndef _VMA_JSON_WRITER_FUNCTIONS +VmaJsonWriter::VmaJsonWriter(const VkAllocationCallbacks* pAllocationCallbacks, VmaStringBuilder& sb) + : m_SB(sb), + m_Stack(VmaStlAllocator(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_Size(size_t n) +{ + VMA_ASSERT(m_InsideString); + // Fix for AppleClang incorrect type casting + // TODO: Change to if constexpr when C++17 used as minimal standard + WriteSize(n, std::is_same{}); +} + +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::WriteSize(size_t n) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + // Fix for AppleClang incorrect type casting + // TODO: Change to if constexpr when C++17 used as minimal standard + WriteSize(n, std::is_same{}); +} + +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) + { + 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 // _VMA_JSON_WRITER_FUNCTIONS + +static void VmaPrintDetailedStatistics(VmaJsonWriter& json, const VmaDetailedStatistics& stat) +{ + json.BeginObject(); + + json.WriteString("BlockCount"); + json.WriteNumber(stat.statistics.blockCount); + json.WriteString("BlockBytes"); + json.WriteNumber(stat.statistics.blockBytes); + json.WriteString("AllocationCount"); + json.WriteNumber(stat.statistics.allocationCount); + json.WriteString("AllocationBytes"); + json.WriteNumber(stat.statistics.allocationBytes); + json.WriteString("UnusedRangeCount"); + json.WriteNumber(stat.unusedRangeCount); + + if (stat.statistics.allocationCount > 1) + { + json.WriteString("AllocationSizeMin"); + json.WriteNumber(stat.allocationSizeMin); + json.WriteString("AllocationSizeMax"); + json.WriteNumber(stat.allocationSizeMax); + } + if (stat.unusedRangeCount > 1) + { + json.WriteString("UnusedRangeSizeMin"); + json.WriteNumber(stat.unusedRangeSizeMin); + json.WriteString("UnusedRangeSizeMax"); + json.WriteNumber(stat.unusedRangeSizeMax); + } + json.EndObject(); +} +#endif // _VMA_JSON_WRITER + +#ifndef _VMA_MAPPING_HYSTERESIS + +class VmaMappingHysteresis +{ + VMA_CLASS_NO_COPY(VmaMappingHysteresis) +public: + VmaMappingHysteresis() = default; + + uint32_t GetExtraMapping() const { return m_ExtraMapping; } + + // Call when Map was called. + // Returns true if switched to extra +1 mapping reference count. + bool PostMap() + { +#if VMA_MAPPING_HYSTERESIS_ENABLED + if(m_ExtraMapping == 0) + { + ++m_MajorCounter; + if(m_MajorCounter >= COUNTER_MIN_EXTRA_MAPPING) + { + m_ExtraMapping = 1; + m_MajorCounter = 0; + m_MinorCounter = 0; + return true; + } + } + else // m_ExtraMapping == 1 + PostMinorCounter(); +#endif // #if VMA_MAPPING_HYSTERESIS_ENABLED + return false; + } + + // Call when Unmap was called. + void PostUnmap() + { +#if VMA_MAPPING_HYSTERESIS_ENABLED + if(m_ExtraMapping == 0) + ++m_MajorCounter; + else // m_ExtraMapping == 1 + PostMinorCounter(); +#endif // #if VMA_MAPPING_HYSTERESIS_ENABLED + } + + // Call when allocation was made from the memory block. + void PostAlloc() + { +#if VMA_MAPPING_HYSTERESIS_ENABLED + if(m_ExtraMapping == 1) + ++m_MajorCounter; + else // m_ExtraMapping == 0 + PostMinorCounter(); +#endif // #if VMA_MAPPING_HYSTERESIS_ENABLED + } + + // Call when allocation was freed from the memory block. + // Returns true if switched to extra -1 mapping reference count. + bool PostFree() + { +#if VMA_MAPPING_HYSTERESIS_ENABLED + if(m_ExtraMapping == 1) + { + ++m_MajorCounter; + if(m_MajorCounter >= COUNTER_MIN_EXTRA_MAPPING && + m_MajorCounter > m_MinorCounter + 1) + { + m_ExtraMapping = 0; + m_MajorCounter = 0; + m_MinorCounter = 0; + return true; + } + } + else // m_ExtraMapping == 0 + PostMinorCounter(); +#endif // #if VMA_MAPPING_HYSTERESIS_ENABLED + return false; + } + +private: + static const int32_t COUNTER_MIN_EXTRA_MAPPING = 7; + + uint32_t m_MinorCounter = 0; + uint32_t m_MajorCounter = 0; + uint32_t m_ExtraMapping = 0; // 0 or 1. + + void PostMinorCounter() + { + if(m_MinorCounter < m_MajorCounter) + { + ++m_MinorCounter; + } + else if(m_MajorCounter > 0) + { + --m_MajorCounter; + --m_MinorCounter; + } + } +}; + +#endif // _VMA_MAPPING_HYSTERESIS + +#ifndef _VMA_DEVICE_MEMORY_BLOCK +/* +Represents a single block of device memory (`VkDeviceMemory`) with all the +data about its regions (aka suballocations, #VmaAllocation), assigned and free. + +Thread-safety: +- Access to m_pMetadata must be externally synchronized. +- Map, Unmap, Bind* are synchronized internally. +*/ +class VmaDeviceMemoryBlock +{ + VMA_CLASS_NO_COPY(VmaDeviceMemoryBlock) +public: + VmaBlockMetadata* m_pMetadata; + + VmaDeviceMemoryBlock(VmaAllocator hAllocator); + ~VmaDeviceMemoryBlock(); + + // Always call after construction. + void Init( + VmaAllocator hAllocator, + VmaPool hParentPool, + uint32_t newMemoryTypeIndex, + VkDeviceMemory newMemory, + VkDeviceSize newSize, + uint32_t id, + uint32_t algorithm, + VkDeviceSize bufferImageGranularity); + // Always call before destruction. + void Destroy(VmaAllocator allocator); + + VmaPool GetParentPool() const { return m_hParentPool; } + 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; } + uint32_t GetMapRefCount() const { return m_MapCount; } + + // Call when allocation/free was made from m_pMetadata. + // Used for m_MappingHysteresis. + void PostAlloc() { m_MappingHysteresis.PostAlloc(); } + void PostFree(VmaAllocator hAllocator); + + // 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 WriteMagicValueAfterAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); + VkResult ValidateMagicValueAfterAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); + + VkResult BindBufferMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkBuffer hBuffer, + const void* pNext); + VkResult BindImageMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkImage hImage, + const void* pNext); + +private: + VmaPool m_hParentPool; // VK_NULL_HANDLE if not belongs to custom pool. + uint32_t m_MemoryTypeIndex; + uint32_t m_Id; + VkDeviceMemory m_hMemory; + + /* + Protects access to m_hMemory so it is 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_MapAndBindMutex; + VmaMappingHysteresis m_MappingHysteresis; + uint32_t m_MapCount; + void* m_pMappedData; +}; +#endif // _VMA_DEVICE_MEMORY_BLOCK + +#ifndef _VMA_ALLOCATION_T struct VmaAllocation_T { -private: - static const uint8_t MAP_COUNT_FLAG_PERSISTENT_MAP = 0x80; + friend struct VmaDedicatedAllocationListItemTraits; enum FLAGS { - FLAG_USER_DATA_STRING = 0x01, + FLAG_PERSISTENT_MAP = 0x01, + FLAG_MAPPING_ALLOWED = 0x02, }; public: @@ -4545,148 +5974,48 @@ public: ALLOCATION_TYPE_DEDICATED, }; - /* - This struct is allocated using VmaPoolAllocator. - */ - - VmaAllocation_T(uint32_t currentFrameIndex, bool userDataString) : - m_Alignment{1}, - m_Size{0}, - m_pUserData{VMA_NULL}, - m_LastUseFrameIndex{currentFrameIndex}, - m_MemoryTypeIndex{0}, - 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 : (uint8_t)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); - } + // This struct is allocated using VmaPoolAllocator. + VmaAllocation_T(bool mappingAllowed); + ~VmaAllocation_T(); void InitBlockAllocation( VmaDeviceMemoryBlock* block, - VkDeviceSize offset, + VmaAllocHandle allocHandle, VkDeviceSize alignment, VkDeviceSize size, uint32_t memoryTypeIndex, 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_MemoryTypeIndex = memoryTypeIndex; - m_MapCount = mapped ? MAP_COUNT_FLAG_PERSISTENT_MAP : 0; - m_SuballocationType = (uint8_t)suballocationType; - 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_MemoryTypeIndex = 0; - 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 ChangeOffset(VkDeviceSize newOffset); - + bool mapped); // pMappedData not null means allocation is created with MAPPED flag. void InitDedicatedAllocation( + VmaPool hParentPool, 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_MemoryTypeIndex = memoryTypeIndex; - m_SuballocationType = (uint8_t)suballocationType; - m_MapCount = (pMappedData != VMA_NULL) ? MAP_COUNT_FLAG_PERSISTENT_MAP : 0; - m_DedicatedAllocation.m_hMemory = hMemory; - m_DedicatedAllocation.m_pMappedData = pMappedData; - m_DedicatedAllocation.m_Prev = VMA_NULL; - m_DedicatedAllocation.m_Next = VMA_NULL; - } + VkDeviceSize size); 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); + const char* GetName() const { return m_pName; } 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; + VmaDeviceMemoryBlock* GetBlock() const { VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); return m_BlockAllocation.m_Block; } uint32_t GetMemoryTypeIndex() const { return m_MemoryTypeIndex; } - bool IsPersistentMap() const { return (m_MapCount & MAP_COUNT_FLAG_PERSISTENT_MAP) != 0; } + bool IsPersistentMap() const { return (m_Flags & FLAG_PERSISTENT_MAP) != 0; } + bool IsMappingAllowed() const { return (m_Flags & FLAG_MAPPING_ALLOWED) != 0; } + + void SetUserData(VmaAllocator hAllocator, void* pUserData) { m_pUserData = pUserData; } + void SetName(VmaAllocator hAllocator, const char* pName); + void FreeName(VmaAllocator hAllocator); + uint8_t SwapBlockAllocation(VmaAllocator hAllocator, VmaAllocation allocation); + VmaAllocHandle GetAllocHandle() const; + VkDeviceSize GetOffset() const; + VmaPool GetParentPool() const; + VkDeviceMemory GetMemory() const; void* GetMappedData() const; - bool CanBecomeLost() 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(); @@ -4694,48 +6023,28 @@ public: 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 InitBufferImageUsage(uint32_t bufferImageUsage); void PrintParameters(class VmaJsonWriter& json) const; #endif private: - VkDeviceSize m_Alignment; - VkDeviceSize m_Size; - void* m_pUserData; - VMA_ATOMIC_UINT32 m_LastUseFrameIndex; - uint32_t m_MemoryTypeIndex; - 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 { VmaDeviceMemoryBlock* m_Block; - VkDeviceSize m_Offset; - bool m_CanBecomeLost; + VmaAllocHandle m_AllocHandle; }; - // Allocation for an object that has its own private VkDeviceMemory. struct DedicatedAllocation { + VmaPool m_hParentPool; // VK_NULL_HANDLE if not belongs to custom pool. VkDeviceMemory m_hMemory; void* m_pMappedData; // Not null means memory is mapped. VmaAllocation_T* m_Prev; VmaAllocation_T* m_Next; }; - union { // Allocation out of VmaDeviceMemoryBlock. @@ -4744,19 +6053,27 @@ private: DedicatedAllocation m_DedicatedAllocation; }; + VkDeviceSize m_Alignment; + VkDeviceSize m_Size; + void* m_pUserData; + char* m_pName; + uint32_t m_MemoryTypeIndex; + uint8_t m_Type; // ALLOCATION_TYPE + uint8_t m_SuballocationType; // VmaSuballocationType + // Reference counter for vmaMapMemory()/vmaUnmapMemory(). + uint8_t m_MapCount; + uint8_t m_Flags; // enum FLAGS #if VMA_STATS_STRING_ENABLED - uint32_t m_CreationFrameIndex; uint32_t m_BufferImageUsage; // 0 if unknown. #endif - - void FreeUserDataString(VmaAllocator hAllocator); - - friend struct VmaDedicatedAllocationListItemTraits; }; +#endif // _VMA_ALLOCATION_T +#ifndef _VMA_DEDICATED_ALLOCATION_LIST_ITEM_TRAITS struct VmaDedicatedAllocationListItemTraits { typedef VmaAllocation_T ItemType; + static ItemType* GetPrev(const ItemType* item) { VMA_HEAVY_ASSERT(item->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); @@ -4772,12 +6089,138 @@ struct VmaDedicatedAllocationListItemTraits VMA_HEAVY_ASSERT(item->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); return item->m_DedicatedAllocation.m_Prev; } - static ItemType*& AccessNext(ItemType* item){ + static ItemType*& AccessNext(ItemType* item) + { VMA_HEAVY_ASSERT(item->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); return item->m_DedicatedAllocation.m_Next; } }; +#endif // _VMA_DEDICATED_ALLOCATION_LIST_ITEM_TRAITS +#ifndef _VMA_DEDICATED_ALLOCATION_LIST +/* +Stores linked list of VmaAllocation_T objects. +Thread-safe, synchronized internally. +*/ +class VmaDedicatedAllocationList +{ +public: + VmaDedicatedAllocationList() {} + ~VmaDedicatedAllocationList(); + + void Init(bool useMutex) { m_UseMutex = useMutex; } + bool Validate(); + + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats); + void AddStatistics(VmaStatistics& inoutStats); +#if VMA_STATS_STRING_ENABLED + // Writes JSON array with the list of allocations. + void BuildStatsString(VmaJsonWriter& json); +#endif + + bool IsEmpty(); + void Register(VmaAllocation alloc); + void Unregister(VmaAllocation alloc); + +private: + typedef VmaIntrusiveLinkedList DedicatedAllocationLinkedList; + + bool m_UseMutex = true; + VMA_RW_MUTEX m_Mutex; + DedicatedAllocationLinkedList m_AllocationList; +}; + +#ifndef _VMA_DEDICATED_ALLOCATION_LIST_FUNCTIONS + +VmaDedicatedAllocationList::~VmaDedicatedAllocationList() +{ + VMA_HEAVY_ASSERT(Validate()); + + if (!m_AllocationList.IsEmpty()) + { + VMA_ASSERT(false && "Unfreed dedicated allocations found!"); + } +} + +bool VmaDedicatedAllocationList::Validate() +{ + const size_t declaredCount = m_AllocationList.GetCount(); + size_t actualCount = 0; + VmaMutexLockRead lock(m_Mutex, m_UseMutex); + for (VmaAllocation alloc = m_AllocationList.Front(); + alloc != VMA_NULL; alloc = m_AllocationList.GetNext(alloc)) + { + ++actualCount; + } + VMA_VALIDATE(actualCount == declaredCount); + + return true; +} + +void VmaDedicatedAllocationList::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) +{ + for(auto* item = m_AllocationList.Front(); item != nullptr; item = DedicatedAllocationLinkedList::GetNext(item)) + { + const VkDeviceSize size = item->GetSize(); + inoutStats.statistics.blockCount++; + inoutStats.statistics.blockBytes += size; + VmaAddDetailedStatisticsAllocation(inoutStats, item->GetSize()); + } +} + +void VmaDedicatedAllocationList::AddStatistics(VmaStatistics& inoutStats) +{ + VmaMutexLockRead lock(m_Mutex, m_UseMutex); + + const uint32_t allocCount = (uint32_t)m_AllocationList.GetCount(); + inoutStats.blockCount += allocCount; + inoutStats.allocationCount += allocCount; + + for(auto* item = m_AllocationList.Front(); item != nullptr; item = DedicatedAllocationLinkedList::GetNext(item)) + { + const VkDeviceSize size = item->GetSize(); + inoutStats.blockBytes += size; + inoutStats.allocationBytes += size; + } +} + +#if VMA_STATS_STRING_ENABLED +void VmaDedicatedAllocationList::BuildStatsString(VmaJsonWriter& json) +{ + VmaMutexLockRead lock(m_Mutex, m_UseMutex); + json.BeginArray(); + for (VmaAllocation alloc = m_AllocationList.Front(); + alloc != VMA_NULL; alloc = m_AllocationList.GetNext(alloc)) + { + json.BeginObject(true); + alloc->PrintParameters(json); + json.EndObject(); + } + json.EndArray(); +} +#endif // VMA_STATS_STRING_ENABLED + +bool VmaDedicatedAllocationList::IsEmpty() +{ + VmaMutexLockRead lock(m_Mutex, m_UseMutex); + return m_AllocationList.IsEmpty(); +} + +void VmaDedicatedAllocationList::Register(VmaAllocation alloc) +{ + VmaMutexLockWrite lock(m_Mutex, m_UseMutex); + m_AllocationList.PushBack(alloc); +} + +void VmaDedicatedAllocationList::Unregister(VmaAllocation alloc) +{ + VmaMutexLockWrite lock(m_Mutex, m_UseMutex); + m_AllocationList.Remove(alloc); +} +#endif // _VMA_DEDICATED_ALLOCATION_LIST_FUNCTIONS +#endif // _VMA_DEDICATED_ALLOCATION_LIST + +#ifndef _VMA_SUBALLOCATION /* Represents a region of VmaDeviceMemoryBlock that is either assigned and returned as allocated memory block or free. @@ -4786,7 +6229,7 @@ struct VmaSuballocation { VkDeviceSize offset; VkDeviceSize size; - VmaAllocation hAllocation; + void* userData; VmaSuballocationType type; }; @@ -4798,6 +6241,7 @@ struct VmaSuballocationOffsetLess return lhs.offset < rhs.offset; } }; + struct VmaSuballocationOffsetGreater { bool operator()(const VmaSuballocation& lhs, const VmaSuballocation& rhs) const @@ -4806,49 +6250,39 @@ struct VmaSuballocationOffsetGreater } }; -typedef VmaList< VmaSuballocation, VmaStlAllocator > VmaSuballocationList; - -// Cost of one additional allocation lost, as equivalent in bytes. -static const VkDeviceSize VMA_LOST_ALLOCATION_COST = 1048576; - -enum class VmaAllocationRequestType +struct VmaSuballocationItemSizeLess { - Normal, - // Used by "Linear" algorithm. - UpperAddress, - EndOf1st, - EndOf2nd, -}; + 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; + } +}; +#endif // _VMA_SUBALLOCATION + +#ifndef _VMA_ALLOCATION_REQUEST /* 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. +item points to a FREE suballocation. */ 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. + VmaAllocHandle allocHandle; + VkDeviceSize size; VmaSuballocationList::iterator item; - size_t itemsToMakeLostCount; void* customData; + uint64_t algorithmData; VmaAllocationRequestType type; - - VkDeviceSize CalcCost() const - { - return sumItemSize + itemsToMakeLostCount * VMA_LOST_ALLOCATION_COST; - } }; +#endif // _VMA_ALLOCATION_REQUEST +#ifndef _VMA_BLOCK_METADATA /* Data structure used for bookkeeping of allocations and unused ranges of memory in a single VkDeviceMemory block. @@ -4856,22 +6290,33 @@ in a single VkDeviceMemory block. class VmaBlockMetadata { public: - VmaBlockMetadata(VmaAllocator hAllocator); - virtual ~VmaBlockMetadata() { } + // pAllocationCallbacks, if not null, must be owned externally - alive and unchanged for the whole lifetime of this object. + VmaBlockMetadata(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); + virtual ~VmaBlockMetadata() = default; + virtual void Init(VkDeviceSize size) { m_Size = size; } + bool IsVirtual() const { return m_IsVirtual; } + VkDeviceSize GetSize() const { return m_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 size_t GetFreeRegionsCount() 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 GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) = 0; + virtual VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const = 0; + virtual void* GetAllocationUserData(VmaAllocHandle allocHandle) const = 0; + + virtual VmaAllocHandle GetAllocationListBegin() const = 0; + virtual VmaAllocHandle GetNextAllocation(VmaAllocHandle prevAlloc) const = 0; + virtual VkDeviceSize GetNextFreeRegionSize(VmaAllocHandle alloc) const = 0; - virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const = 0; // Shouldn't modify blockCount. - virtual void AddPoolStats(VmaPoolStats& inoutStats) const = 0; + virtual void AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const = 0; + virtual void AddStatistics(VmaStatistics& inoutStats) const = 0; #if VMA_STATS_STRING_ENABLED virtual void PrintDetailedMap(class VmaJsonWriter& json) const = 0; @@ -4881,49 +6326,46 @@ public: // 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, - VmaAllocation hAllocation) = 0; + void* userData) = 0; // Frees suballocation assigned to given memory region. - virtual void Free(const VmaAllocation allocation) = 0; - virtual void FreeAtOffset(VkDeviceSize offset) = 0; + virtual void Free(VmaAllocHandle allocHandle) = 0; + + // Frees all allocations. + // Careful! Don't call it if there are VmaAllocation objects owned by userData of cleared allocations! + virtual void Clear() = 0; + + virtual void SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) = 0; + virtual void DebugLogAllAllocations() const = 0; protected: const VkAllocationCallbacks* GetAllocationCallbacks() const { return m_pAllocationCallbacks; } + VkDeviceSize GetBufferImageGranularity() const { return m_BufferImageGranularity; } + VkDeviceSize GetDebugMargin() const { return IsVirtual() ? 0 : VMA_DEBUG_MARGIN; } + void DebugLogAllocation(VkDeviceSize offset, VkDeviceSize size, void* userData) const; #if VMA_STATS_STRING_ENABLED + // mapRefCount == UINT32_MAX means unspecified. 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; + VkDeviceSize offset, VkDeviceSize size, void* userData) const; void PrintDetailedMap_UnusedRange(class VmaJsonWriter& json, VkDeviceSize offset, VkDeviceSize size) const; @@ -4933,99 +6375,426 @@ protected: private: VkDeviceSize m_Size; const VkAllocationCallbacks* m_pAllocationCallbacks; + const VkDeviceSize m_BufferImageGranularity; + const bool m_IsVirtual; }; -#define VMA_VALIDATE(cond) do { if(!(cond)) { \ - VMA_ASSERT(0 && "Validation failed: " #cond); \ - return false; \ - } } while(false) +#ifndef _VMA_BLOCK_METADATA_FUNCTIONS +VmaBlockMetadata::VmaBlockMetadata(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual) + : m_Size(0), + m_pAllocationCallbacks(pAllocationCallbacks), + m_BufferImageGranularity(bufferImageGranularity), + m_IsVirtual(isVirtual) {} -class VmaBlockMetadata_Generic : public VmaBlockMetadata +void VmaBlockMetadata::DebugLogAllocation(VkDeviceSize offset, VkDeviceSize size, void* userData) const { - VMA_CLASS_NO_COPY(VmaBlockMetadata_Generic) -public: - VmaBlockMetadata_Generic(VmaAllocator hAllocator); - virtual ~VmaBlockMetadata_Generic(); - virtual void Init(VkDeviceSize size); + if (IsVirtual()) + { + VMA_DEBUG_LOG("UNFREED VIRTUAL ALLOCATION; Offset: %llu; Size: %llu; UserData: %p", offset, size, userData); + } + else + { + VMA_ASSERT(userData != VMA_NULL); + VmaAllocation allocation = reinterpret_cast(userData); - 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; + userData = allocation->GetUserData(); + const char* name = allocation->GetName(); #if VMA_STATS_STRING_ENABLED - virtual void PrintDetailedMap(class VmaJsonWriter& json) const; + VMA_DEBUG_LOG("UNFREED ALLOCATION; Offset: %llu; Size: %llu; UserData: %p; Name: %s; Type: %s; Usage: %u", + offset, size, userData, name ? name : "vma_empty", + VMA_SUBALLOCATION_TYPE_NAMES[allocation->GetSuballocationType()], + allocation->GetBufferImageUsage()); +#else + VMA_DEBUG_LOG("UNFREED ALLOCATION; Offset: %llu; Size: %llu; UserData: %p; Name: %s; Type: %u", + offset, size, userData, name ? name : "vma_empty", + (uint32_t)allocation->GetSuballocationType()); +#endif // VMA_STATS_STRING_ENABLED + } + +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata::PrintDetailedMap_Begin(class VmaJsonWriter& json, + VkDeviceSize unusedBytes, size_t allocationCount, size_t unusedRangeCount) const +{ + json.WriteString("TotalBytes"); + json.WriteNumber(GetSize()); + + json.WriteString("UnusedBytes"); + json.WriteSize(unusedBytes); + + json.WriteString("Allocations"); + json.WriteSize(allocationCount); + + json.WriteString("UnusedRanges"); + json.WriteSize(unusedRangeCount); + + json.WriteString("Suballocations"); + json.BeginArray(); +} + +void VmaBlockMetadata::PrintDetailedMap_Allocation(class VmaJsonWriter& json, + VkDeviceSize offset, VkDeviceSize size, void* userData) const +{ + json.BeginObject(true); + + json.WriteString("Offset"); + json.WriteNumber(offset); + + if (IsVirtual()) + { + json.WriteString("Size"); + json.WriteNumber(size); + if (userData) + { + json.WriteString("CustomData"); + json.BeginString(); + json.ContinueString_Pointer(userData); + json.EndString(); + } + } + else + { + ((VmaAllocation)userData)->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(); +} +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_BLOCK_METADATA_FUNCTIONS +#endif // _VMA_BLOCK_METADATA + +#ifndef _VMA_BLOCK_BUFFER_IMAGE_GRANULARITY +// Before deleting object of this class remember to call 'Destroy()' +class VmaBlockBufferImageGranularity final +{ +public: + struct ValidationContext + { + const VkAllocationCallbacks* allocCallbacks; + uint16_t* pageAllocs; + }; + + VmaBlockBufferImageGranularity(VkDeviceSize bufferImageGranularity); + ~VmaBlockBufferImageGranularity(); + + bool IsEnabled() const { return m_BufferImageGranularity > MAX_LOW_BUFFER_IMAGE_GRANULARITY; } + + void Init(const VkAllocationCallbacks* pAllocationCallbacks, VkDeviceSize size); + // Before destroying object you must call free it's memory + void Destroy(const VkAllocationCallbacks* pAllocationCallbacks); + + void RoundupAllocRequest(VmaSuballocationType allocType, + VkDeviceSize& inOutAllocSize, + VkDeviceSize& inOutAllocAlignment) const; + + bool CheckConflictAndAlignUp(VkDeviceSize& inOutAllocOffset, + VkDeviceSize allocSize, + VkDeviceSize blockOffset, + VkDeviceSize blockSize, + VmaSuballocationType allocType) const; + + void AllocPages(uint8_t allocType, VkDeviceSize offset, VkDeviceSize size); + void FreePages(VkDeviceSize offset, VkDeviceSize size); + void Clear(); + + ValidationContext StartValidation(const VkAllocationCallbacks* pAllocationCallbacks, + bool isVirutal) const; + bool Validate(ValidationContext& ctx, VkDeviceSize offset, VkDeviceSize size) const; + bool FinishValidation(ValidationContext& ctx) const; + +private: + static const uint16_t MAX_LOW_BUFFER_IMAGE_GRANULARITY = 256; + + struct RegionInfo + { + uint8_t allocType; + uint16_t allocCount; + }; + + VkDeviceSize m_BufferImageGranularity; + uint32_t m_RegionCount; + RegionInfo* m_RegionInfo; + + uint32_t GetStartPage(VkDeviceSize offset) const { return OffsetToPageIndex(offset & ~(m_BufferImageGranularity - 1)); } + uint32_t GetEndPage(VkDeviceSize offset, VkDeviceSize size) const { return OffsetToPageIndex((offset + size - 1) & ~(m_BufferImageGranularity - 1)); } + + uint32_t OffsetToPageIndex(VkDeviceSize offset) const; + void AllocPage(RegionInfo& page, uint8_t allocType); +}; + +#ifndef _VMA_BLOCK_BUFFER_IMAGE_GRANULARITY_FUNCTIONS +VmaBlockBufferImageGranularity::VmaBlockBufferImageGranularity(VkDeviceSize bufferImageGranularity) + : m_BufferImageGranularity(bufferImageGranularity), + m_RegionCount(0), + m_RegionInfo(VMA_NULL) {} + +VmaBlockBufferImageGranularity::~VmaBlockBufferImageGranularity() +{ + VMA_ASSERT(m_RegionInfo == VMA_NULL && "Free not called before destroying object!"); +} + +void VmaBlockBufferImageGranularity::Init(const VkAllocationCallbacks* pAllocationCallbacks, VkDeviceSize size) +{ + if (IsEnabled()) + { + m_RegionCount = static_cast(VmaDivideRoundingUp(size, m_BufferImageGranularity)); + m_RegionInfo = vma_new_array(pAllocationCallbacks, RegionInfo, m_RegionCount); + memset(m_RegionInfo, 0, m_RegionCount * sizeof(RegionInfo)); + } +} + +void VmaBlockBufferImageGranularity::Destroy(const VkAllocationCallbacks* pAllocationCallbacks) +{ + if (m_RegionInfo) + { + vma_delete_array(pAllocationCallbacks, m_RegionInfo, m_RegionCount); + m_RegionInfo = VMA_NULL; + } +} + +void VmaBlockBufferImageGranularity::RoundupAllocRequest(VmaSuballocationType allocType, + VkDeviceSize& inOutAllocSize, + VkDeviceSize& inOutAllocAlignment) const +{ + if (m_BufferImageGranularity > 1 && + m_BufferImageGranularity <= MAX_LOW_BUFFER_IMAGE_GRANULARITY) + { + if (allocType == VMA_SUBALLOCATION_TYPE_UNKNOWN || + allocType == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + allocType == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL) + { + inOutAllocAlignment = VMA_MAX(inOutAllocAlignment, m_BufferImageGranularity); + inOutAllocSize = VmaAlignUp(inOutAllocSize, m_BufferImageGranularity); + } + } +} + +bool VmaBlockBufferImageGranularity::CheckConflictAndAlignUp(VkDeviceSize& inOutAllocOffset, + VkDeviceSize allocSize, + VkDeviceSize blockOffset, + VkDeviceSize blockSize, + VmaSuballocationType allocType) const +{ + if (IsEnabled()) + { + uint32_t startPage = GetStartPage(inOutAllocOffset); + if (m_RegionInfo[startPage].allocCount > 0 && + VmaIsBufferImageGranularityConflict(static_cast(m_RegionInfo[startPage].allocType), allocType)) + { + inOutAllocOffset = VmaAlignUp(inOutAllocOffset, m_BufferImageGranularity); + if (blockSize < allocSize + inOutAllocOffset - blockOffset) + return true; + ++startPage; + } + uint32_t endPage = GetEndPage(inOutAllocOffset, allocSize); + if (endPage != startPage && + m_RegionInfo[endPage].allocCount > 0 && + VmaIsBufferImageGranularityConflict(static_cast(m_RegionInfo[endPage].allocType), allocType)) + { + return true; + } + } + return false; +} + +void VmaBlockBufferImageGranularity::AllocPages(uint8_t allocType, VkDeviceSize offset, VkDeviceSize size) +{ + if (IsEnabled()) + { + uint32_t startPage = GetStartPage(offset); + AllocPage(m_RegionInfo[startPage], allocType); + + uint32_t endPage = GetEndPage(offset, size); + if (startPage != endPage) + AllocPage(m_RegionInfo[endPage], allocType); + } +} + +void VmaBlockBufferImageGranularity::FreePages(VkDeviceSize offset, VkDeviceSize size) +{ + if (IsEnabled()) + { + uint32_t startPage = GetStartPage(offset); + --m_RegionInfo[startPage].allocCount; + if (m_RegionInfo[startPage].allocCount == 0) + m_RegionInfo[startPage].allocType = VMA_SUBALLOCATION_TYPE_FREE; + uint32_t endPage = GetEndPage(offset, size); + if (startPage != endPage) + { + --m_RegionInfo[endPage].allocCount; + if (m_RegionInfo[endPage].allocCount == 0) + m_RegionInfo[endPage].allocType = VMA_SUBALLOCATION_TYPE_FREE; + } + } +} + +void VmaBlockBufferImageGranularity::Clear() +{ + if (m_RegionInfo) + memset(m_RegionInfo, 0, m_RegionCount * sizeof(RegionInfo)); +} + +VmaBlockBufferImageGranularity::ValidationContext VmaBlockBufferImageGranularity::StartValidation( + const VkAllocationCallbacks* pAllocationCallbacks, bool isVirutal) const +{ + ValidationContext ctx{ pAllocationCallbacks, VMA_NULL }; + if (!isVirutal && IsEnabled()) + { + ctx.pageAllocs = vma_new_array(pAllocationCallbacks, uint16_t, m_RegionCount); + memset(ctx.pageAllocs, 0, m_RegionCount * sizeof(uint16_t)); + } + return ctx; +} + +bool VmaBlockBufferImageGranularity::Validate(ValidationContext& ctx, + VkDeviceSize offset, VkDeviceSize size) const +{ + if (IsEnabled()) + { + uint32_t start = GetStartPage(offset); + ++ctx.pageAllocs[start]; + VMA_VALIDATE(m_RegionInfo[start].allocCount > 0); + + uint32_t end = GetEndPage(offset, size); + if (start != end) + { + ++ctx.pageAllocs[end]; + VMA_VALIDATE(m_RegionInfo[end].allocCount > 0); + } + } + return true; +} + +bool VmaBlockBufferImageGranularity::FinishValidation(ValidationContext& ctx) const +{ + // Check proper page structure + if (IsEnabled()) + { + VMA_ASSERT(ctx.pageAllocs != VMA_NULL && "Validation context not initialized!"); + + for (uint32_t page = 0; page < m_RegionCount; ++page) + { + VMA_VALIDATE(ctx.pageAllocs[page] == m_RegionInfo[page].allocCount); + } + vma_delete_array(ctx.allocCallbacks, ctx.pageAllocs, m_RegionCount); + ctx.pageAllocs = VMA_NULL; + } + return true; +} + +uint32_t VmaBlockBufferImageGranularity::OffsetToPageIndex(VkDeviceSize offset) const +{ + return static_cast(offset >> VMA_BITSCAN_MSB(m_BufferImageGranularity)); +} + +void VmaBlockBufferImageGranularity::AllocPage(RegionInfo& page, uint8_t allocType) +{ + // When current alloc type is free then it can be overriden by new type + if (page.allocCount == 0 || (page.allocCount > 0 && page.allocType == VMA_SUBALLOCATION_TYPE_FREE)) + page.allocType = allocType; + + ++page.allocCount; +} +#endif // _VMA_BLOCK_BUFFER_IMAGE_GRANULARITY_FUNCTIONS +#endif // _VMA_BLOCK_BUFFER_IMAGE_GRANULARITY + +#if 0 +#ifndef _VMA_BLOCK_METADATA_GENERIC +class VmaBlockMetadata_Generic : public VmaBlockMetadata +{ + friend class VmaDefragmentationAlgorithm_Generic; + friend class VmaDefragmentationAlgorithm_Fast; + VMA_CLASS_NO_COPY(VmaBlockMetadata_Generic) +public: + VmaBlockMetadata_Generic(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); + virtual ~VmaBlockMetadata_Generic() = default; + + size_t GetAllocationCount() const override { return m_Suballocations.size() - m_FreeCount; } + VkDeviceSize GetSumFreeSize() const override { return m_SumFreeSize; } + bool IsEmpty() const override { return (m_Suballocations.size() == 1) && (m_FreeCount == 1); } + void Free(VmaAllocHandle allocHandle) override { FreeSuballocation(FindAtOffset((VkDeviceSize)allocHandle - 1)); } + VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const override { return (VkDeviceSize)allocHandle - 1; }; + + void Init(VkDeviceSize size) override; + bool Validate() const override; + + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const override; + void AddStatistics(VmaStatistics& inoutStats) const override; + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json, uint32_t mapRefCount) const override; #endif - virtual bool CreateAllocationRequest( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, + bool CreateAllocationRequest( VkDeviceSize allocSize, VkDeviceSize allocAlignment, bool upperAddress, VmaSuballocationType allocType, - bool canMakeOtherLost, uint32_t strategy, - VmaAllocationRequest* pAllocationRequest); + VmaAllocationRequest* pAllocationRequest) override; - virtual bool MakeRequestedAllocationsLost( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VmaAllocationRequest* pAllocationRequest); + VkResult CheckCorruption(const void* pBlockData) override; - virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); - - virtual VkResult CheckCorruption(const void* pBlockData); - - virtual void Alloc( + void Alloc( const VmaAllocationRequest& request, VmaSuballocationType type, - VkDeviceSize allocSize, - VmaAllocation hAllocation); + void* userData) override; - virtual void Free(const VmaAllocation allocation); - virtual void FreeAtOffset(VkDeviceSize offset); - - //////////////////////////////////////////////////////////////////////////////// - // For defragmentation - - bool IsBufferImageGranularityConflictPossible( - VkDeviceSize bufferImageGranularity, - VmaSuballocationType& inOutPrevSuballocType) const; + void GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) override; + void* GetAllocationUserData(VmaAllocHandle allocHandle) const override; + VmaAllocHandle GetAllocationListBegin() const override; + VmaAllocHandle GetNextAllocation(VmaAllocHandle prevAlloc) const override; + void Clear() override; + void SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) override; + void DebugLogAllAllocations() const override; 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; + // Suballocations that are free. Sorted by size, ascending. + VmaVector> m_FreeSuballocationsBySize; + VkDeviceSize AlignAllocationSize(VkDeviceSize size) const { return IsVirtual() ? size : VmaAlignUp(size, (VkDeviceSize)16); } + + VmaSuballocationList::iterator FindAtOffset(VkDeviceSize offset) const; 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; + VmaAllocHandle* pAllocHandle) 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. @@ -5040,6 +6809,685 @@ private: void UnregisterFreeSuballocation(VmaSuballocationList::iterator item); }; +#ifndef _VMA_BLOCK_METADATA_GENERIC_FUNCTIONS +VmaBlockMetadata_Generic::VmaBlockMetadata_Generic(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual) + : VmaBlockMetadata(pAllocationCallbacks, bufferImageGranularity, isVirtual), + m_FreeCount(0), + m_SumFreeSize(0), + m_Suballocations(VmaStlAllocator(pAllocationCallbacks)), + m_FreeSuballocationsBySize(VmaStlAllocator(pAllocationCallbacks)) {} + +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; + + m_Suballocations.push_back(suballoc); + m_FreeSuballocationsBySize.push_back(m_Suballocations.begin()); +} + +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; + + const VkDeviceSize debugMargin = GetDebugMargin(); + + for (const auto& subAlloc : m_Suballocations) + { + // 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); + + VmaAllocation alloc = (VmaAllocation)subAlloc.userData; + if (!IsVirtual()) + { + VMA_VALIDATE(currFree == (alloc == VK_NULL_HANDLE)); + } + + if (currFree) + { + calculatedSumFreeSize += subAlloc.size; + ++calculatedFreeCount; + ++freeSuballocationsToRegister; + + // Margin required between allocations - every free space must be at least that large. + VMA_VALIDATE(subAlloc.size >= debugMargin); + } + else + { + if (!IsVirtual()) + { + VMA_VALIDATE((VkDeviceSize)alloc->GetAllocHandle() == subAlloc.offset + 1); + VMA_VALIDATE(alloc->GetSize() == subAlloc.size); + } + + // Margin required between allocations - previous allocation must be free. + VMA_VALIDATE(debugMargin == 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 calculated values. + VMA_VALIDATE(ValidateFreeSuballocationList()); + VMA_VALIDATE(calculatedOffset == GetSize()); + VMA_VALIDATE(calculatedSumFreeSize == m_SumFreeSize); + VMA_VALIDATE(calculatedFreeCount == m_FreeCount); + + return true; +} + +void VmaBlockMetadata_Generic::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const +{ + const uint32_t rangeCount = (uint32_t)m_Suballocations.size(); + inoutStats.statistics.blockCount++; + inoutStats.statistics.blockBytes += GetSize(); + + for (const auto& suballoc : m_Suballocations) + { + if (suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + VmaAddDetailedStatisticsAllocation(inoutStats, suballoc.size); + else + VmaAddDetailedStatisticsUnusedRange(inoutStats, suballoc.size); + } +} + +void VmaBlockMetadata_Generic::AddStatistics(VmaStatistics& inoutStats) const +{ + inoutStats.blockCount++; + inoutStats.allocationCount += (uint32_t)m_Suballocations.size() - m_FreeCount; + inoutStats.blockBytes += GetSize(); + inoutStats.allocationBytes += GetSize() - m_SumFreeSize; +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata_Generic::PrintDetailedMap(class VmaJsonWriter& json, uint32_t mapRefCount) const +{ + PrintDetailedMap_Begin(json, + m_SumFreeSize, // unusedBytes + m_Suballocations.size() - (size_t)m_FreeCount, // allocationCount + m_FreeCount, // unusedRangeCount + mapRefCount); + + for (const auto& suballoc : m_Suballocations) + { + if (suballoc.type == VMA_SUBALLOCATION_TYPE_FREE) + { + PrintDetailedMap_UnusedRange(json, suballoc.offset, suballoc.size); + } + else + { + PrintDetailedMap_Allocation(json, suballoc.offset, suballoc.size, suballoc.userData); + } + } + + PrintDetailedMap_End(json); +} +#endif // VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Generic::CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(!upperAddress); + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(pAllocationRequest != VMA_NULL); + VMA_HEAVY_ASSERT(Validate()); + + allocSize = AlignAllocationSize(allocSize); + + pAllocationRequest->type = VmaAllocationRequestType::Normal; + pAllocationRequest->size = allocSize; + + const VkDeviceSize debugMargin = GetDebugMargin(); + + // There is not enough total free space in this block to fulfill the request: Early return. + if (m_SumFreeSize < allocSize + debugMargin) + { + return false; + } + + // New algorithm, efficiently searching freeSuballocationsBySize. + const size_t freeSuballocCount = m_FreeSuballocationsBySize.size(); + if (freeSuballocCount > 0) + { + if (strategy == 0 || + strategy == VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT) + { + // Find first free suballocation with size not less than allocSize + debugMargin. + VmaSuballocationList::iterator* const it = VmaBinaryFindFirstNotLess( + m_FreeSuballocationsBySize.data(), + m_FreeSuballocationsBySize.data() + freeSuballocCount, + allocSize + debugMargin, + VmaSuballocationItemSizeLess()); + size_t index = it - m_FreeSuballocationsBySize.data(); + for (; index < freeSuballocCount; ++index) + { + if (CheckAllocation( + allocSize, + allocAlignment, + allocType, + m_FreeSuballocationsBySize[index], + &pAllocationRequest->allocHandle)) + { + 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( + allocSize, + allocAlignment, + allocType, + it, + &pAllocationRequest->allocHandle)) + { + pAllocationRequest->item = it; + return true; + } + } + } + else + { + VMA_ASSERT(strategy & (VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT | VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT )); + // Search staring from biggest suballocations. + for (size_t index = freeSuballocCount; index--; ) + { + if (CheckAllocation( + allocSize, + allocAlignment, + allocType, + m_FreeSuballocationsBySize[index], + &pAllocationRequest->allocHandle)) + { + pAllocationRequest->item = m_FreeSuballocationsBySize[index]; + return true; + } + } + } + } + + return false; +} + +VkResult VmaBlockMetadata_Generic::CheckCorruption(const void* pBlockData) +{ + for (auto& suballoc : m_Suballocations) + { + if (suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + { + if (!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_UNKNOWN_COPY; + } + } + } + + return VK_SUCCESS; +} + +void VmaBlockMetadata_Generic::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + void* userData) +{ + VMA_ASSERT(request.type == VmaAllocationRequestType::Normal); + 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((VkDeviceSize)request.allocHandle - 1 >= suballoc.offset); + const VkDeviceSize paddingBegin = (VkDeviceSize)request.allocHandle - suballoc.offset - 1; + VMA_ASSERT(suballoc.size >= paddingBegin + request.size); + const VkDeviceSize paddingEnd = suballoc.size - paddingBegin - request.size; + + // Unregister this free suballocation from m_FreeSuballocationsBySize and update + // it to become used. + UnregisterFreeSuballocation(request.item); + + suballoc.offset = (VkDeviceSize)request.allocHandle - 1; + suballoc.size = request.size; + suballoc.type = type; + suballoc.userData = userData; + + // If there are any free bytes remaining at the end, insert new free suballocation after current one. + if (paddingEnd) + { + VmaSuballocation paddingSuballoc = {}; + paddingSuballoc.offset = suballoc.offset + suballoc.size; + 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 = suballoc.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 -= request.size; +} + +void VmaBlockMetadata_Generic::GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) +{ + outInfo.offset = (VkDeviceSize)allocHandle - 1; + const VmaSuballocation& suballoc = *FindAtOffset(outInfo.offset); + outInfo.size = suballoc.size; + outInfo.pUserData = suballoc.userData; +} + +void* VmaBlockMetadata_Generic::GetAllocationUserData(VmaAllocHandle allocHandle) const +{ + return FindAtOffset((VkDeviceSize)allocHandle - 1)->userData; +} + +VmaAllocHandle VmaBlockMetadata_Generic::GetAllocationListBegin() const +{ + if (IsEmpty()) + return VK_NULL_HANDLE; + + for (const auto& suballoc : m_Suballocations) + { + if (suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + return (VmaAllocHandle)(suballoc.offset + 1); + } + VMA_ASSERT(false && "Should contain at least 1 allocation!"); + return VK_NULL_HANDLE; +} + +VmaAllocHandle VmaBlockMetadata_Generic::GetNextAllocation(VmaAllocHandle prevAlloc) const +{ + VmaSuballocationList::const_iterator prev = FindAtOffset((VkDeviceSize)prevAlloc - 1); + + for (VmaSuballocationList::const_iterator it = ++prev; it != m_Suballocations.end(); ++it) + { + if (it->type != VMA_SUBALLOCATION_TYPE_FREE) + return (VmaAllocHandle)(it->offset + 1); + } + return VK_NULL_HANDLE; +} + +void VmaBlockMetadata_Generic::Clear() +{ + const VkDeviceSize size = GetSize(); + + VMA_ASSERT(IsVirtual()); + m_FreeCount = 1; + m_SumFreeSize = size; + m_Suballocations.clear(); + m_FreeSuballocationsBySize.clear(); + + VmaSuballocation suballoc = {}; + suballoc.offset = 0; + suballoc.size = size; + suballoc.type = VMA_SUBALLOCATION_TYPE_FREE; + m_Suballocations.push_back(suballoc); + + m_FreeSuballocationsBySize.push_back(m_Suballocations.begin()); +} + +void VmaBlockMetadata_Generic::SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) +{ + VmaSuballocation& suballoc = *FindAtOffset((VkDeviceSize)allocHandle - 1); + suballoc.userData = userData; +} + +void VmaBlockMetadata_Generic::DebugLogAllAllocations() const +{ + for (const auto& suballoc : m_Suballocations) + { + if (suballoc.type != VMA_SUBALLOCATION_TYPE_FREE) + DebugLogAllocation(suballoc.offset, suballoc.size, suballoc.userData); + } +} + +VmaSuballocationList::iterator VmaBlockMetadata_Generic::FindAtOffset(VkDeviceSize offset) const +{ + VMA_HEAVY_ASSERT(!m_Suballocations.empty()); + const VkDeviceSize last = m_Suballocations.rbegin()->offset; + if (last == offset) + return m_Suballocations.rbegin().drop_const(); + const VkDeviceSize first = m_Suballocations.begin()->offset; + if (first == offset) + return m_Suballocations.begin().drop_const(); + + const size_t suballocCount = m_Suballocations.size(); + const VkDeviceSize step = (last - first + m_Suballocations.begin()->size) / suballocCount; + auto findSuballocation = [&](auto begin, auto end) -> VmaSuballocationList::iterator + { + for (auto suballocItem = begin; + suballocItem != end; + ++suballocItem) + { + if (suballocItem->offset == offset) + return suballocItem.drop_const(); + } + VMA_ASSERT(false && "Not found!"); + return m_Suballocations.end().drop_const(); + }; + // If requested offset is closer to the end of range, search from the end + if (offset - first > suballocCount * step / 2) + { + return findSuballocation(m_Suballocations.rbegin(), m_Suballocations.rend()); + } + return findSuballocation(m_Suballocations.begin(), m_Suballocations.end()); +} + +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 >= lastSize); + lastSize = it->size; + } + return true; +} + +bool VmaBlockMetadata_Generic::CheckAllocation( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + VmaSuballocationList::const_iterator suballocItem, + VmaAllocHandle* pAllocHandle) const +{ + VMA_ASSERT(allocSize > 0); + VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); + VMA_ASSERT(suballocItem != m_Suballocations.cend()); + VMA_ASSERT(pAllocHandle != VMA_NULL); + + const VkDeviceSize debugMargin = GetDebugMargin(); + const VkDeviceSize bufferImageGranularity = GetBufferImageGranularity(); + + const VmaSuballocation& suballoc = *suballocItem; + VMA_ASSERT(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE); + + // 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. + VkDeviceSize offset = suballoc.offset + (suballocItem == m_Suballocations.cbegin() ? 0 : GetDebugMargin()); + + // Apply debugMargin from the end of previous alloc. + if (debugMargin > 0) + { + offset += debugMargin; + } + + // Apply alignment. + offset = VmaAlignUp(offset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if (bufferImageGranularity > 1 && bufferImageGranularity != allocAlignment) + { + bool bufferImageGranularityConflict = false; + VmaSuballocationList::const_iterator prevSuballocItem = suballocItem; + while (prevSuballocItem != m_Suballocations.cbegin()) + { + --prevSuballocItem; + const VmaSuballocation& prevSuballoc = *prevSuballocItem; + if (VmaBlocksOnSamePage(prevSuballoc.offset, prevSuballoc.size, offset, bufferImageGranularity)) + { + if (VmaIsBufferImageGranularityConflict(prevSuballoc.type, allocType)) + { + bufferImageGranularityConflict = true; + break; + } + } + else + // Already on previous page. + break; + } + if (bufferImageGranularityConflict) + { + offset = VmaAlignUp(offset, bufferImageGranularity); + } + } + + // Calculate padding at the beginning based on current offset. + const VkDeviceSize paddingBegin = offset - suballoc.offset; + + // Fail if requested size plus margin after is bigger than size of this suballocation. + if (paddingBegin + allocSize + debugMargin > suballoc.size) + { + return false; + } + + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if (allocSize % bufferImageGranularity || offset % bufferImageGranularity) + { + VmaSuballocationList::const_iterator nextSuballocItem = suballocItem; + ++nextSuballocItem; + while (nextSuballocItem != m_Suballocations.cend()) + { + const VmaSuballocation& nextSuballoc = *nextSuballocItem; + if (VmaBlocksOnSamePage(offset, allocSize, nextSuballoc.offset, bufferImageGranularity)) + { + if (VmaIsBufferImageGranularityConflict(allocType, nextSuballoc.type)) + { + return false; + } + } + else + { + // Already on next page. + break; + } + ++nextSuballocItem; + } + } + + *pAllocHandle = (VmaAllocHandle)(offset + 1); + // All tests passed: Success. pAllocHandle 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.userData = VMA_NULL; + + // 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 (m_FreeSuballocationsBySize.empty()) + { + m_FreeSuballocationsBySize.push_back(item); + } + else + { + VmaVectorInsertSorted(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()); + + 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()); +} +#endif // _VMA_BLOCK_METADATA_GENERIC_FUNCTIONS +#endif // _VMA_BLOCK_METADATA_GENERIC +#endif // #if 0 + +#ifndef _VMA_BLOCK_METADATA_LINEAR /* Allocations and their references in internal data structure look like this: @@ -5122,52 +7570,50 @@ class VmaBlockMetadata_Linear : public VmaBlockMetadata { VMA_CLASS_NO_COPY(VmaBlockMetadata_Linear) public: - VmaBlockMetadata_Linear(VmaAllocator hAllocator); - virtual ~VmaBlockMetadata_Linear(); - virtual void Init(VkDeviceSize size); + VmaBlockMetadata_Linear(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); + virtual ~VmaBlockMetadata_Linear() = default; - 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; } + VkDeviceSize GetSumFreeSize() const override { return m_SumFreeSize; } + bool IsEmpty() const override { return GetAllocationCount() == 0; } + VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const override { return (VkDeviceSize)allocHandle - 1; }; - virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const; - virtual void AddPoolStats(VmaPoolStats& inoutStats) const; + void Init(VkDeviceSize size) override; + bool Validate() const override; + size_t GetAllocationCount() const override; + size_t GetFreeRegionsCount() const override; + + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const override; + void AddStatistics(VmaStatistics& inoutStats) const override; #if VMA_STATS_STRING_ENABLED - virtual void PrintDetailedMap(class VmaJsonWriter& json) const; + void PrintDetailedMap(class VmaJsonWriter& json) const override; #endif - virtual bool CreateAllocationRequest( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, + bool CreateAllocationRequest( VkDeviceSize allocSize, VkDeviceSize allocAlignment, bool upperAddress, VmaSuballocationType allocType, - bool canMakeOtherLost, uint32_t strategy, - VmaAllocationRequest* pAllocationRequest); + VmaAllocationRequest* pAllocationRequest) override; - virtual bool MakeRequestedAllocationsLost( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VmaAllocationRequest* pAllocationRequest); + VkResult CheckCorruption(const void* pBlockData) override; - virtual uint32_t MakeAllocationsLost(uint32_t currentFrameIndex, uint32_t frameInUseCount); - - virtual VkResult CheckCorruption(const void* pBlockData); - - virtual void Alloc( + void Alloc( const VmaAllocationRequest& request, VmaSuballocationType type, - VkDeviceSize allocSize, - VmaAllocation hAllocation); + void* userData) override; - virtual void Free(const VmaAllocation allocation); - virtual void FreeAtOffset(VkDeviceSize offset); + void Free(VmaAllocHandle allocHandle) override; + void GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) override; + void* GetAllocationUserData(VmaAllocHandle allocHandle) const override; + VmaAllocHandle GetAllocationListBegin() const override; + VmaAllocHandle GetNextAllocation(VmaAllocHandle prevAlloc) const override; + VkDeviceSize GetNextFreeRegionSize(VmaAllocHandle alloc) const override; + void Clear() override; + void SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) override; + void DebugLogAllAllocations() const override; private: /* @@ -5177,7 +7623,7 @@ private: 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 > SuballocationVectorType; + typedef VmaVector> SuballocationVectorType; enum SECOND_VECTOR_MODE { @@ -5199,12 +7645,6 @@ private: 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. @@ -5212,31 +7652,1567 @@ private: // Number of items in 2nd vector with hAllocation = null. size_t m_2ndNullItemsCount; + 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; } + + VmaSuballocation& FindSuballocation(VkDeviceSize offset) const; bool ShouldCompact1st() const; void CleanupAfterFree(); bool CreateAllocationRequest_LowerAddress( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, VkDeviceSize allocSize, VkDeviceSize allocAlignment, VmaSuballocationType allocType, - bool canMakeOtherLost, uint32_t strategy, VmaAllocationRequest* pAllocationRequest); bool CreateAllocationRequest_UpperAddress( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, VkDeviceSize allocSize, VkDeviceSize allocAlignment, VmaSuballocationType allocType, - bool canMakeOtherLost, uint32_t strategy, VmaAllocationRequest* pAllocationRequest); }; +#ifndef _VMA_BLOCK_METADATA_LINEAR_FUNCTIONS +VmaBlockMetadata_Linear::VmaBlockMetadata_Linear(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual) + : VmaBlockMetadata(pAllocationCallbacks, bufferImageGranularity, isVirtual), + m_SumFreeSize(0), + m_Suballocations0(VmaStlAllocator(pAllocationCallbacks)), + m_Suballocations1(VmaStlAllocator(pAllocationCallbacks)), + m_1stVectorIndex(0), + m_2ndVectorMode(SECOND_VECTOR_EMPTY), + m_1stNullItemsBeginCount(0), + m_1stNullItemsMiddleCount(0), + m_2ndNullItemsCount(0) {} + +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].type != VMA_SUBALLOCATION_TYPE_FREE); + // Null item at the end should be just pop_back(). + VMA_VALIDATE(suballocations1st.back().type != VMA_SUBALLOCATION_TYPE_FREE); + } + if (!suballocations2nd.empty()) + { + // Null item at the end should be just pop_back(). + VMA_VALIDATE(suballocations2nd.back().type != VMA_SUBALLOCATION_TYPE_FREE); + } + + VMA_VALIDATE(m_1stNullItemsBeginCount + m_1stNullItemsMiddleCount <= suballocations1st.size()); + VMA_VALIDATE(m_2ndNullItemsCount <= suballocations2nd.size()); + + VkDeviceSize sumUsedSize = 0; + const size_t suballoc1stCount = suballocations1st.size(); + const VkDeviceSize debugMargin = GetDebugMargin(); + VkDeviceSize offset = 0; + + 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); + + VmaAllocation const alloc = (VmaAllocation)suballoc.userData; + if (!IsVirtual()) + { + VMA_VALIDATE(currFree == (alloc == VK_NULL_HANDLE)); + } + VMA_VALIDATE(suballoc.offset >= offset); + + if (!currFree) + { + if (!IsVirtual()) + { + VMA_VALIDATE((VkDeviceSize)alloc->GetAllocHandle() == suballoc.offset + 1); + VMA_VALIDATE(alloc->GetSize() == suballoc.size); + } + sumUsedSize += suballoc.size; + } + else + { + ++nullItem2ndCount; + } + + offset = suballoc.offset + suballoc.size + debugMargin; + } + + 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.userData == VMA_NULL); + } + + 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); + + VmaAllocation const alloc = (VmaAllocation)suballoc.userData; + if (!IsVirtual()) + { + VMA_VALIDATE(currFree == (alloc == VK_NULL_HANDLE)); + } + VMA_VALIDATE(suballoc.offset >= offset); + VMA_VALIDATE(i >= m_1stNullItemsBeginCount || currFree); + + if (!currFree) + { + if (!IsVirtual()) + { + VMA_VALIDATE((VkDeviceSize)alloc->GetAllocHandle() == suballoc.offset + 1); + VMA_VALIDATE(alloc->GetSize() == suballoc.size); + } + sumUsedSize += suballoc.size; + } + else + { + ++nullItem1stCount; + } + + offset = suballoc.offset + suballoc.size + debugMargin; + } + 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); + + VmaAllocation const alloc = (VmaAllocation)suballoc.userData; + if (!IsVirtual()) + { + VMA_VALIDATE(currFree == (alloc == VK_NULL_HANDLE)); + } + VMA_VALIDATE(suballoc.offset >= offset); + + if (!currFree) + { + if (!IsVirtual()) + { + VMA_VALIDATE((VkDeviceSize)alloc->GetAllocHandle() == suballoc.offset + 1); + VMA_VALIDATE(alloc->GetSize() == suballoc.size); + } + sumUsedSize += suballoc.size; + } + else + { + ++nullItem2ndCount; + } + + offset = suballoc.offset + suballoc.size + debugMargin; + } + + 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; +} + +size_t VmaBlockMetadata_Linear::GetFreeRegionsCount() const +{ + // Function only used for defragmentation, which is disabled for this algorithm + VMA_ASSERT(0); + return SIZE_MAX; +} + +void VmaBlockMetadata_Linear::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) 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(); + + inoutStats.statistics.blockCount++; + inoutStats.statistics.blockBytes += size; + + 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].userData == VMA_NULL) + { + ++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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + VmaAddDetailedStatisticsAllocation(inoutStats, 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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, 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].userData == VMA_NULL) + { + ++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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + VmaAddDetailedStatisticsAllocation(inoutStats, 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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, 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].userData == VMA_NULL) + { + --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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, unusedRangeSize); + } + + // 2. Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + VmaAddDetailedStatisticsAllocation(inoutStats, 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; + VmaAddDetailedStatisticsUnusedRange(inoutStats, unusedRangeSize); + } + + // End of loop. + lastOffset = size; + } + } + } +} + +void VmaBlockMetadata_Linear::AddStatistics(VmaStatistics& 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.blockCount++; + inoutStats.blockBytes += size; + inoutStats.allocationBytes += size - m_SumFreeSize; + + 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].userData == VMA_NULL) + { + ++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; + } + + // 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; + } + + // 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].userData == VMA_NULL) + { + ++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; + } + + // 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; + } + + // 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].userData == VMA_NULL) + { + --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; + } + + // 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; + } + + // 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].userData == VMA_NULL) + { + ++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].userData == VMA_NULL) + { + ++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].userData == VMA_NULL) + { + --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].userData == VMA_NULL) + { + ++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.size, suballoc.userData); + + // 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].userData == VMA_NULL) + { + ++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.size, suballoc.userData); + + // 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].userData == VMA_NULL) + { + --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.size, suballoc.userData); + + // 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 // VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Linear::CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + 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()); + pAllocationRequest->size = allocSize; + return upperAddress ? + CreateAllocationRequest_UpperAddress( + allocSize, allocAlignment, allocType, strategy, pAllocationRequest) : + CreateAllocationRequest_LowerAddress( + allocSize, allocAlignment, allocType, strategy, pAllocationRequest); +} + +VkResult VmaBlockMetadata_Linear::CheckCorruption(const void* pBlockData) +{ + VMA_ASSERT(!IsVirtual()); + 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 + suballoc.size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_UNKNOWN_COPY; + } + } + } + + 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 + suballoc.size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_UNKNOWN_COPY; + } + } + } + + return VK_SUCCESS; +} + +void VmaBlockMetadata_Linear::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + void* userData) +{ + const VkDeviceSize offset = (VkDeviceSize)request.allocHandle - 1; + const VmaSuballocation newSuballoc = { offset, request.size, userData, type }; + + switch (request.type) + { + case VmaAllocationRequestType::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; + } + break; + case VmaAllocationRequestType::EndOf1st: + { + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + + VMA_ASSERT(suballocations1st.empty() || + offset >= suballocations1st.back().offset + suballocations1st.back().size); + // Check if it fits before the end of the block. + VMA_ASSERT(offset + request.size <= GetSize()); + + suballocations1st.push_back(newSuballoc); + } + break; + case VmaAllocationRequestType::EndOf2nd: + { + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + // New allocation at the end of 2-part ring buffer, so before first allocation from 1st vector. + VMA_ASSERT(!suballocations1st.empty() && + offset + request.size <= 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); + } + break; + default: + VMA_ASSERT(0 && "CRITICAL INTERNAL ERROR."); + } + + m_SumFreeSize -= newSuballoc.size; +} + +void VmaBlockMetadata_Linear::Free(VmaAllocHandle allocHandle) +{ + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + VkDeviceSize offset = (VkDeviceSize)allocHandle - 1; + + 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.userData = VMA_NULL; + 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; + } + } + + VmaSuballocation refSuballoc; + refSuballoc.offset = offset; + // Rest of members stays uninitialized intentionally for better performance. + + // Item from the middle of 1st vector. + { + const SuballocationVectorType::iterator it = VmaBinaryFindSorted( + suballocations1st.begin() + m_1stNullItemsBeginCount, + suballocations1st.end(), + refSuballoc, + VmaSuballocationOffsetLess()); + if (it != suballocations1st.end()) + { + it->type = VMA_SUBALLOCATION_TYPE_FREE; + it->userData = VMA_NULL; + ++m_1stNullItemsMiddleCount; + m_SumFreeSize += it->size; + CleanupAfterFree(); + return; + } + } + + if (m_2ndVectorMode != SECOND_VECTOR_EMPTY) + { + // Item from the middle of 2nd vector. + const SuballocationVectorType::iterator it = m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER ? + VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetLess()) : + VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetGreater()); + if (it != suballocations2nd.end()) + { + it->type = VMA_SUBALLOCATION_TYPE_FREE; + it->userData = VMA_NULL; + ++m_2ndNullItemsCount; + m_SumFreeSize += it->size; + CleanupAfterFree(); + return; + } + } + + VMA_ASSERT(0 && "Allocation to free not found in linear allocator!"); +} + +void VmaBlockMetadata_Linear::GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) +{ + outInfo.offset = (VkDeviceSize)allocHandle - 1; + VmaSuballocation& suballoc = FindSuballocation(outInfo.offset); + outInfo.size = suballoc.size; + outInfo.pUserData = suballoc.userData; +} + +void* VmaBlockMetadata_Linear::GetAllocationUserData(VmaAllocHandle allocHandle) const +{ + return FindSuballocation((VkDeviceSize)allocHandle - 1).userData; +} + +VmaAllocHandle VmaBlockMetadata_Linear::GetAllocationListBegin() const +{ + // Function only used for defragmentation, which is disabled for this algorithm + VMA_ASSERT(0); + return VK_NULL_HANDLE; +} + +VmaAllocHandle VmaBlockMetadata_Linear::GetNextAllocation(VmaAllocHandle prevAlloc) const +{ + // Function only used for defragmentation, which is disabled for this algorithm + VMA_ASSERT(0); + return VK_NULL_HANDLE; +} + +VkDeviceSize VmaBlockMetadata_Linear::GetNextFreeRegionSize(VmaAllocHandle alloc) const +{ + // Function only used for defragmentation, which is disabled for this algorithm + VMA_ASSERT(0); + return 0; +} + +void VmaBlockMetadata_Linear::Clear() +{ + m_SumFreeSize = GetSize(); + m_Suballocations0.clear(); + m_Suballocations1.clear(); + // Leaving m_1stVectorIndex unchanged - it doesn't matter. + m_2ndVectorMode = SECOND_VECTOR_EMPTY; + m_1stNullItemsBeginCount = 0; + m_1stNullItemsMiddleCount = 0; + m_2ndNullItemsCount = 0; +} + +void VmaBlockMetadata_Linear::SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) +{ + VmaSuballocation& suballoc = FindSuballocation((VkDeviceSize)allocHandle - 1); + suballoc.userData = userData; +} + +void VmaBlockMetadata_Linear::DebugLogAllAllocations() const +{ + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + for (auto it = suballocations1st.begin() + m_1stNullItemsBeginCount; it != suballocations1st.end(); ++it) + if (it->type != VMA_SUBALLOCATION_TYPE_FREE) + DebugLogAllocation(it->offset, it->size, it->userData); + + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + for (auto it = suballocations2nd.begin(); it != suballocations2nd.end(); ++it) + if (it->type != VMA_SUBALLOCATION_TYPE_FREE) + DebugLogAllocation(it->offset, it->size, it->userData); +} + +VmaSuballocation& VmaBlockMetadata_Linear::FindSuballocation(VkDeviceSize offset) const +{ + const SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + const SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + VmaSuballocation refSuballoc; + refSuballoc.offset = offset; + // Rest of members stays uninitialized intentionally for better performance. + + // Item from the 1st vector. + { + SuballocationVectorType::const_iterator it = VmaBinaryFindSorted( + suballocations1st.begin() + m_1stNullItemsBeginCount, + suballocations1st.end(), + refSuballoc, + VmaSuballocationOffsetLess()); + if (it != suballocations1st.end()) + { + return const_cast(*it); + } + } + + if (m_2ndVectorMode != SECOND_VECTOR_EMPTY) + { + // Rest of members stays uninitialized intentionally for better performance. + SuballocationVectorType::const_iterator it = m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER ? + VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetLess()) : + VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetGreater()); + if (it != suballocations2nd.end()) + { + return const_cast(*it); + } + } + + VMA_ASSERT(0 && "Allocation not found in linear allocator!"); + return const_cast(suballocations1st.back()); // Should never occur. +} + +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].type == VMA_SUBALLOCATION_TYPE_FREE) + { + ++m_1stNullItemsBeginCount; + --m_1stNullItemsMiddleCount; + } + + // Find more null items at the end of 1st vector. + while (m_1stNullItemsMiddleCount > 0 && + suballocations1st.back().type == VMA_SUBALLOCATION_TYPE_FREE) + { + --m_1stNullItemsMiddleCount; + suballocations1st.pop_back(); + } + + // Find more null items at the end of 2nd vector. + while (m_2ndNullItemsCount > 0 && + suballocations2nd.back().type == VMA_SUBALLOCATION_TYPE_FREE) + { + --m_2ndNullItemsCount; + suballocations2nd.pop_back(); + } + + // Find more null items at the beginning of 2nd vector. + while (m_2ndNullItemsCount > 0 && + suballocations2nd[0].type == VMA_SUBALLOCATION_TYPE_FREE) + { + --m_2ndNullItemsCount; + VmaVectorRemove(suballocations2nd, 0); + } + + if (ShouldCompact1st()) + { + const size_t nonNullItemCount = suballoc1stCount - nullItem1stCount; + size_t srcIndex = m_1stNullItemsBeginCount; + for (size_t dstIndex = 0; dstIndex < nonNullItemCount; ++dstIndex) + { + while (suballocations1st[srcIndex].type == VMA_SUBALLOCATION_TYPE_FREE) + { + ++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].type == VMA_SUBALLOCATION_TYPE_FREE) + { + ++m_1stNullItemsBeginCount; + --m_1stNullItemsMiddleCount; + } + m_2ndNullItemsCount = 0; + m_1stVectorIndex ^= 1; + } + } + } + + VMA_HEAVY_ASSERT(Validate()); +} + +bool VmaBlockMetadata_Linear::CreateAllocationRequest_LowerAddress( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + const VkDeviceSize blockSize = GetSize(); + const VkDeviceSize debugMargin = GetDebugMargin(); + const VkDeviceSize bufferImageGranularity = GetBufferImageGranularity(); + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + 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 + debugMargin; + } + + // Start from offset equal to beginning of free space. + VkDeviceSize resultOffset = resultBaseOffset; + + // Apply alignment. + resultOffset = VmaAlignUp(resultOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if (bufferImageGranularity > 1 && bufferImageGranularity != allocAlignment && !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 : blockSize; + + // There is enough free space at the end after alignment. + if (resultOffset + allocSize + debugMargin <= freeSpaceEnd) + { + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if ((allocSize % bufferImageGranularity || resultOffset % bufferImageGranularity) && 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->allocHandle = (VmaAllocHandle)(resultOffset + 1); + // pAllocationRequest->item, customData unused. + pAllocationRequest->type = VmaAllocationRequestType::EndOf1st; + 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 + debugMargin; + } + + // Start from offset equal to beginning of free space. + VkDeviceSize resultOffset = resultBaseOffset; + + // Apply alignment. + resultOffset = VmaAlignUp(resultOffset, allocAlignment); + + // Check previous suballocations for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if (bufferImageGranularity > 1 && bufferImageGranularity != allocAlignment && !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); + } + } + + size_t index1st = m_1stNullItemsBeginCount; + + // There is enough free space at the end after alignment. + if ((index1st == suballocations1st.size() && resultOffset + allocSize + debugMargin <= blockSize) || + (index1st < suballocations1st.size() && resultOffset + allocSize + debugMargin <= suballocations1st[index1st].offset)) + { + // Check next suballocations for BufferImageGranularity conflicts. + // If conflict exists, allocation cannot be made here. + if (allocSize % bufferImageGranularity || resultOffset % bufferImageGranularity) + { + 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->allocHandle = (VmaAllocHandle)(resultOffset + 1); + pAllocationRequest->type = VmaAllocationRequestType::EndOf2nd; + // pAllocationRequest->item, customData unused. + return true; + } + } + + return false; +} + +bool VmaBlockMetadata_Linear::CreateAllocationRequest_UpperAddress( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + const VkDeviceSize blockSize = GetSize(); + const VkDeviceSize bufferImageGranularity = GetBufferImageGranularity(); + SuballocationVectorType& suballocations1st = AccessSuballocations1st(); + SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); + + 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 > blockSize) + { + return false; + } + VkDeviceSize resultBaseOffset = blockSize - 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; + + const VkDeviceSize debugMargin = GetDebugMargin(); + + // Apply debugMargin at the end. + if (debugMargin > 0) + { + if (resultOffset < debugMargin) + { + return false; + } + resultOffset -= debugMargin; + } + + // Apply alignment. + resultOffset = VmaAlignDown(resultOffset, allocAlignment); + + // Check next suballocations from 2nd for BufferImageGranularity conflicts. + // Make bigger alignment if necessary. + if (bufferImageGranularity > 1 && bufferImageGranularity != allocAlignment && !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 + debugMargin <= 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->allocHandle = (VmaAllocHandle)(resultOffset + 1); + // pAllocationRequest->item unused. + pAllocationRequest->type = VmaAllocationRequestType::UpperAddress; + return true; + } + + return false; +} +#endif // _VMA_BLOCK_METADATA_LINEAR_FUNCTIONS +#endif // _VMA_BLOCK_METADATA_LINEAR + +#if 0 +#ifndef _VMA_BLOCK_METADATA_BUDDY /* - GetSize() is the original size of allocated memory block. - m_UsableSize is this size aligned down to a power of two. @@ -5252,69 +9228,57 @@ class VmaBlockMetadata_Buddy : public VmaBlockMetadata { VMA_CLASS_NO_COPY(VmaBlockMetadata_Buddy) public: - VmaBlockMetadata_Buddy(VmaAllocator hAllocator); + VmaBlockMetadata_Buddy(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); 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; } + size_t GetAllocationCount() const override { return m_AllocationCount; } + VkDeviceSize GetSumFreeSize() const override { return m_SumFreeSize + GetUnusableSize(); } + bool IsEmpty() const override { return m_Root->type == Node::TYPE_FREE; } + VkResult CheckCorruption(const void* pBlockData) override { return VK_ERROR_FEATURE_NOT_PRESENT; } + VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const override { return (VkDeviceSize)allocHandle - 1; }; + void DebugLogAllAllocations() const override { DebugLogAllAllocationNode(m_Root, 0); } - virtual void CalcAllocationStatInfo(VmaStatInfo& outInfo) const; - virtual void AddPoolStats(VmaPoolStats& inoutStats) const; + void Init(VkDeviceSize size) override; + bool Validate() const override; + + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const override; + void AddStatistics(VmaStatistics& inoutStats) const override; #if VMA_STATS_STRING_ENABLED - virtual void PrintDetailedMap(class VmaJsonWriter& json) const; + void PrintDetailedMap(class VmaJsonWriter& json, uint32_t mapRefCount) const override; #endif - virtual bool CreateAllocationRequest( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, + bool CreateAllocationRequest( VkDeviceSize allocSize, VkDeviceSize allocAlignment, bool upperAddress, VmaSuballocationType allocType, - bool canMakeOtherLost, uint32_t strategy, - VmaAllocationRequest* pAllocationRequest); + VmaAllocationRequest* pAllocationRequest) override; - 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( + void Alloc( const VmaAllocationRequest& request, VmaSuballocationType type, - VkDeviceSize allocSize, - VmaAllocation hAllocation); + void* userData) override; - virtual void Free(const VmaAllocation allocation) { FreeAtOffset(allocation, allocation->GetOffset()); } - virtual void FreeAtOffset(VkDeviceSize offset) { FreeAtOffset(VMA_NULL, offset); } + void Free(VmaAllocHandle allocHandle) override; + void GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) override; + void* GetAllocationUserData(VmaAllocHandle allocHandle) const override; + VmaAllocHandle GetAllocationListBegin() const override; + VmaAllocHandle GetNextAllocation(VmaAllocHandle prevAlloc) const override; + void Clear() override; + void SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) override; private: - static const VkDeviceSize MIN_NODE_SIZE = 32; - static const size_t MAX_LEVELS = 30; + static const size_t MAX_LEVELS = 48; struct ValidationContext { - size_t calculatedAllocationCount; - size_t calculatedFreeCount; - VkDeviceSize calculatedSumFreeSize; - - ValidationContext() : - calculatedAllocationCount(0), - calculatedFreeCount(0), - calculatedSumFreeSize(0) { } + size_t calculatedAllocationCount = 0; + size_t calculatedFreeCount = 0; + VkDeviceSize calculatedSumFreeSize = 0; }; - struct Node { VkDeviceSize offset; @@ -5337,7 +9301,7 @@ private: } free; struct { - VmaAllocation alloc; + void* userData; } allocation; struct { @@ -5349,27 +9313,38 @@ private: // Size of the memory block aligned down to a power of two. VkDeviceSize m_UsableSize; uint32_t m_LevelCount; - + VmaPoolAllocator m_NodeAllocator; Node* m_Root; - struct { + 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. + // Doesn't include space wasted due to internal fragmentation - allocation sizes are just aligned up to node sizes. + // Doesn't include unusable size. VkDeviceSize m_SumFreeSize; VkDeviceSize GetUnusableSize() const { return GetSize() - m_UsableSize; } - void DeleteNode(Node* node); + VkDeviceSize LevelToNodeSize(uint32_t level) const { return m_UsableSize >> level; } + + VkDeviceSize AlignAllocationSize(VkDeviceSize size) const + { + if (!IsVirtual()) + { + size = VmaAlignUp(size, (VkDeviceSize)16); + } + return VmaNextPow2(size); + } + Node* FindAllocationNode(VkDeviceSize offset, uint32_t& outLevel) const; + void DeleteNodeChildren(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; + void AddNodeToDetailedStatistics(VmaDetailedStatistics& inoutStats, 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. @@ -5378,113 +9353,1517 @@ private: // node->type must be FREE. // node->free.prev, next stay untouched. void RemoveFromFreeList(uint32_t level, Node* node); + void DebugLogAllAllocationNode(Node* node, uint32_t level) const; #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 +#ifndef _VMA_BLOCK_METADATA_BUDDY_FUNCTIONS +VmaBlockMetadata_Buddy::VmaBlockMetadata_Buddy(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual) + : VmaBlockMetadata(pAllocationCallbacks, bufferImageGranularity, isVirtual), + m_NodeAllocator(pAllocationCallbacks, 32), // firstBlockCapacity + m_Root(VMA_NULL), + m_AllocationCount(0), + m_FreeCount(1), + m_SumFreeSize(0) { - VMA_CLASS_NO_COPY(VmaDeviceMemoryBlock) -public: - VmaBlockMetadata* m_pMetadata; + memset(m_FreeList, 0, sizeof(m_FreeList)); +} - VmaDeviceMemoryBlock(VmaAllocator hAllocator); +VmaBlockMetadata_Buddy::~VmaBlockMetadata_Buddy() +{ + DeleteNodeChildren(m_Root); + m_NodeAllocator.Free(m_Root); +} - ~VmaDeviceMemoryBlock() +void VmaBlockMetadata_Buddy::Init(VkDeviceSize size) +{ + VmaBlockMetadata::Init(size); + + m_UsableSize = VmaPrevPow2(size); + m_SumFreeSize = m_UsableSize; + + // Calculate m_LevelCount. + const VkDeviceSize minNodeSize = IsVirtual() ? 1 : 16; + m_LevelCount = 1; + while (m_LevelCount < MAX_LEVELS && + LevelToNodeSize(m_LevelCount) >= minNodeSize) { - VMA_ASSERT(m_MapCount == 0 && "VkDeviceMemory block is being destroyed while it is still mapped."); - VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); + ++m_LevelCount; } - // Always call after construction. - void Init( - VmaAllocator hAllocator, - VmaPool hParentPool, - uint32_t newMemoryTypeIndex, - VkDeviceMemory newMemory, - VkDeviceSize newSize, - uint32_t id, - uint32_t algorithm); - // Always call before destruction. - void Destroy(VmaAllocator allocator); + Node* rootNode = m_NodeAllocator.Alloc(); + rootNode->offset = 0; + rootNode->type = Node::TYPE_FREE; + rootNode->parent = VMA_NULL; + rootNode->buddy = VMA_NULL; - VmaPool GetParentPool() const { return m_hParentPool; } - 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; } + m_Root = rootNode; + AddToFreeListFront(0, rootNode); +} - // Validates all data structures inside this object. If not valid, returns false. - bool Validate() const; +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); - VkResult CheckCorruption(VmaAllocator hAllocator); + // 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); - // ppData can be null. - VkResult Map(VmaAllocator hAllocator, uint32_t count, void** ppData); - void Unmap(VmaAllocator hAllocator, uint32_t count); + for (Node* node = m_FreeList[level].front; + node != VMA_NULL; + node = node->free.next) + { + VMA_VALIDATE(node->type == Node::TYPE_FREE); - VkResult WriteMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); - VkResult ValidateMagicValueAroundAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize); + if (node->free.next == VMA_NULL) + { + VMA_VALIDATE(m_FreeList[level].back == node); + } + else + { + VMA_VALIDATE(node->free.next->free.prev == node); + } + } + } - VkResult BindBufferMemory( - const VmaAllocator hAllocator, - const VmaAllocation hAllocation, - VkDeviceSize allocationLocalOffset, - VkBuffer hBuffer, - const void* pNext); - VkResult BindImageMemory( - const VmaAllocator hAllocator, - const VmaAllocation hAllocation, - VkDeviceSize allocationLocalOffset, - VkImage hImage, - const void* pNext); + // 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; +} + +void VmaBlockMetadata_Buddy::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const +{ + inoutStats.statistics.blockCount++; + inoutStats.statistics.blockBytes += GetSize(); + + AddNodeToDetailedStatistics(inoutStats, m_Root, LevelToNodeSize(0)); + + const VkDeviceSize unusableSize = GetUnusableSize(); + if (unusableSize > 0) + VmaAddDetailedStatisticsUnusedRange(inoutStats, unusableSize); +} + +void VmaBlockMetadata_Buddy::AddStatistics(VmaStatistics& inoutStats) const +{ + inoutStats.blockCount++; + inoutStats.allocationCount += (uint32_t)m_AllocationCount; + inoutStats.blockBytes += GetSize(); + inoutStats.allocationBytes += GetSize() - m_SumFreeSize; +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata_Buddy::PrintDetailedMap(class VmaJsonWriter& json, uint32_t mapRefCount) const +{ + VmaDetailedStatistics stats; + VmaClearDetailedStatistics(stats); + AddDetailedStatistics(stats); + + PrintDetailedMap_Begin( + json, + stats.statistics.blockBytes - stats.statistics.allocationBytes, + stats.statistics.allocationCount, + stats.unusedRangeCount, + mapRefCount); + + 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 // VMA_STATS_STRING_ENABLED + +bool VmaBlockMetadata_Buddy::CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(!upperAddress && "VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT can be used only with linear algorithm."); + + allocSize = AlignAllocationSize(allocSize); + + // 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, GetBufferImageGranularity()); + allocSize = VmaAlignUp(allocSize, GetBufferImageGranularity()); + } + + if (allocSize > m_UsableSize) + { + return false; + } + + const uint32_t targetLevel = AllocSizeToLevel(allocSize); + for (uint32_t level = targetLevel; level--; ) + { + for (Node* freeNode = m_FreeList[level].front; + freeNode != VMA_NULL; + freeNode = freeNode->free.next) + { + if (freeNode->offset % allocAlignment == 0) + { + pAllocationRequest->type = VmaAllocationRequestType::Normal; + pAllocationRequest->allocHandle = (VmaAllocHandle)(freeNode->offset + 1); + pAllocationRequest->size = allocSize; + pAllocationRequest->customData = (void*)(uintptr_t)level; + return true; + } + } + } + + return false; +} + +void VmaBlockMetadata_Buddy::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + void* userData) +{ + VMA_ASSERT(request.type == VmaAllocationRequestType::Normal); + + const uint32_t targetLevel = AllocSizeToLevel(request.size); + 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); + const VkDeviceSize offset = (VkDeviceSize)request.allocHandle - 1; + while (currNode->offset != 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 = m_NodeAllocator.Alloc(); + Node* rightChild = m_NodeAllocator.Alloc(); + + 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; + ++currLevel; + currNode = m_FreeList[currLevel].front; + + /* + We can be sure that currNode, as left child of node previously split, + also fulfills 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.userData = userData; + + ++m_AllocationCount; + --m_FreeCount; + m_SumFreeSize -= request.size; +} + +void VmaBlockMetadata_Buddy::GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) +{ + uint32_t level = 0; + outInfo.offset = (VkDeviceSize)allocHandle - 1; + const Node* const node = FindAllocationNode(outInfo.offset, level); + outInfo.size = LevelToNodeSize(level); + outInfo.pUserData = node->allocation.userData; +} + +void* VmaBlockMetadata_Buddy::GetAllocationUserData(VmaAllocHandle allocHandle) const +{ + uint32_t level = 0; + const Node* const node = FindAllocationNode((VkDeviceSize)allocHandle - 1, level); + return node->allocation.userData; +} + +VmaAllocHandle VmaBlockMetadata_Buddy::GetAllocationListBegin() const +{ + // Function only used for defragmentation, which is disabled for this algorithm + return VK_NULL_HANDLE; +} + +VmaAllocHandle VmaBlockMetadata_Buddy::GetNextAllocation(VmaAllocHandle prevAlloc) const +{ + // Function only used for defragmentation, which is disabled for this algorithm + return VK_NULL_HANDLE; +} + +void VmaBlockMetadata_Buddy::DeleteNodeChildren(Node* node) +{ + if (node->type == Node::TYPE_SPLIT) + { + DeleteNodeChildren(node->split.leftChild->buddy); + DeleteNodeChildren(node->split.leftChild); + const VkAllocationCallbacks* allocationCallbacks = GetAllocationCallbacks(); + m_NodeAllocator.Free(node->split.leftChild->buddy); + m_NodeAllocator.Free(node->split.leftChild); + } +} + +void VmaBlockMetadata_Buddy::Clear() +{ + DeleteNodeChildren(m_Root); + m_Root->type = Node::TYPE_FREE; + m_AllocationCount = 0; + m_FreeCount = 1; + m_SumFreeSize = m_UsableSize; +} + +void VmaBlockMetadata_Buddy::SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) +{ + uint32_t level = 0; + Node* const node = FindAllocationNode((VkDeviceSize)allocHandle - 1, level); + node->allocation.userData = userData; +} + +VmaBlockMetadata_Buddy::Node* VmaBlockMetadata_Buddy::FindAllocationNode(VkDeviceSize offset, uint32_t& outLevel) const +{ + Node* node = m_Root; + VkDeviceSize nodeOffset = 0; + outLevel = 0; + VkDeviceSize levelNodeSize = LevelToNodeSize(0); + while (node->type == Node::TYPE_SPLIT) + { + const VkDeviceSize nextLevelNodeSize = levelNodeSize >> 1; + if (offset < nodeOffset + nextLevelNodeSize) + { + node = node->split.leftChild; + } + else + { + node = node->split.leftChild->buddy; + nodeOffset += nextLevelNodeSize; + } + ++outLevel; + levelNodeSize = nextLevelNodeSize; + } + + VMA_ASSERT(node != VMA_NULL && node->type == Node::TYPE_ALLOCATION); + return 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; + if (!IsVirtual()) + { + VMA_VALIDATE(curr->allocation.userData != VMA_NULL); + } + break; + case Node::TYPE_SPLIT: + { + const uint32_t childrenLevel = level + 1; + const VkDeviceSize childrenLevelNodeSize = levelNodeSize >> 1; + 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 >>= 1; + nextLevelNodeSize >>= 1; + } + return level; +} + +void VmaBlockMetadata_Buddy::Free(VmaAllocHandle allocHandle) +{ + uint32_t level = 0; + Node* node = FindAllocationNode((VkDeviceSize)allocHandle - 1, level); + + ++m_FreeCount; + --m_AllocationCount; + m_SumFreeSize += LevelToNodeSize(level); + + 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; + + m_NodeAllocator.Free(node->buddy); + m_NodeAllocator.Free(node); + parent->type = Node::TYPE_FREE; + + node = parent; + --level; + --m_FreeCount; + } + + AddToFreeListFront(level, node); +} + +void VmaBlockMetadata_Buddy::AddNodeToDetailedStatistics(VmaDetailedStatistics& inoutStats, const Node* node, VkDeviceSize levelNodeSize) const +{ + switch (node->type) + { + case Node::TYPE_FREE: + VmaAddDetailedStatisticsUnusedRange(inoutStats, levelNodeSize); + break; + case Node::TYPE_ALLOCATION: + VmaAddDetailedStatisticsAllocation(inoutStats, levelNodeSize); + break; + case Node::TYPE_SPLIT: + { + const VkDeviceSize childrenNodeSize = levelNodeSize / 2; + const Node* const leftChild = node->split.leftChild; + AddNodeToDetailedStatistics(inoutStats, leftChild, childrenNodeSize); + const Node* const rightChild = leftChild->buddy; + AddNodeToDetailedStatistics(inoutStats, 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; + } +} + +void VmaBlockMetadata_Buddy::DebugLogAllAllocationNode(Node* node, uint32_t level) const +{ + switch (node->type) + { + case Node::TYPE_FREE: + break; + case Node::TYPE_ALLOCATION: + DebugLogAllocation(node->offset, LevelToNodeSize(level), node->allocation.userData); + break; + case Node::TYPE_SPLIT: + { + ++level; + DebugLogAllAllocationNode(node->split.leftChild, level); + DebugLogAllAllocationNode(node->split.leftChild->buddy, level); + } + break; + default: + VMA_ASSERT(0); + } +} + +#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, levelNodeSize, node->allocation.userData); + 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 // VMA_STATS_STRING_ENABLED +#endif // _VMA_BLOCK_METADATA_BUDDY_FUNCTIONS +#endif // _VMA_BLOCK_METADATA_BUDDY +#endif // #if 0 + +#ifndef _VMA_BLOCK_METADATA_TLSF +// To not search current larger region if first allocation won't succeed and skip to smaller range +// use with VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT as strategy in CreateAllocationRequest(). +// When fragmentation and reusal of previous blocks doesn't matter then use with +// VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT for fastest alloc time possible. +class VmaBlockMetadata_TLSF : public VmaBlockMetadata +{ + VMA_CLASS_NO_COPY(VmaBlockMetadata_TLSF) +public: + VmaBlockMetadata_TLSF(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); + virtual ~VmaBlockMetadata_TLSF(); + + size_t GetAllocationCount() const override { return m_AllocCount; } + size_t GetFreeRegionsCount() const override { return m_BlocksFreeCount + 1; } + VkDeviceSize GetSumFreeSize() const override { return m_BlocksFreeSize + m_NullBlock->size; } + bool IsEmpty() const override { return m_NullBlock->offset == 0; } + VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const override { return ((Block*)allocHandle)->offset; }; + + void Init(VkDeviceSize size) override; + bool Validate() const override; + + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const override; + void AddStatistics(VmaStatistics& inoutStats) const override; + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json) const override; +#endif + + bool CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) override; + + VkResult CheckCorruption(const void* pBlockData) override; + void Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + void* userData) override; + + void Free(VmaAllocHandle allocHandle) override; + void GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) override; + void* GetAllocationUserData(VmaAllocHandle allocHandle) const override; + VmaAllocHandle GetAllocationListBegin() const override; + VmaAllocHandle GetNextAllocation(VmaAllocHandle prevAlloc) const override; + VkDeviceSize GetNextFreeRegionSize(VmaAllocHandle alloc) const override; + void Clear() override; + void SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) override; + void DebugLogAllAllocations() const override; private: - VmaPool m_hParentPool; // VK_NULL_HANDLE if not belongs to custom pool. - uint32_t m_MemoryTypeIndex; - uint32_t m_Id; - VkDeviceMemory m_hMemory; + // According to original paper it should be preferable 4 or 5: + // M. Masmano, I. Ripoll, A. Crespo, and J. Real "TLSF: a New Dynamic Memory Allocator for Real-Time Systems" + // http://www.gii.upv.es/tlsf/files/ecrts04_tlsf.pdf + static const uint8_t SECOND_LEVEL_INDEX = 5; + static const uint16_t SMALL_BUFFER_SIZE = 256; + static const uint32_t INITIAL_BLOCK_ALLOC_COUNT = 16; + static const uint8_t MEMORY_CLASS_SHIFT = 7; + static const uint8_t MAX_MEMORY_CLASSES = 65 - MEMORY_CLASS_SHIFT; + class Block + { + public: + VkDeviceSize offset; + VkDeviceSize size; + Block* prevPhysical; + Block* nextPhysical; + + void MarkFree() { prevFree = VMA_NULL; } + void MarkTaken() { prevFree = this; } + bool IsFree() const { return prevFree != this; } + void*& UserData() { VMA_HEAVY_ASSERT(!IsFree()); return userData; } + Block*& PrevFree() { return prevFree; } + Block*& NextFree() { VMA_HEAVY_ASSERT(IsFree()); return nextFree; } + + private: + Block* prevFree; // Address of the same block here indicates that block is taken + union + { + Block* nextFree; + void* userData; + }; + }; + + size_t m_AllocCount; + // Total number of free blocks besides null block + size_t m_BlocksFreeCount; + // Total size of free blocks excluding null block + VkDeviceSize m_BlocksFreeSize; + uint32_t m_IsFreeBitmap; + uint8_t m_MemoryClasses; + uint32_t m_InnerIsFreeBitmap[MAX_MEMORY_CLASSES]; + uint32_t m_ListsCount; /* - Protects access to m_hMemory so it is 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. + * 0: 0-3 lists for small buffers + * 1+: 0-(2^SLI-1) lists for normal buffers */ - VMA_MUTEX m_Mutex; - uint32_t m_MapCount; - void* m_pMappedData; + Block** m_FreeList; + VmaPoolAllocator m_BlockAllocator; + Block* m_NullBlock; + VmaBlockBufferImageGranularity m_GranularityHandler; + + uint8_t SizeToMemoryClass(VkDeviceSize size) const; + uint16_t SizeToSecondIndex(VkDeviceSize size, uint8_t memoryClass) const; + uint32_t GetListIndex(uint8_t memoryClass, uint16_t secondIndex) const; + uint32_t GetListIndex(VkDeviceSize size) const; + + void RemoveFreeBlock(Block* block); + void InsertFreeBlock(Block* block); + void MergeBlock(Block* block, Block* prev); + + Block* FindFreeBlock(VkDeviceSize size, uint32_t& listIndex) const; + bool CheckBlock( + Block& block, + uint32_t listIndex, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + VmaAllocationRequest* pAllocationRequest); }; -struct VmaDefragmentationMove +#ifndef _VMA_BLOCK_METADATA_TLSF_FUNCTIONS +VmaBlockMetadata_TLSF::VmaBlockMetadata_TLSF(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual) + : VmaBlockMetadata(pAllocationCallbacks, bufferImageGranularity, isVirtual), + m_AllocCount(0), + m_BlocksFreeCount(0), + m_BlocksFreeSize(0), + m_IsFreeBitmap(0), + m_MemoryClasses(0), + m_ListsCount(0), + m_FreeList(VMA_NULL), + m_BlockAllocator(pAllocationCallbacks, INITIAL_BLOCK_ALLOC_COUNT), + m_NullBlock(VMA_NULL), + m_GranularityHandler(bufferImageGranularity) {} + +VmaBlockMetadata_TLSF::~VmaBlockMetadata_TLSF() { - size_t srcBlockIndex; - size_t dstBlockIndex; - VkDeviceSize srcOffset; - VkDeviceSize dstOffset; - VkDeviceSize size; - VmaAllocation hAllocation; - VmaDeviceMemoryBlock* pSrcBlock; - VmaDeviceMemoryBlock* pDstBlock; -}; + if (m_FreeList) + vma_delete_array(GetAllocationCallbacks(), m_FreeList, m_ListsCount); + m_GranularityHandler.Destroy(GetAllocationCallbacks()); +} -class VmaDefragmentationAlgorithm; +void VmaBlockMetadata_TLSF::Init(VkDeviceSize size) +{ + VmaBlockMetadata::Init(size); + if (!IsVirtual()) + m_GranularityHandler.Init(GetAllocationCallbacks(), size); + + m_NullBlock = m_BlockAllocator.Alloc(); + m_NullBlock->size = size; + m_NullBlock->offset = 0; + m_NullBlock->prevPhysical = VMA_NULL; + m_NullBlock->nextPhysical = VMA_NULL; + m_NullBlock->MarkFree(); + m_NullBlock->NextFree() = VMA_NULL; + m_NullBlock->PrevFree() = VMA_NULL; + uint8_t memoryClass = SizeToMemoryClass(size); + uint16_t sli = SizeToSecondIndex(size, memoryClass); + m_ListsCount = (memoryClass == 0 ? 0 : (memoryClass - 1) * (1UL << SECOND_LEVEL_INDEX) + sli) + 1; + if (IsVirtual()) + m_ListsCount += 1UL << SECOND_LEVEL_INDEX; + else + m_ListsCount += 4; + + m_MemoryClasses = memoryClass + 2; + memset(m_InnerIsFreeBitmap, 0, MAX_MEMORY_CLASSES * sizeof(uint32_t)); + + m_FreeList = vma_new_array(GetAllocationCallbacks(), Block*, m_ListsCount); + memset(m_FreeList, 0, m_ListsCount * sizeof(Block*)); +} + +bool VmaBlockMetadata_TLSF::Validate() const +{ + VMA_VALIDATE(GetSumFreeSize() <= GetSize()); + + VkDeviceSize calculatedSize = m_NullBlock->size; + VkDeviceSize calculatedFreeSize = m_NullBlock->size; + size_t allocCount = 0; + size_t freeCount = 0; + + // Check integrity of free lists + for (uint32_t list = 0; list < m_ListsCount; ++list) + { + Block* block = m_FreeList[list]; + if (block != VMA_NULL) + { + VMA_VALIDATE(block->IsFree()); + VMA_VALIDATE(block->PrevFree() == VMA_NULL); + while (block->NextFree()) + { + VMA_VALIDATE(block->NextFree()->IsFree()); + VMA_VALIDATE(block->NextFree()->PrevFree() == block); + block = block->NextFree(); + } + } + } + + VkDeviceSize nextOffset = m_NullBlock->offset; + auto validateCtx = m_GranularityHandler.StartValidation(GetAllocationCallbacks(), IsVirtual()); + + VMA_VALIDATE(m_NullBlock->nextPhysical == VMA_NULL); + if (m_NullBlock->prevPhysical) + { + VMA_VALIDATE(m_NullBlock->prevPhysical->nextPhysical == m_NullBlock); + } + // Check all blocks + for (Block* prev = m_NullBlock->prevPhysical; prev != VMA_NULL; prev = prev->prevPhysical) + { + VMA_VALIDATE(prev->offset + prev->size == nextOffset); + nextOffset = prev->offset; + calculatedSize += prev->size; + + uint32_t listIndex = GetListIndex(prev->size); + if (prev->IsFree()) + { + ++freeCount; + // Check if free block belongs to free list + Block* freeBlock = m_FreeList[listIndex]; + VMA_VALIDATE(freeBlock != VMA_NULL); + + bool found = false; + do + { + if (freeBlock == prev) + found = true; + + freeBlock = freeBlock->NextFree(); + } while (!found && freeBlock != VMA_NULL); + + VMA_VALIDATE(found); + calculatedFreeSize += prev->size; + } + else + { + ++allocCount; + // Check if taken block is not on a free list + Block* freeBlock = m_FreeList[listIndex]; + while (freeBlock) + { + VMA_VALIDATE(freeBlock != prev); + freeBlock = freeBlock->NextFree(); + } + + if (!IsVirtual()) + { + VMA_VALIDATE(m_GranularityHandler.Validate(validateCtx, prev->offset, prev->size)); + } + } + + if (prev->prevPhysical) + { + VMA_VALIDATE(prev->prevPhysical->nextPhysical == prev); + } + } + + if (!IsVirtual()) + { + VMA_VALIDATE(m_GranularityHandler.FinishValidation(validateCtx)); + } + + VMA_VALIDATE(nextOffset == 0); + VMA_VALIDATE(calculatedSize == GetSize()); + VMA_VALIDATE(calculatedFreeSize == GetSumFreeSize()); + VMA_VALIDATE(allocCount == m_AllocCount); + VMA_VALIDATE(freeCount == m_BlocksFreeCount); + + return true; +} + +void VmaBlockMetadata_TLSF::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) const +{ + inoutStats.statistics.blockCount++; + inoutStats.statistics.blockBytes += GetSize(); + if (m_NullBlock->size > 0) + VmaAddDetailedStatisticsUnusedRange(inoutStats, m_NullBlock->size); + + for (Block* block = m_NullBlock->prevPhysical; block != VMA_NULL; block = block->prevPhysical) + { + if (block->IsFree()) + VmaAddDetailedStatisticsUnusedRange(inoutStats, block->size); + else + VmaAddDetailedStatisticsAllocation(inoutStats, block->size); + } +} + +void VmaBlockMetadata_TLSF::AddStatistics(VmaStatistics& inoutStats) const +{ + inoutStats.blockCount++; + inoutStats.allocationCount += (uint32_t)m_AllocCount; + inoutStats.blockBytes += GetSize(); + inoutStats.allocationBytes += GetSize() - GetSumFreeSize(); +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockMetadata_TLSF::PrintDetailedMap(class VmaJsonWriter& json) const +{ + size_t blockCount = m_AllocCount + m_BlocksFreeCount; + VmaStlAllocator allocator(GetAllocationCallbacks()); + VmaVector> blockList(blockCount, allocator); + + size_t i = blockCount; + for (Block* block = m_NullBlock->prevPhysical; block != VMA_NULL; block = block->prevPhysical) + { + blockList[--i] = block; + } + VMA_ASSERT(i == 0); + + VmaDetailedStatistics stats; + VmaClearDetailedStatistics(stats); + AddDetailedStatistics(stats); + + PrintDetailedMap_Begin(json, + stats.statistics.blockBytes - stats.statistics.allocationBytes, + stats.statistics.allocationCount, + stats.unusedRangeCount); + + for (; i < blockCount; ++i) + { + Block* block = blockList[i]; + if (block->IsFree()) + PrintDetailedMap_UnusedRange(json, block->offset, block->size); + else + PrintDetailedMap_Allocation(json, block->offset, block->size, block->UserData()); + } + if (m_NullBlock->size > 0) + PrintDetailedMap_UnusedRange(json, m_NullBlock->offset, m_NullBlock->size); + + PrintDetailedMap_End(json); +} +#endif + +bool VmaBlockMetadata_TLSF::CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(allocSize > 0 && "Cannot allocate empty block!"); + VMA_ASSERT(!upperAddress && "VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT can be used only with linear algorithm."); + + // For small granularity round up + if (!IsVirtual()) + m_GranularityHandler.RoundupAllocRequest(allocType, allocSize, allocAlignment); + + allocSize += GetDebugMargin(); + // Quick check for too small pool + if (allocSize > GetSumFreeSize()) + return false; + + // If no free blocks in pool then check only null block + if (m_BlocksFreeCount == 0) + return CheckBlock(*m_NullBlock, m_ListsCount, allocSize, allocAlignment, allocType, pAllocationRequest); + + // Round up to the next block + VkDeviceSize sizeForNextList = allocSize; + VkDeviceSize smallSizeStep = SMALL_BUFFER_SIZE / (IsVirtual() ? 1 << SECOND_LEVEL_INDEX : 4); + if (allocSize > SMALL_BUFFER_SIZE) + { + sizeForNextList += (1ULL << (VMA_BITSCAN_MSB(allocSize) - SECOND_LEVEL_INDEX)); + } + else if (allocSize > SMALL_BUFFER_SIZE - smallSizeStep) + sizeForNextList = SMALL_BUFFER_SIZE + 1; + else + sizeForNextList += smallSizeStep; + + uint32_t nextListIndex = 0; + uint32_t prevListIndex = 0; + Block* nextListBlock = VMA_NULL; + Block* prevListBlock = VMA_NULL; + + // Check blocks according to strategies + if (strategy & VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT) + { + // Quick check for larger block first + nextListBlock = FindFreeBlock(sizeForNextList, nextListIndex); + if (nextListBlock != VMA_NULL && CheckBlock(*nextListBlock, nextListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + + // If not fitted then null block + if (CheckBlock(*m_NullBlock, m_ListsCount, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + + // Null block failed, search larger bucket + while (nextListBlock) + { + if (CheckBlock(*nextListBlock, nextListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + nextListBlock = nextListBlock->NextFree(); + } + + // Failed again, check best fit bucket + prevListBlock = FindFreeBlock(allocSize, prevListIndex); + while (prevListBlock) + { + if (CheckBlock(*prevListBlock, prevListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + prevListBlock = prevListBlock->NextFree(); + } + } + else if (strategy & VMA_ALLOCATION_CREATE_STRATEGY_MIN_MEMORY_BIT) + { + // Check best fit bucket + prevListBlock = FindFreeBlock(allocSize, prevListIndex); + while (prevListBlock) + { + if (CheckBlock(*prevListBlock, prevListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + prevListBlock = prevListBlock->NextFree(); + } + + // If failed check null block + if (CheckBlock(*m_NullBlock, m_ListsCount, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + + // Check larger bucket + nextListBlock = FindFreeBlock(sizeForNextList, nextListIndex); + while (nextListBlock) + { + if (CheckBlock(*nextListBlock, nextListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + nextListBlock = nextListBlock->NextFree(); + } + } + else if (strategy & VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT ) + { + // Perform search from the start + VmaStlAllocator allocator(GetAllocationCallbacks()); + VmaVector> blockList(m_BlocksFreeCount, allocator); + + size_t i = m_BlocksFreeCount; + for (Block* block = m_NullBlock->prevPhysical; block != VMA_NULL; block = block->prevPhysical) + { + if (block->IsFree() && block->size >= allocSize) + blockList[--i] = block; + } + + for (; i < m_BlocksFreeCount; ++i) + { + Block& block = *blockList[i]; + if (CheckBlock(block, GetListIndex(block.size), allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + } + + // If failed check null block + if (CheckBlock(*m_NullBlock, m_ListsCount, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + + // Whole range searched, no more memory + return false; + } + else + { + // Check larger bucket + nextListBlock = FindFreeBlock(sizeForNextList, nextListIndex); + while (nextListBlock) + { + if (CheckBlock(*nextListBlock, nextListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + nextListBlock = nextListBlock->NextFree(); + } + + // If failed check null block + if (CheckBlock(*m_NullBlock, m_ListsCount, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + + // Check best fit bucket + prevListBlock = FindFreeBlock(allocSize, prevListIndex); + while (prevListBlock) + { + if (CheckBlock(*prevListBlock, prevListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + prevListBlock = prevListBlock->NextFree(); + } + } + + // Worst case, full search has to be done + while (++nextListIndex < m_ListsCount) + { + nextListBlock = m_FreeList[nextListIndex]; + while (nextListBlock) + { + if (CheckBlock(*nextListBlock, nextListIndex, allocSize, allocAlignment, allocType, pAllocationRequest)) + return true; + nextListBlock = nextListBlock->NextFree(); + } + } + + // No more memory sadly + return false; +} + +VkResult VmaBlockMetadata_TLSF::CheckCorruption(const void* pBlockData) +{ + for (Block* block = m_NullBlock->prevPhysical; block != VMA_NULL; block = block->prevPhysical) + { + if (!block->IsFree()) + { + if (!VmaValidateMagicValue(pBlockData, block->offset + block->size)) + { + VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); + return VK_ERROR_UNKNOWN_COPY; + } + } + } + + return VK_SUCCESS; +} + +void VmaBlockMetadata_TLSF::Alloc( + const VmaAllocationRequest& request, + VmaSuballocationType type, + void* userData) +{ + VMA_ASSERT(request.type == VmaAllocationRequestType::TLSF); + + // Get block and pop it from the free list + Block* currentBlock = (Block*)request.allocHandle; + VkDeviceSize offset = request.algorithmData; + VMA_ASSERT(currentBlock != VMA_NULL); + VMA_ASSERT(currentBlock->offset <= offset); + + if (currentBlock != m_NullBlock) + RemoveFreeBlock(currentBlock); + + VkDeviceSize debugMargin = GetDebugMargin(); + VkDeviceSize misssingAlignment = offset - currentBlock->offset; + + // Append missing alignment to prev block or create new one + if (misssingAlignment) + { + Block* prevBlock = currentBlock->prevPhysical; + VMA_ASSERT(prevBlock != VMA_NULL && "There should be no missing alignment at offset 0!"); + + if (prevBlock->IsFree() && prevBlock->size != debugMargin) + { + uint32_t oldList = GetListIndex(prevBlock->size); + prevBlock->size += misssingAlignment; + // Check if new size crosses list bucket + if (oldList != GetListIndex(prevBlock->size)) + { + prevBlock->size -= misssingAlignment; + RemoveFreeBlock(prevBlock); + prevBlock->size += misssingAlignment; + InsertFreeBlock(prevBlock); + } + else + m_BlocksFreeSize += misssingAlignment; + } + else + { + Block* newBlock = m_BlockAllocator.Alloc(); + currentBlock->prevPhysical = newBlock; + prevBlock->nextPhysical = newBlock; + newBlock->prevPhysical = prevBlock; + newBlock->nextPhysical = currentBlock; + newBlock->size = misssingAlignment; + newBlock->offset = currentBlock->offset; + newBlock->MarkTaken(); + + InsertFreeBlock(newBlock); + } + + currentBlock->size -= misssingAlignment; + currentBlock->offset += misssingAlignment; + } + + VkDeviceSize size = request.size + debugMargin; + if (currentBlock->size == size) + { + if (currentBlock == m_NullBlock) + { + // Setup new null block + m_NullBlock = m_BlockAllocator.Alloc(); + m_NullBlock->size = 0; + m_NullBlock->offset = currentBlock->offset + size; + m_NullBlock->prevPhysical = currentBlock; + m_NullBlock->nextPhysical = VMA_NULL; + m_NullBlock->MarkFree(); + m_NullBlock->PrevFree() = VMA_NULL; + m_NullBlock->NextFree() = VMA_NULL; + currentBlock->nextPhysical = m_NullBlock; + currentBlock->MarkTaken(); + } + } + else + { + VMA_ASSERT(currentBlock->size > size && "Proper block already found, shouldn't find smaller one!"); + + // Create new free block + Block* newBlock = m_BlockAllocator.Alloc(); + newBlock->size = currentBlock->size - size; + newBlock->offset = currentBlock->offset + size; + newBlock->prevPhysical = currentBlock; + newBlock->nextPhysical = currentBlock->nextPhysical; + currentBlock->nextPhysical = newBlock; + currentBlock->size = size; + + if (currentBlock == m_NullBlock) + { + m_NullBlock = newBlock; + m_NullBlock->MarkFree(); + m_NullBlock->NextFree() = VMA_NULL; + m_NullBlock->PrevFree() = VMA_NULL; + currentBlock->MarkTaken(); + } + else + { + newBlock->nextPhysical->prevPhysical = newBlock; + newBlock->MarkTaken(); + InsertFreeBlock(newBlock); + } + } + currentBlock->UserData() = userData; + + if (debugMargin > 0) + { + currentBlock->size -= debugMargin; + Block* newBlock = m_BlockAllocator.Alloc(); + newBlock->size = debugMargin; + newBlock->offset = currentBlock->offset + currentBlock->size; + newBlock->prevPhysical = currentBlock; + newBlock->nextPhysical = currentBlock->nextPhysical; + newBlock->MarkTaken(); + currentBlock->nextPhysical->prevPhysical = newBlock; + currentBlock->nextPhysical = newBlock; + InsertFreeBlock(newBlock); + } + + if (!IsVirtual()) + m_GranularityHandler.AllocPages((uint8_t)(uintptr_t)request.customData, + currentBlock->offset, currentBlock->size); + ++m_AllocCount; +} + +void VmaBlockMetadata_TLSF::Free(VmaAllocHandle allocHandle) +{ + Block* block = (Block*)allocHandle; + Block* next = block->nextPhysical; + VMA_ASSERT(!block->IsFree() && "Block is already free!"); + + if (!IsVirtual()) + m_GranularityHandler.FreePages(block->offset, block->size); + --m_AllocCount; + + VkDeviceSize debugMargin = GetDebugMargin(); + if (debugMargin > 0) + { + RemoveFreeBlock(next); + MergeBlock(next, block); + block = next; + next = next->nextPhysical; + } + + // Try merging + Block* prev = block->prevPhysical; + if (prev != VMA_NULL && prev->IsFree() && prev->size != debugMargin) + { + RemoveFreeBlock(prev); + MergeBlock(block, prev); + } + + if (!next->IsFree()) + InsertFreeBlock(block); + else if (next == m_NullBlock) + MergeBlock(m_NullBlock, block); + else + { + RemoveFreeBlock(next); + MergeBlock(next, block); + InsertFreeBlock(next); + } +} + +void VmaBlockMetadata_TLSF::GetAllocationInfo(VmaAllocHandle allocHandle, VmaVirtualAllocationInfo& outInfo) +{ + Block* block = (Block*)allocHandle; + VMA_ASSERT(!block->IsFree() && "Cannot get allocation info for free block!"); + outInfo.offset = block->offset; + outInfo.size = block->size; + outInfo.pUserData = block->UserData(); +} + +void* VmaBlockMetadata_TLSF::GetAllocationUserData(VmaAllocHandle allocHandle) const +{ + Block* block = (Block*)allocHandle; + VMA_ASSERT(!block->IsFree() && "Cannot get user data for free block!"); + return block->UserData(); +} + +VmaAllocHandle VmaBlockMetadata_TLSF::GetAllocationListBegin() const +{ + if (m_AllocCount == 0) + return VK_NULL_HANDLE; + + for (Block* block = m_NullBlock->prevPhysical; block; block = block->prevPhysical) + { + if (!block->IsFree()) + return (VmaAllocHandle)block; + } + VMA_ASSERT(false && "If m_AllocCount > 0 then should find any allocation!"); + return VK_NULL_HANDLE; +} + +VmaAllocHandle VmaBlockMetadata_TLSF::GetNextAllocation(VmaAllocHandle prevAlloc) const +{ + Block* startBlock = (Block*)prevAlloc; + VMA_ASSERT(!startBlock->IsFree() && "Incorrect block!"); + + for (Block* block = startBlock->prevPhysical; block; block = block->prevPhysical) + { + if (!block->IsFree()) + return (VmaAllocHandle)block; + } + return VK_NULL_HANDLE; +} + +VkDeviceSize VmaBlockMetadata_TLSF::GetNextFreeRegionSize(VmaAllocHandle alloc) const +{ + Block* block = (Block*)alloc; + VMA_ASSERT(!block->IsFree() && "Incorrect block!"); + + if (block->prevPhysical) + return block->prevPhysical->IsFree() ? block->prevPhysical->size : 0; + return 0; +} + +void VmaBlockMetadata_TLSF::Clear() +{ + m_AllocCount = 0; + m_BlocksFreeCount = 0; + m_BlocksFreeSize = 0; + m_IsFreeBitmap = 0; + m_NullBlock->offset = 0; + m_NullBlock->size = GetSize(); + Block* block = m_NullBlock->prevPhysical; + m_NullBlock->prevPhysical = VMA_NULL; + while (block) + { + Block* prev = block->prevPhysical; + m_BlockAllocator.Free(block); + block = prev; + } + memset(m_FreeList, 0, m_ListsCount * sizeof(Block*)); + memset(m_InnerIsFreeBitmap, 0, m_MemoryClasses * sizeof(uint32_t)); + m_GranularityHandler.Clear(); +} + +void VmaBlockMetadata_TLSF::SetAllocationUserData(VmaAllocHandle allocHandle, void* userData) +{ + Block* block = (Block*)allocHandle; + VMA_ASSERT(!block->IsFree() && "Trying to set user data for not allocated block!"); + block->UserData() = userData; +} + +void VmaBlockMetadata_TLSF::DebugLogAllAllocations() const +{ + for (Block* block = m_NullBlock->prevPhysical; block != VMA_NULL; block = block->prevPhysical) + if (!block->IsFree()) + DebugLogAllocation(block->offset, block->size, block->UserData()); +} + +uint8_t VmaBlockMetadata_TLSF::SizeToMemoryClass(VkDeviceSize size) const +{ + if (size > SMALL_BUFFER_SIZE) + return VMA_BITSCAN_MSB(size) - MEMORY_CLASS_SHIFT; + return 0; +} + +uint16_t VmaBlockMetadata_TLSF::SizeToSecondIndex(VkDeviceSize size, uint8_t memoryClass) const +{ + if (memoryClass == 0) + { + if (IsVirtual()) + return static_cast((size - 1) / 8); + else + return static_cast((size - 1) / 64); + } + return static_cast((size >> (memoryClass + MEMORY_CLASS_SHIFT - SECOND_LEVEL_INDEX)) ^ (1U << SECOND_LEVEL_INDEX)); +} + +uint32_t VmaBlockMetadata_TLSF::GetListIndex(uint8_t memoryClass, uint16_t secondIndex) const +{ + if (memoryClass == 0) + return secondIndex; + + const uint32_t index = static_cast(memoryClass - 1) * (1 << SECOND_LEVEL_INDEX) + secondIndex; + if (IsVirtual()) + return index + (1 << SECOND_LEVEL_INDEX); + else + return index + 4; +} + +uint32_t VmaBlockMetadata_TLSF::GetListIndex(VkDeviceSize size) const +{ + uint8_t memoryClass = SizeToMemoryClass(size); + return GetListIndex(memoryClass, SizeToSecondIndex(size, memoryClass)); +} + +void VmaBlockMetadata_TLSF::RemoveFreeBlock(Block* block) +{ + VMA_ASSERT(block != m_NullBlock); + VMA_ASSERT(block->IsFree()); + + if (block->NextFree() != VMA_NULL) + block->NextFree()->PrevFree() = block->PrevFree(); + if (block->PrevFree() != VMA_NULL) + block->PrevFree()->NextFree() = block->NextFree(); + else + { + uint8_t memClass = SizeToMemoryClass(block->size); + uint16_t secondIndex = SizeToSecondIndex(block->size, memClass); + uint32_t index = GetListIndex(memClass, secondIndex); + VMA_ASSERT(m_FreeList[index] == block); + m_FreeList[index] = block->NextFree(); + if (block->NextFree() == VMA_NULL) + { + m_InnerIsFreeBitmap[memClass] &= ~(1U << secondIndex); + if (m_InnerIsFreeBitmap[memClass] == 0) + m_IsFreeBitmap &= ~(1UL << memClass); + } + } + block->MarkTaken(); + block->UserData() = VMA_NULL; + --m_BlocksFreeCount; + m_BlocksFreeSize -= block->size; +} + +void VmaBlockMetadata_TLSF::InsertFreeBlock(Block* block) +{ + VMA_ASSERT(block != m_NullBlock); + VMA_ASSERT(!block->IsFree() && "Cannot insert block twice!"); + + uint8_t memClass = SizeToMemoryClass(block->size); + uint16_t secondIndex = SizeToSecondIndex(block->size, memClass); + uint32_t index = GetListIndex(memClass, secondIndex); + VMA_ASSERT(index < m_ListsCount); + block->PrevFree() = VMA_NULL; + block->NextFree() = m_FreeList[index]; + m_FreeList[index] = block; + if (block->NextFree() != VMA_NULL) + block->NextFree()->PrevFree() = block; + else + { + m_InnerIsFreeBitmap[memClass] |= 1U << secondIndex; + m_IsFreeBitmap |= 1UL << memClass; + } + ++m_BlocksFreeCount; + m_BlocksFreeSize += block->size; +} + +void VmaBlockMetadata_TLSF::MergeBlock(Block* block, Block* prev) +{ + VMA_ASSERT(block->prevPhysical == prev && "Cannot merge seperate physical regions!"); + VMA_ASSERT(!prev->IsFree() && "Cannot merge block that belongs to free list!"); + + block->offset = prev->offset; + block->size += prev->size; + block->prevPhysical = prev->prevPhysical; + if (block->prevPhysical) + block->prevPhysical->nextPhysical = block; + m_BlockAllocator.Free(prev); +} + +VmaBlockMetadata_TLSF::Block* VmaBlockMetadata_TLSF::FindFreeBlock(VkDeviceSize size, uint32_t& listIndex) const +{ + uint8_t memoryClass = SizeToMemoryClass(size); + uint32_t innerFreeMap = m_InnerIsFreeBitmap[memoryClass] & (~0U << SizeToSecondIndex(size, memoryClass)); + if (!innerFreeMap) + { + // Check higher levels for avaiable blocks + uint32_t freeMap = m_IsFreeBitmap & (~0UL << (memoryClass + 1)); + if (!freeMap) + return VMA_NULL; // No more memory avaible + + // Find lowest free region + memoryClass = VMA_BITSCAN_LSB(freeMap); + innerFreeMap = m_InnerIsFreeBitmap[memoryClass]; + VMA_ASSERT(innerFreeMap != 0); + } + // Find lowest free subregion + listIndex = GetListIndex(memoryClass, VMA_BITSCAN_LSB(innerFreeMap)); + VMA_ASSERT(m_FreeList[listIndex]); + return m_FreeList[listIndex]; +} + +bool VmaBlockMetadata_TLSF::CheckBlock( + Block& block, + uint32_t listIndex, + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + VmaAllocationRequest* pAllocationRequest) +{ + VMA_ASSERT(block.IsFree() && "Block is already taken!"); + + VkDeviceSize alignedOffset = VmaAlignUp(block.offset, allocAlignment); + if (block.size < allocSize + alignedOffset - block.offset) + return false; + + // Check for granularity conflicts + if (!IsVirtual() && + m_GranularityHandler.CheckConflictAndAlignUp(alignedOffset, allocSize, block.offset, block.size, allocType)) + return false; + + // Alloc successful + pAllocationRequest->type = VmaAllocationRequestType::TLSF; + pAllocationRequest->allocHandle = (VmaAllocHandle)█ + pAllocationRequest->size = allocSize - GetDebugMargin(); + pAllocationRequest->customData = (void*)allocType; + pAllocationRequest->algorithmData = alignedOffset; + + // Place block at the start of list if it's normal block + if (listIndex != m_ListsCount && block.PrevFree()) + { + block.PrevFree()->NextFree() = block.NextFree(); + if (block.NextFree()) + block.NextFree()->PrevFree() = block.PrevFree(); + block.PrevFree() = VMA_NULL; + block.NextFree() = m_FreeList[listIndex]; + m_FreeList[listIndex] = █ + if (block.NextFree()) + block.NextFree()->PrevFree() = █ + } + + return true; +} +#endif // _VMA_BLOCK_METADATA_TLSF_FUNCTIONS +#endif // _VMA_BLOCK_METADATA_TLSF + +#ifndef _VMA_BLOCK_VECTOR /* Sequence of VmaDeviceMemoryBlock. Represents memory blocks allocated for a specific Vulkan memory type. Synchronized internally with a mutex. */ -struct VmaBlockVector +class VmaBlockVector { + friend struct VmaDefragmentationContext_T; VMA_CLASS_NO_COPY(VmaBlockVector) public: VmaBlockVector( @@ -5495,7 +10874,6 @@ public: size_t minBlockCount, size_t maxBlockCount, VkDeviceSize bufferImageGranularity, - uint32_t frameInUseCount, bool explicitBlockSize, uint32_t algorithm, float priority, @@ -5503,24 +10881,29 @@ public: void* pMemoryAllocateNext); ~VmaBlockVector(); - VkResult CreateMinBlocks(); - VmaAllocator GetAllocator() const { return m_hAllocator; } VmaPool GetParentPool() const { return m_hParentPool; } bool IsCustomPool() const { return m_hParentPool != VMA_NULL; } 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; } + bool HasExplicitBlockSize() const { return m_ExplicitBlockSize; } + float GetPriority() const { return m_Priority; } + const void* GetAllocationNextPtr() const { return m_pMemoryAllocateNext; } + // To be used only while the m_Mutex is locked. Used during defragmentation. + size_t GetBlockCount() const { return m_Blocks.size(); } + // To be used only while the m_Mutex is locked. Used during defragmentation. + VmaDeviceMemoryBlock* GetBlock(size_t index) const { return m_Blocks[index]; } + VMA_RW_MUTEX &GetMutex() { return m_Mutex; } - void GetPoolStats(VmaPoolStats* pStats); - + VkResult CreateMinBlocks(); + void AddStatistics(VmaStatistics& inoutStats); + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats); bool IsEmpty(); bool IsCorruptionDetectionEnabled() const; VkResult Allocate( - uint32_t currentFrameIndex, VkDeviceSize size, VkDeviceSize alignment, const VmaAllocationCreateInfo& createInfo, @@ -5530,49 +10913,13 @@ public: void Free(const 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, VmaDefragmentationFlags flags, - VkDeviceSize& maxCpuBytesToMove, uint32_t& maxCpuAllocationsToMove, - VkDeviceSize& maxGpuBytesToMove, uint32_t& maxGpuAllocationsToMove, - VkCommandBuffer commandBuffer); - void DefragmentationEnd( - class VmaBlockVectorDefragmentationContext* pCtx, - uint32_t flags, - VmaDefragmentationStats* pStats); - - uint32_t ProcessDefragmentations( - class VmaBlockVectorDefragmentationContext *pCtx, - VmaDefragmentationPassMoveInfo* pMove, uint32_t maxMoves); - - void CommitDefragmentations( - 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 VmaPool m_hParentPool; const uint32_t m_MemoryTypeIndex; @@ -5580,42 +10927,37 @@ private: const size_t m_MinBlockCount; const size_t m_MaxBlockCount; const VkDeviceSize m_BufferImageGranularity; - const uint32_t m_FrameInUseCount; const bool m_ExplicitBlockSize; const uint32_t m_Algorithm; const float m_Priority; const VkDeviceSize m_MinAllocationAlignment; + void* const m_pMemoryAllocateNext; VMA_RW_MUTEX m_Mutex; - - /* There can be at most one allocation that is completely empty (except when minBlockCount > 0) - - a hysteresis to avoid pessimistic case of alternating creation and destruction of a VkDeviceMemory. */ - bool m_HasEmptyBlock; // Incrementally sorted by sumFreeSize, ascending. - VmaVector< VmaDeviceMemoryBlock*, VmaStlAllocator > m_Blocks; + VmaVector> m_Blocks; uint32_t m_NextBlockId; + bool m_IncrementalSort = true; + + void SetIncrementalSort(bool val) { m_IncrementalSort = val; } 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(); + void SortByFreeSize(); VkResult AllocatePage( - 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, - uint32_t currentFrameIndex, VkDeviceSize size, VkDeviceSize alignment, VmaAllocationCreateFlags allocFlags, @@ -5624,32 +10966,114 @@ private: uint32_t strategy, VmaAllocation* pAllocation); + VkResult CommitAllocationRequest( + VmaAllocationRequest& allocRequest, + VmaDeviceMemoryBlock* pBlock, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation); + VkResult CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIndex); - - // Saves result to pCtx->res. - void ApplyDefragmentationMovesCpu( - class VmaBlockVectorDefragmentationContext* pDefragCtx, - const VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves); - // Saves result to pCtx->res. - void ApplyDefragmentationMovesGpu( - class VmaBlockVectorDefragmentationContext* pDefragCtx, - VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves, - VkCommandBuffer commandBuffer); - - /* - Used during defragmentation. pDefragmentationStats is optional. It is in/out - - updated with new data. - */ - void FreeEmptyBlocks(VmaDefragmentationStats* pDefragmentationStats); - - void UpdateHasEmptyBlock(); + bool HasEmptyBlock(); }; +#endif // _VMA_BLOCK_VECTOR +#ifndef _VMA_DEFRAGMENTATION_CONTEXT +struct VmaDefragmentationContext_T +{ + VMA_CLASS_NO_COPY(VmaDefragmentationContext_T) +public: + VmaDefragmentationContext_T( + VmaAllocator hAllocator, + const VmaDefragmentationInfo& info); + ~VmaDefragmentationContext_T(); + + void GetStats(VmaDefragmentationStats& outStats) { outStats = m_GlobalStats; } + + VkResult DefragmentPassBegin(VmaDefragmentationPassMoveInfo& moveInfo); + VkResult DefragmentPassEnd(VmaDefragmentationPassMoveInfo& moveInfo); + +private: + // Max number of allocations to ignore due to size constraints before ending single pass + static const uint8_t MAX_ALLOCS_TO_IGNORE = 16; + enum class CounterStatus { Pass, Ignore, End }; + + struct FragmentedBlock + { + uint32_t data; + VmaDeviceMemoryBlock* block; + }; + struct StateBalanced + { + VkDeviceSize avgFreeSize = 0; + VkDeviceSize avgAllocSize = UINT64_MAX; + }; + struct StateExtensive + { + enum class Operation : uint8_t + { + FindFreeBlockBuffer, FindFreeBlockTexture, FindFreeBlockAll, + MoveBuffers, MoveTextures, MoveAll, + Cleanup, Done + }; + + Operation operation = Operation::FindFreeBlockTexture; + size_t firstFreeBlock = SIZE_MAX; + }; + struct MoveAllocationData + { + VkDeviceSize size; + VkDeviceSize alignment; + VmaSuballocationType type; + VmaAllocationCreateFlags flags; + VmaDefragmentationMove move = {}; + }; + + const VkDeviceSize m_MaxPassBytes; + const uint32_t m_MaxPassAllocations; + + VmaStlAllocator m_MoveAllocator; + VmaVector> m_Moves; + + uint8_t m_IgnoredAllocs = 0; + uint32_t m_Algorithm; + uint32_t m_BlockVectorCount; + VmaBlockVector* m_PoolBlockVector; + VmaBlockVector** m_pBlockVectors; + size_t m_ImmovableBlockCount = 0; + VmaDefragmentationStats m_GlobalStats = { 0 }; + VmaDefragmentationStats m_PassStats = { 0 }; + void* m_AlgorithmState = VMA_NULL; + + static MoveAllocationData GetMoveData(VmaAllocHandle handle, VmaBlockMetadata* metadata); + CounterStatus CheckCounters(VkDeviceSize bytes); + bool IncrementCounters(VkDeviceSize bytes); + bool ReallocWithinBlock(VmaBlockVector& vector, VmaDeviceMemoryBlock* block); + bool AllocInOtherBlock(size_t start, size_t end, MoveAllocationData& data, VmaBlockVector& vector); + + bool ComputeDefragmentation(VmaBlockVector& vector, size_t index); + bool ComputeDefragmentation_Fast(VmaBlockVector& vector); + bool ComputeDefragmentation_Balanced(VmaBlockVector& vector, size_t index, bool update); + bool ComputeDefragmentation_Full(VmaBlockVector& vector); + bool ComputeDefragmentation_Extensive(VmaBlockVector& vector, size_t index); + + void UpdateVectorStatistics(VmaBlockVector& vector, StateBalanced& state); + bool MoveDataToFreeBlocks(VmaSuballocationType currentType, + VmaBlockVector& vector, size_t firstFreeBlock, + bool& texturePresent, bool& bufferPresent, bool& otherPresent); +}; +#endif // _VMA_DEFRAGMENTATION_CONTEXT + +#ifndef _VMA_POOL_T struct VmaPool_T { + friend struct VmaPoolListItemTraits; VMA_CLASS_NO_COPY(VmaPool_T) public: VmaBlockVector m_BlockVector; + VmaDedicatedAllocationList m_DedicatedAllocations; VmaPool_T( VmaAllocator hAllocator, @@ -5672,616 +11096,24 @@ private: char* m_Name; VmaPool_T* m_PrevPool = VMA_NULL; VmaPool_T* m_NextPool = VMA_NULL; - friend struct VmaPoolListItemTraits; }; struct VmaPoolListItemTraits { typedef VmaPool_T ItemType; + static ItemType* GetPrev(const ItemType* item) { return item->m_PrevPool; } static ItemType* GetNext(const ItemType* item) { return item->m_NextPool; } static ItemType*& AccessPrev(ItemType* item) { return item->m_PrevPool; } static ItemType*& AccessNext(ItemType* item) { return item->m_NextPool; } }; +#endif // _VMA_POOL_T -/* -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 >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - VmaDefragmentationFlags flags) = 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 >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - VmaDefragmentationFlags flags); - - 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 > 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 > BlockInfoVector; - BlockInfoVector m_Blocks; - - VkResult DefragmentRound( - VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - bool freeOldAllocations); - - 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 >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - VmaDefragmentationFlags flags); - - 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 > 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; -}; - -class VmaBlockVectorDefragmentationContext -{ - VMA_CLASS_NO_COPY(VmaBlockVectorDefragmentationContext) -public: - VkResult res; - bool mutexLocked; - VmaVector< VmaBlockDefragmentationContext, VmaStlAllocator > blockContexts; - VmaVector< VmaDefragmentationMove, VmaStlAllocator > defragmentationMoves; - uint32_t defragmentationMovesProcessed; - uint32_t defragmentationMovesCommitted; - bool hasDefragmentationPlan; - - VmaBlockVectorDefragmentationContext( - VmaAllocator hAllocator, - VmaPool hCustomPool, // Optional. - VmaBlockVector* pBlockVector, - uint32_t currFrameIndex); - ~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, VmaDefragmentationFlags flags); - -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; - // Owner of this object. - VmaDefragmentationAlgorithm* m_pAlgorithm; - - struct AllocInfo - { - VmaAllocation hAlloc; - VkBool32* pChanged; - }; - // Used between constructor and Begin. - VmaVector< AllocInfo, VmaStlAllocator > 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, const VmaPool* pPools); - void AddAllocations( - uint32_t allocationCount, - const 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 occurred and object can be destroyed immediately. - */ - VkResult Defragment( - VkDeviceSize maxCpuBytesToMove, uint32_t maxCpuAllocationsToMove, - VkDeviceSize maxGpuBytesToMove, uint32_t maxGpuAllocationsToMove, - VkCommandBuffer commandBuffer, VmaDefragmentationStats* pStats, VmaDefragmentationFlags flags); - - VkResult DefragmentPassBegin(VmaDefragmentationPassInfo* pInfo); - VkResult DefragmentPassEnd(); - -private: - const VmaAllocator m_hAllocator; - const uint32_t m_CurrFrameIndex; - const uint32_t m_Flags; - VmaDefragmentationStats* const m_pStats; - - VkDeviceSize m_MaxCpuBytesToMove; - uint32_t m_MaxCpuAllocationsToMove; - VkDeviceSize m_MaxGpuBytesToMove; - uint32_t m_MaxGpuAllocationsToMove; - - // Owner of these objects. - VmaBlockVectorDefragmentationContext* m_DefaultPoolContexts[VK_MAX_MEMORY_TYPES]; - // Owner of these objects. - VmaVector< VmaBlockVectorDefragmentationContext*, VmaStlAllocator > 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, - uint32_t vulkanApiVersion, - bool dedicatedAllocationExtensionEnabled, - bool bindMemory2ExtensionEnabled, - bool memoryBudgetExtensionEnabled, - bool deviceCoherentMemoryExtensionEnabled); - ~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 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); - void RecordSetPoolName(uint32_t frameIndex, - VmaPool pool, - const char* name); - -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; - std::chrono::time_point m_RecordingStartTime; - - void GetBasicParams(CallParams& outParams); - - // T must be a pointer type, e.g. VmaAllocation, VmaPool. - template - 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 - -/* -Thread-safe wrapper over VmaPoolAllocator free list, for allocation of VmaAllocation_T objects. -*/ -class VmaAllocationObjectAllocator -{ - VMA_CLASS_NO_COPY(VmaAllocationObjectAllocator) -public: - VmaAllocationObjectAllocator(const VkAllocationCallbacks* pAllocationCallbacks); - - template VmaAllocation Allocate(Types&&... args); - void Free(VmaAllocation hAlloc); - -private: - VMA_MUTEX m_Mutex; - VmaPoolAllocator m_Allocator; -}; - +#ifndef _VMA_CURRENT_BUDGET_DATA struct VmaCurrentBudgetData { + VMA_ATOMIC_UINT32 m_BlockCount[VK_MAX_MEMORY_HEAPS]; + VMA_ATOMIC_UINT32 m_AllocationCount[VK_MAX_MEMORY_HEAPS]; VMA_ATOMIC_UINT64 m_BlockBytes[VK_MAX_MEMORY_HEAPS]; VMA_ATOMIC_UINT64 m_AllocationBytes[VK_MAX_MEMORY_HEAPS]; @@ -6291,44 +11123,229 @@ struct VmaCurrentBudgetData uint64_t m_VulkanUsage[VK_MAX_MEMORY_HEAPS]; uint64_t m_VulkanBudget[VK_MAX_MEMORY_HEAPS]; uint64_t m_BlockBytesAtBudgetFetch[VK_MAX_MEMORY_HEAPS]; -#endif // #if VMA_MEMORY_BUDGET +#endif // VMA_MEMORY_BUDGET - VmaCurrentBudgetData() - { - for(uint32_t heapIndex = 0; heapIndex < VK_MAX_MEMORY_HEAPS; ++heapIndex) - { - m_BlockBytes[heapIndex] = 0; - m_AllocationBytes[heapIndex] = 0; -#if VMA_MEMORY_BUDGET - m_VulkanUsage[heapIndex] = 0; - m_VulkanBudget[heapIndex] = 0; - m_BlockBytesAtBudgetFetch[heapIndex] = 0; -#endif - } + VmaCurrentBudgetData(); -#if VMA_MEMORY_BUDGET - m_OperationsSinceBudgetFetch = 0; -#endif - } - - void AddAllocation(uint32_t heapIndex, VkDeviceSize allocationSize) - { - m_AllocationBytes[heapIndex] += allocationSize; -#if VMA_MEMORY_BUDGET - ++m_OperationsSinceBudgetFetch; -#endif - } - - void RemoveAllocation(uint32_t heapIndex, VkDeviceSize allocationSize) - { - VMA_ASSERT(m_AllocationBytes[heapIndex] >= allocationSize); // DELME - m_AllocationBytes[heapIndex] -= allocationSize; -#if VMA_MEMORY_BUDGET - ++m_OperationsSinceBudgetFetch; -#endif - } + void AddAllocation(uint32_t heapIndex, VkDeviceSize allocationSize); + void RemoveAllocation(uint32_t heapIndex, VkDeviceSize allocationSize); }; +#ifndef _VMA_CURRENT_BUDGET_DATA_FUNCTIONS +VmaCurrentBudgetData::VmaCurrentBudgetData() +{ + for (uint32_t heapIndex = 0; heapIndex < VK_MAX_MEMORY_HEAPS; ++heapIndex) + { + m_BlockCount[heapIndex] = 0; + m_AllocationCount[heapIndex] = 0; + m_BlockBytes[heapIndex] = 0; + m_AllocationBytes[heapIndex] = 0; +#if VMA_MEMORY_BUDGET + m_VulkanUsage[heapIndex] = 0; + m_VulkanBudget[heapIndex] = 0; + m_BlockBytesAtBudgetFetch[heapIndex] = 0; +#endif + } + +#if VMA_MEMORY_BUDGET + m_OperationsSinceBudgetFetch = 0; +#endif +} + +void VmaCurrentBudgetData::AddAllocation(uint32_t heapIndex, VkDeviceSize allocationSize) +{ + m_AllocationBytes[heapIndex] += allocationSize; + ++m_AllocationCount[heapIndex]; +#if VMA_MEMORY_BUDGET + ++m_OperationsSinceBudgetFetch; +#endif +} + +void VmaCurrentBudgetData::RemoveAllocation(uint32_t heapIndex, VkDeviceSize allocationSize) +{ + VMA_ASSERT(m_AllocationBytes[heapIndex] >= allocationSize); + m_AllocationBytes[heapIndex] -= allocationSize; + VMA_ASSERT(m_AllocationCount[heapIndex] > 0); + --m_AllocationCount[heapIndex]; +#if VMA_MEMORY_BUDGET + ++m_OperationsSinceBudgetFetch; +#endif +} +#endif // _VMA_CURRENT_BUDGET_DATA_FUNCTIONS +#endif // _VMA_CURRENT_BUDGET_DATA + +#ifndef _VMA_ALLOCATION_OBJECT_ALLOCATOR +/* +Thread-safe wrapper over VmaPoolAllocator free list, for allocation of VmaAllocation_T objects. +*/ +class VmaAllocationObjectAllocator +{ + VMA_CLASS_NO_COPY(VmaAllocationObjectAllocator) +public: + VmaAllocationObjectAllocator(const VkAllocationCallbacks* pAllocationCallbacks) + : m_Allocator(pAllocationCallbacks, 1024) {} + + template VmaAllocation Allocate(Types&&... args); + void Free(VmaAllocation hAlloc); + +private: + VMA_MUTEX m_Mutex; + VmaPoolAllocator m_Allocator; +}; + +template +VmaAllocation VmaAllocationObjectAllocator::Allocate(Types&&... args) +{ + VmaMutexLock mutexLock(m_Mutex); + return m_Allocator.Alloc(std::forward(args)...); +} + +void VmaAllocationObjectAllocator::Free(VmaAllocation hAlloc) +{ + VmaMutexLock mutexLock(m_Mutex); + m_Allocator.Free(hAlloc); +} +#endif // _VMA_ALLOCATION_OBJECT_ALLOCATOR + +#ifndef _VMA_VIRTUAL_BLOCK_T +struct VmaVirtualBlock_T +{ + VMA_CLASS_NO_COPY(VmaVirtualBlock_T) +public: + const bool m_AllocationCallbacksSpecified; + const VkAllocationCallbacks m_AllocationCallbacks; + + VmaVirtualBlock_T(const VmaVirtualBlockCreateInfo& createInfo); + ~VmaVirtualBlock_T(); + + VkResult Init() { return VK_SUCCESS; } + bool IsEmpty() const { return m_Metadata->IsEmpty(); } + void Free(VmaVirtualAllocation allocation) { m_Metadata->Free((VmaAllocHandle)allocation); } + void SetAllocationUserData(VmaVirtualAllocation allocation, void* userData) { m_Metadata->SetAllocationUserData((VmaAllocHandle)allocation, userData); } + void Clear() { m_Metadata->Clear(); } + + const VkAllocationCallbacks* GetAllocationCallbacks() const; + void GetAllocationInfo(VmaVirtualAllocation allocation, VmaVirtualAllocationInfo& outInfo); + VkResult Allocate(const VmaVirtualAllocationCreateInfo& createInfo, VmaVirtualAllocation& outAllocation, + VkDeviceSize* outOffset); + void GetStatistics(VmaStatistics& outStats) const; + void CalculateDetailedStatistics(VmaDetailedStatistics& outStats) const; +#if VMA_STATS_STRING_ENABLED + void BuildStatsString(bool detailedMap, VmaStringBuilder& sb) const; +#endif + +private: + VmaBlockMetadata* m_Metadata; +}; + +#ifndef _VMA_VIRTUAL_BLOCK_T_FUNCTIONS +VmaVirtualBlock_T::VmaVirtualBlock_T(const VmaVirtualBlockCreateInfo& createInfo) + : m_AllocationCallbacksSpecified(createInfo.pAllocationCallbacks != VMA_NULL), + m_AllocationCallbacks(createInfo.pAllocationCallbacks != VMA_NULL ? *createInfo.pAllocationCallbacks : VmaEmptyAllocationCallbacks) +{ + const uint32_t algorithm = createInfo.flags & VMA_VIRTUAL_BLOCK_CREATE_ALGORITHM_MASK; + switch (algorithm) + { + default: + VMA_ASSERT(0); + case 0: + m_Metadata = vma_new(GetAllocationCallbacks(), VmaBlockMetadata_TLSF)(VK_NULL_HANDLE, 1, true); + break; + case VMA_VIRTUAL_BLOCK_CREATE_LINEAR_ALGORITHM_BIT: + m_Metadata = vma_new(GetAllocationCallbacks(), VmaBlockMetadata_Linear)(VK_NULL_HANDLE, 1, true); + break; + } + + m_Metadata->Init(createInfo.size); +} + +VmaVirtualBlock_T::~VmaVirtualBlock_T() +{ + // Define macro VMA_DEBUG_LOG to receive the list of the unfreed allocations + if (!m_Metadata->IsEmpty()) + m_Metadata->DebugLogAllAllocations(); + // This is the most important assert in the entire library. + // Hitting it means you have some memory leak - unreleased virtual allocations. + VMA_ASSERT(m_Metadata->IsEmpty() && "Some virtual allocations were not freed before destruction of this virtual block!"); + + vma_delete(GetAllocationCallbacks(), m_Metadata); +} + +const VkAllocationCallbacks* VmaVirtualBlock_T::GetAllocationCallbacks() const +{ + return m_AllocationCallbacksSpecified ? &m_AllocationCallbacks : VMA_NULL; +} + +void VmaVirtualBlock_T::GetAllocationInfo(VmaVirtualAllocation allocation, VmaVirtualAllocationInfo& outInfo) +{ + m_Metadata->GetAllocationInfo((VmaAllocHandle)allocation, outInfo); +} + +VkResult VmaVirtualBlock_T::Allocate(const VmaVirtualAllocationCreateInfo& createInfo, VmaVirtualAllocation& outAllocation, + VkDeviceSize* outOffset) +{ + VmaAllocationRequest request = {}; + if (m_Metadata->CreateAllocationRequest( + createInfo.size, // allocSize + VMA_MAX(createInfo.alignment, (VkDeviceSize)1), // allocAlignment + (createInfo.flags & VMA_VIRTUAL_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0, // upperAddress + VMA_SUBALLOCATION_TYPE_UNKNOWN, // allocType - unimportant + createInfo.flags & VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MASK, // strategy + &request)) + { + m_Metadata->Alloc(request, + VMA_SUBALLOCATION_TYPE_UNKNOWN, // type - unimportant + createInfo.pUserData); + outAllocation = (VmaVirtualAllocation)request.allocHandle; + if(outOffset) + *outOffset = m_Metadata->GetAllocationOffset(request.allocHandle); + return VK_SUCCESS; + } + outAllocation = (VmaVirtualAllocation)VK_NULL_HANDLE; + if (outOffset) + *outOffset = UINT64_MAX; + return VK_ERROR_OUT_OF_DEVICE_MEMORY; +} + +void VmaVirtualBlock_T::GetStatistics(VmaStatistics& outStats) const +{ + VmaClearStatistics(outStats); + m_Metadata->AddStatistics(outStats); +} + +void VmaVirtualBlock_T::CalculateDetailedStatistics(VmaDetailedStatistics& outStats) const +{ + VmaClearDetailedStatistics(outStats); + m_Metadata->AddDetailedStatistics(outStats); +} + +#if VMA_STATS_STRING_ENABLED +void VmaVirtualBlock_T::BuildStatsString(bool detailedMap, VmaStringBuilder& sb) const +{ + VmaJsonWriter json(GetAllocationCallbacks(), sb); + json.BeginObject(); + + VmaDetailedStatistics stats; + CalculateDetailedStatistics(stats); + + json.WriteString("Stats"); + VmaPrintDetailedStatistics(json, stats); + + if (detailedMap) + { + json.WriteString("Details"); + json.BeginObject(); + m_Metadata->PrintDetailedMap(json); + json.EndObject(); + } + + json.EndObject(); +} +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_VIRTUAL_BLOCK_T_FUNCTIONS +#endif // _VMA_VIRTUAL_BLOCK_T + + // Main allocator object. struct VmaAllocator_T { @@ -6357,10 +11374,7 @@ public: // Default pools. VmaBlockVector* m_pBlockVectors[VK_MAX_MEMORY_TYPES]; - - typedef VmaIntrusiveLinkedList DedicatedAllocationLinkedList; - DedicatedAllocationLinkedList m_DedicatedAllocations[VK_MAX_MEMORY_TYPES]; - VMA_RW_MUTEX m_DedicatedAllocationsMutex[VK_MAX_MEMORY_TYPES]; + VmaDedicatedAllocationList m_DedicatedAllocations[VK_MAX_MEMORY_TYPES]; VmaCurrentBudgetData m_Budget; VMA_ATOMIC_UINT32 m_DeviceMemoryCount; // Total number of VkDeviceMemory objects. @@ -6371,7 +11385,7 @@ public: const VkAllocationCallbacks* GetAllocationCallbacks() const { - return m_AllocationCallbacksSpecified ? &m_AllocationCallbacks : 0; + return m_AllocationCallbacksSpecified ? &m_AllocationCallbacks : VMA_NULL; } const VmaVulkanFunctions& GetVulkanFunctions() const { @@ -6416,10 +11430,6 @@ public: uint32_t GetGlobalMemoryTypeBits() const { return m_GlobalMemoryTypeBits; } -#if VMA_RECORDING_ENABLED - VmaRecorder* GetRecorder() const { return m_pRecorder; } -#endif - void GetBufferMemoryRequirements( VkBuffer hBuffer, VkMemoryRequirements& memReq, @@ -6430,6 +11440,11 @@ public: VkMemoryRequirements& memReq, bool& requiresDedicatedAllocation, bool& prefersDedicatedAllocation) const; + VkResult FindMemoryTypeIndex( + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkFlags bufImgUsage, // VkBufferCreateInfo::usage or VkImageCreateInfo::usage. UINT32_MAX if unknown. + uint32_t* pMemoryTypeIndex) const; // Main allocation function. VkResult AllocateMemory( @@ -6437,8 +11452,8 @@ public: bool requiresDedicatedAllocation, bool prefersDedicatedAllocation, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, // UINT32_MAX when unknown. VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, // UINT32_MAX if unknown. const VmaAllocationCreateInfo& createInfo, VmaSuballocationType suballocType, size_t allocationCount, @@ -6449,46 +11464,28 @@ public: size_t allocationCount, const VmaAllocation* pAllocations); - void CalculateStats(VmaStats* pStats); + void CalculateStatistics(VmaTotalStatistics* pStats); - void GetBudget( - VmaBudget* outBudget, uint32_t firstHeap, uint32_t heapCount); + void GetHeapBudgets( + VmaBudget* outBudgets, uint32_t firstHeap, uint32_t heapCount); #if VMA_STATS_STRING_ENABLED void PrintDetailedMap(class VmaJsonWriter& json); #endif - VkResult DefragmentationBegin( - const VmaDefragmentationInfo2& info, - VmaDefragmentationStats* pStats, - VmaDefragmentationContext* pContext); - VkResult DefragmentationEnd( - VmaDefragmentationContext context); - - VkResult DefragmentationPassBegin( - VmaDefragmentationPassInfo* pInfo, - VmaDefragmentationContext context); - VkResult DefragmentationPassEnd( - 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 GetPoolStatistics(VmaPool pool, VmaStatistics* pPoolStats); + void CalculatePoolStatistics(VmaPool pool, VmaDetailedStatistics* 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); - // Call to Vulkan function vkAllocateMemory with accompanying bookkeeping. VkResult AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory); // Call to Vulkan function vkFreeMemory with accompanying bookkeeping. @@ -6566,10 +11563,6 @@ private: // Global bit mask AND-ed with any memoryTypeBits to disallow certain memory types. uint32_t m_GlobalMemoryTypeBits; -#if VMA_RECORDING_ENABLED - VmaRecorder* m_pRecorder; -#endif - void ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunctions); #if VMA_STATIC_VULKAN_FUNCTIONS == 1 @@ -6587,53 +11580,71 @@ private: VkDeviceSize CalcPreferredBlockSize(uint32_t memTypeIndex); VkResult AllocateMemoryOfType( + VmaPool pool, VkDeviceSize size, VkDeviceSize alignment, - bool dedicatedAllocation, + bool dedicatedPreferred, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, const VmaAllocationCreateInfo& createInfo, uint32_t memTypeIndex, VmaSuballocationType suballocType, + VmaDedicatedAllocationList& dedicatedAllocations, + VmaBlockVector& blockVector, size_t allocationCount, VmaAllocation* pAllocations); // Helper function only to be used inside AllocateDedicatedMemory. VkResult AllocateDedicatedMemoryPage( + VmaPool pool, VkDeviceSize size, VmaSuballocationType suballocType, uint32_t memTypeIndex, const VkMemoryAllocateInfo& allocInfo, bool map, bool isUserDataString, + bool isMappingAllowed, void* pUserData, VmaAllocation* pAllocation); // Allocates and registers new VkDeviceMemory specifically for dedicated allocations. VkResult AllocateDedicatedMemory( + VmaPool pool, VkDeviceSize size, VmaSuballocationType suballocType, + VmaDedicatedAllocationList& dedicatedAllocations, uint32_t memTypeIndex, - bool withinBudget, bool map, bool isUserDataString, + bool isMappingAllowed, + bool canAliasMemory, void* pUserData, float priority, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, size_t allocationCount, - VmaAllocation* pAllocations); + VmaAllocation* pAllocations, + const void* pNextChain = nullptr); void FreeDedicatedMemory(const VmaAllocation allocation); + VkResult CalcMemTypeParams( + VmaAllocationCreateInfo& outCreateInfo, + uint32_t memTypeIndex, + VkDeviceSize size, + size_t allocationCount); + VkResult CalcAllocationParams( + VmaAllocationCreateInfo& outCreateInfo, + bool dedicatedRequired, + bool dedicatedPreferred); + /* Calculates and returns bit mask of memory types that can support defragmentation on GPU as they support creation of required buffer for copy operations. */ uint32_t CalculateGpuDefragmentationMemoryTypeBits() const; - uint32_t CalculateGlobalMemoryTypeBits() const; bool GetFlushOrInvalidateRange( @@ -6646,9 +11657,8 @@ private: #endif // #if VMA_MEMORY_BUDGET }; -//////////////////////////////////////////////////////////////////////////////// -// Memory allocation #2 after VmaAllocator_T definition +#ifndef _VMA_MEMORY_FUNCTIONS static void* VmaMalloc(VmaAllocator hAllocator, size_t size, size_t alignment) { return VmaMalloc(&hAllocator->m_AllocationCallbacks, size, alignment); @@ -6691,421 +11701,377 @@ static void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count) VmaFree(hAllocator, ptr); } } +#endif // _VMA_MEMORY_FUNCTIONS -//////////////////////////////////////////////////////////////////////////////// -// VmaStringBuilder +#ifndef _VMA_DEVICE_MEMORY_BLOCK_FUNCTIONS +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) {} -#if VMA_STATS_STRING_ENABLED - -class VmaStringBuilder +VmaDeviceMemoryBlock::~VmaDeviceMemoryBlock() { -public: - VmaStringBuilder(VmaAllocator alloc) : m_Data(VmaStlAllocator(alloc->GetAllocationCallbacks())) { } - size_t GetLength() const { return m_Data.size(); } - const char* GetData() const { return m_Data.data(); } + VMA_ASSERT(m_MapCount == 0 && "VkDeviceMemory block is being destroyed while it is still mapped."); + VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); +} - 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 > m_Data; -}; - -void VmaStringBuilder::Add(const char* pStr) +void VmaDeviceMemoryBlock::Init( + VmaAllocator hAllocator, + VmaPool hParentPool, + uint32_t newMemoryTypeIndex, + VkDeviceMemory newMemory, + VkDeviceSize newSize, + uint32_t id, + uint32_t algorithm, + VkDeviceSize bufferImageGranularity) { - const size_t strLen = strlen(pStr); - if(strLen > 0) + VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); + + m_hParentPool = hParentPool; + m_MemoryTypeIndex = newMemoryTypeIndex; + m_Id = id; + m_hMemory = newMemory; + + switch (algorithm) { - const size_t oldCount = m_Data.size(); - m_Data.resize(oldCount + strLen); - memcpy(m_Data.data() + oldCount, pStr, strLen); + case VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_Linear)(hAllocator->GetAllocationCallbacks(), + bufferImageGranularity, false); // isVirtual + break; + default: + VMA_ASSERT(0); + // Fall-through. + case 0: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_TLSF)(hAllocator->GetAllocationCallbacks(), + bufferImageGranularity, false); // isVirtual } + m_pMetadata->Init(newSize); } -void VmaStringBuilder::AddNumber(uint32_t num) +void VmaDeviceMemoryBlock::Destroy(VmaAllocator allocator) { - char buf[11]; - buf[10] = '\0'; - char *p = &buf[10]; - do + // Define macro VMA_DEBUG_LOG to receive the list of the unfreed allocations + if (!m_pMetadata->IsEmpty()) + m_pMetadata->DebugLogAllAllocations(); + // 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; +} + +void VmaDeviceMemoryBlock::PostFree(VmaAllocator hAllocator) +{ + if(m_MappingHysteresis.PostFree()) { - *--p = '0' + (num % 10); - num /= 10; - } - while(num); - Add(p); -} - -void VmaStringBuilder::AddNumber(uint64_t num) -{ - char buf[21]; - buf[20] = '\0'; - char *p = &buf[20]; - do - { - *--p = '0' + (num % 10); - num /= 10; - } - while(num); - Add(p); -} - -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 > 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(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 == '\\') + VMA_ASSERT(m_MappingHysteresis.GetExtraMapping() == 0); + if (m_MapCount == 0) { - 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; + m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)(hAllocator->m_hDevice, m_hMemory); } } } -void VmaJsonWriter::ContinueString(uint32_t n) +bool VmaDeviceMemoryBlock::Validate() const { - VMA_ASSERT(m_InsideString); - m_SB.AddNumber(n); + VMA_VALIDATE((m_hMemory != VK_NULL_HANDLE) && + (m_pMetadata->GetSize() != 0)); + + return m_pMetadata->Validate(); } -void VmaJsonWriter::ContinueString(uint64_t n) +VkResult VmaDeviceMemoryBlock::CheckCorruption(VmaAllocator hAllocator) { - 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') + void* pData = nullptr; + VkResult res = Map(hAllocator, 1, &pData); + if (res != VK_SUCCESS) { - ContinueString(pStr); + return res; } - m_SB.Add('"'); - m_InsideString = false; + + res = m_pMetadata->CheckCorruption(pData); + + Unmap(hAllocator, 1); + + return res; } -void VmaJsonWriter::WriteNumber(uint32_t n) +VkResult VmaDeviceMemoryBlock::Map(VmaAllocator hAllocator, uint32_t count, void** ppData) { - 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()) + if (count == 0) { - StackItem& currItem = m_Stack.back(); - if(currItem.type == COLLECTION_TYPE_OBJECT && - currItem.valueCount % 2 == 0) - { - 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; + return VK_SUCCESS; } -} -void VmaJsonWriter::WriteIndent(bool oneLess) -{ - if(!m_Stack.empty() && !m_Stack.back().singleLineMode) + VmaMutexLock lock(m_MapAndBindMutex, hAllocator->m_UseMutex); + const uint32_t oldTotalMapCount = m_MapCount + m_MappingHysteresis.GetExtraMapping(); + m_MappingHysteresis.PostMap(); + if (oldTotalMapCount != 0) { - m_SB.AddNewLine(); - - size_t count = m_Stack.size(); - if(count > 0 && oneLess) + m_MapCount += count; + VMA_ASSERT(m_pMappedData != VMA_NULL); + if (ppData != VMA_NULL) { - --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) - { - m_pUserData = VmaCreateStringCopy(hAllocator->GetAllocationCallbacks(), (const char*)pUserData); + *ppData = m_pMappedData; } + return VK_SUCCESS; } else { - m_pUserData = pUserData; + 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 VmaAllocation_T::ChangeBlockAllocation( - VmaAllocator hAllocator, - VmaDeviceMemoryBlock* block, - VkDeviceSize offset) +void VmaDeviceMemoryBlock::Unmap(VmaAllocator hAllocator, uint32_t count) { - 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) + if (count == 0) { - 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); + return; } - m_BlockAllocation.m_Block = block; - m_BlockAllocation.m_Offset = offset; + VmaMutexLock lock(m_MapAndBindMutex, hAllocator->m_UseMutex); + if (m_MapCount >= count) + { + m_MapCount -= count; + const uint32_t totalMapCount = m_MapCount + m_MappingHysteresis.GetExtraMapping(); + if (totalMapCount == 0) + { + m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)(hAllocator->m_hDevice, m_hMemory); + } + m_MappingHysteresis.PostUnmap(); + } + else + { + VMA_ASSERT(0 && "VkDeviceMemory block is being unmapped while it was not previously mapped."); + } } -void VmaAllocation_T::ChangeOffset(VkDeviceSize newOffset) +VkResult VmaDeviceMemoryBlock::WriteMagicValueAfterAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize) { + VMA_ASSERT(VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_MARGIN % 4 == 0 && VMA_DEBUG_DETECT_CORRUPTION); + + void* pData; + VkResult res = Map(hAllocator, 1, &pData); + if (res != VK_SUCCESS) + { + return res; + } + + VmaWriteMagicValue(pData, allocOffset + allocSize); + + Unmap(hAllocator, 1); + return VK_SUCCESS; +} + +VkResult VmaDeviceMemoryBlock::ValidateMagicValueAfterAllocation(VmaAllocator hAllocator, VkDeviceSize allocOffset, VkDeviceSize allocSize) +{ + VMA_ASSERT(VMA_DEBUG_MARGIN > 0 && VMA_DEBUG_MARGIN % 4 == 0 && VMA_DEBUG_DETECT_CORRUPTION); + + void* pData; + VkResult res = Map(hAllocator, 1, &pData); + if (res != VK_SUCCESS) + { + return res; + } + + 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, + VkDeviceSize allocationLocalOffset, + VkBuffer hBuffer, + const void* pNext) +{ + VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && + hAllocation->GetBlock() == this); + VMA_ASSERT(allocationLocalOffset < hAllocation->GetSize() && + "Invalid allocationLocalOffset. Did you forget that this offset is relative to the beginning of the allocation, not the whole memory block?"); + const VkDeviceSize memoryOffset = hAllocation->GetOffset() + allocationLocalOffset; + // 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_MapAndBindMutex, hAllocator->m_UseMutex); + return hAllocator->BindVulkanBuffer(m_hMemory, memoryOffset, hBuffer, pNext); +} + +VkResult VmaDeviceMemoryBlock::BindImageMemory( + const VmaAllocator hAllocator, + const VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkImage hImage, + const void* pNext) +{ + VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && + hAllocation->GetBlock() == this); + VMA_ASSERT(allocationLocalOffset < hAllocation->GetSize() && + "Invalid allocationLocalOffset. Did you forget that this offset is relative to the beginning of the allocation, not the whole memory block?"); + const VkDeviceSize memoryOffset = hAllocation->GetOffset() + allocationLocalOffset; + // 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_MapAndBindMutex, hAllocator->m_UseMutex); + return hAllocator->BindVulkanImage(m_hMemory, memoryOffset, hImage, pNext); +} +#endif // _VMA_DEVICE_MEMORY_BLOCK_FUNCTIONS + +#ifndef _VMA_ALLOCATION_T_FUNCTIONS +VmaAllocation_T::VmaAllocation_T(bool mappingAllowed) + : m_Alignment{ 1 }, + m_Size{ 0 }, + m_pUserData{ VMA_NULL }, + m_pName{ VMA_NULL }, + m_MemoryTypeIndex{ 0 }, + m_Type{ (uint8_t)ALLOCATION_TYPE_NONE }, + m_SuballocationType{ (uint8_t)VMA_SUBALLOCATION_TYPE_UNKNOWN }, + m_MapCount{ 0 }, + m_Flags{ 0 } +{ + if(mappingAllowed) + m_Flags |= (uint8_t)FLAG_MAPPING_ALLOWED; + +#if VMA_STATS_STRING_ENABLED + m_BufferImageUsage = 0; +#endif +} + +VmaAllocation_T::~VmaAllocation_T() +{ + VMA_ASSERT(m_MapCount == 0 && "Allocation was not unmapped before destruction."); + + // Check if owned string was freed. + VMA_ASSERT(m_pName == VMA_NULL); +} + +void VmaAllocation_T::InitBlockAllocation( + VmaDeviceMemoryBlock* block, + VmaAllocHandle allocHandle, + VkDeviceSize alignment, + VkDeviceSize size, + uint32_t memoryTypeIndex, + VmaSuballocationType suballocationType, + bool mapped) +{ + 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_MemoryTypeIndex = memoryTypeIndex; + if(mapped) + { + VMA_ASSERT(IsMappingAllowed() && "Mapping is not allowed on this allocation! Please use one of the new VMA_ALLOCATION_CREATE_HOST_ACCESS_* flags when creating it."); + m_Flags |= (uint8_t)FLAG_PERSISTENT_MAP; + } + m_SuballocationType = (uint8_t)suballocationType; + m_BlockAllocation.m_Block = block; + m_BlockAllocation.m_AllocHandle = allocHandle; +} + +void VmaAllocation_T::InitDedicatedAllocation( + VmaPool hParentPool, + 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_MemoryTypeIndex = memoryTypeIndex; + m_SuballocationType = (uint8_t)suballocationType; + if(pMappedData != VMA_NULL) + { + VMA_ASSERT(IsMappingAllowed() && "Mapping is not allowed on this allocation! Please use one of the new VMA_ALLOCATION_CREATE_HOST_ACCESS_* flags when creating it."); + m_Flags |= (uint8_t)FLAG_PERSISTENT_MAP; + } + m_DedicatedAllocation.m_hParentPool = hParentPool; + m_DedicatedAllocation.m_hMemory = hMemory; + m_DedicatedAllocation.m_pMappedData = pMappedData; + m_DedicatedAllocation.m_Prev = VMA_NULL; + m_DedicatedAllocation.m_Next = VMA_NULL; +} + +void VmaAllocation_T::SetName(VmaAllocator hAllocator, const char* pName) +{ + VMA_ASSERT(pName == VMA_NULL || pName != m_pName); + + FreeName(hAllocator); + + if (pName != VMA_NULL) + m_pName = VmaCreateStringCopy(hAllocator->GetAllocationCallbacks(), pName); +} + +uint8_t VmaAllocation_T::SwapBlockAllocation(VmaAllocator hAllocator, VmaAllocation allocation) +{ + VMA_ASSERT(allocation != VMA_NULL); VMA_ASSERT(m_Type == ALLOCATION_TYPE_BLOCK); - m_BlockAllocation.m_Offset = newOffset; + VMA_ASSERT(allocation->m_Type == ALLOCATION_TYPE_BLOCK); + + if (m_MapCount != 0) + m_BlockAllocation.m_Block->Unmap(hAllocator, m_MapCount); + + m_BlockAllocation.m_Block->m_pMetadata->SetAllocationUserData(m_BlockAllocation.m_AllocHandle, allocation); + VMA_SWAP(m_BlockAllocation, allocation->m_BlockAllocation); + m_BlockAllocation.m_Block->m_pMetadata->SetAllocationUserData(m_BlockAllocation.m_AllocHandle, this); + +#if VMA_STATS_STRING_ENABLED + VMA_SWAP(m_BufferImageUsage, allocation->m_BufferImageUsage); +#endif + return m_MapCount; +} + +VmaAllocHandle VmaAllocation_T::GetAllocHandle() const +{ + switch (m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_AllocHandle; + case ALLOCATION_TYPE_DEDICATED: + return VK_NULL_HANDLE; + default: + VMA_ASSERT(0); + return VK_NULL_HANDLE; + } } VkDeviceSize VmaAllocation_T::GetOffset() const { - switch(m_Type) + switch (m_Type) { case ALLOCATION_TYPE_BLOCK: - return m_BlockAllocation.m_Offset; + return m_BlockAllocation.m_Block->m_pMetadata->GetAllocationOffset(m_BlockAllocation.m_AllocHandle); case ALLOCATION_TYPE_DEDICATED: return 0; default: @@ -7114,9 +12080,23 @@ VkDeviceSize VmaAllocation_T::GetOffset() const } } +VmaPool VmaAllocation_T::GetParentPool() const +{ + switch (m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->GetParentPool(); + case ALLOCATION_TYPE_DEDICATED: + return m_DedicatedAllocation.m_hParentPool; + default: + VMA_ASSERT(0); + return VK_NULL_HANDLE; + } +} + VkDeviceMemory VmaAllocation_T::GetMemory() const { - switch(m_Type) + switch (m_Type) { case ALLOCATION_TYPE_BLOCK: return m_BlockAllocation.m_Block->GetDeviceMemory(); @@ -7130,14 +12110,14 @@ VkDeviceMemory VmaAllocation_T::GetMemory() const void* VmaAllocation_T::GetMappedData() const { - switch(m_Type) + switch (m_Type) { case ALLOCATION_TYPE_BLOCK: - if(m_MapCount != 0) + if (m_MapCount != 0 || IsPersistentMap()) { void* pBlockData = m_BlockAllocation.m_Block->GetMappedData(); VMA_ASSERT(pBlockData != VMA_NULL); - return (char*)pBlockData + m_BlockAllocation.m_Offset; + return (char*)pBlockData + GetOffset(); } else { @@ -7145,7 +12125,7 @@ void* VmaAllocation_T::GetMappedData() const } break; case ALLOCATION_TYPE_DEDICATED: - VMA_ASSERT((m_DedicatedAllocation.m_pMappedData != VMA_NULL) == (m_MapCount != 0)); + VMA_ASSERT((m_DedicatedAllocation.m_pMappedData != VMA_NULL) == (m_MapCount != 0 || IsPersistentMap())); return m_DedicatedAllocation.m_pMappedData; default: VMA_ASSERT(0); @@ -7153,114 +12133,12 @@ void* VmaAllocation_T::GetMappedData() const } } -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; - } -} - -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()); - VmaFreeString(hAllocator->GetAllocationCallbacks(), (char*)m_pUserData); - m_pUserData = VMA_NULL; -} - void VmaAllocation_T::BlockAllocMap() { VMA_ASSERT(GetType() == ALLOCATION_TYPE_BLOCK); + VMA_ASSERT(IsMappingAllowed() && "Mapping is not allowed on this allocation! Please use one of the new VMA_ALLOCATION_CREATE_HOST_ACCESS_* flags when creating it."); - if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) < 0x7F) + if (m_MapCount < 0xFF) { ++m_MapCount; } @@ -7274,7 +12152,7 @@ void VmaAllocation_T::BlockAllocUnmap() { VMA_ASSERT(GetType() == ALLOCATION_TYPE_BLOCK); - if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) != 0) + if (m_MapCount > 0) { --m_MapCount; } @@ -7287,10 +12165,11 @@ void VmaAllocation_T::BlockAllocUnmap() VkResult VmaAllocation_T::DedicatedAllocMap(VmaAllocator hAllocator, void** ppData) { VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); + VMA_ASSERT(IsMappingAllowed() && "Mapping is not allowed on this allocation! Please use one of the new VMA_ALLOCATION_CREATE_HOST_ACCESS_* flags when creating it."); - if(m_MapCount != 0) + if (m_MapCount != 0 || IsPersistentMap()) { - if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) < 0x7F) + if (m_MapCount < 0xFF) { VMA_ASSERT(m_DedicatedAllocation.m_pMappedData != VMA_NULL); *ppData = m_DedicatedAllocation.m_pMappedData; @@ -7312,7 +12191,7 @@ VkResult VmaAllocation_T::DedicatedAllocMap(VmaAllocator hAllocator, void** ppDa VK_WHOLE_SIZE, 0, // flags ppData); - if(result == VK_SUCCESS) + if (result == VK_SUCCESS) { m_DedicatedAllocation.m_pMappedData = *ppData; m_MapCount = 1; @@ -7325,10 +12204,10 @@ void VmaAllocation_T::DedicatedAllocUnmap(VmaAllocator hAllocator) { VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); - if((m_MapCount & ~MAP_COUNT_FLAG_PERSISTENT_MAP) != 0) + if (m_MapCount > 0) { --m_MapCount; - if(m_MapCount == 0) + if (m_MapCount == 0 && !IsPersistentMap()) { m_DedicatedAllocation.m_pMappedData = VMA_NULL; (*hAllocator->GetVulkanFunctions().vkUnmapMemory)( @@ -7343,3739 +12222,48 @@ void VmaAllocation_T::DedicatedAllocUnmap(VmaAllocator hAllocator) } #if VMA_STATS_STRING_ENABLED - -static void VmaPrintStatInfo(VmaJsonWriter& json, const VmaStatInfo& stat) +void VmaAllocation_T::InitBufferImageUsage(uint32_t bufferImageUsage) { - 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(); + VMA_ASSERT(m_BufferImageUsage == 0); + m_BufferImageUsage = bufferImageUsage; } -#endif // #if VMA_STATS_STRING_ENABLED - -struct VmaSuballocationItemSizeLess +void VmaAllocation_T::PrintParameters(class VmaJsonWriter& json) const { - 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(VMA_SUBALLOCATION_TYPE_NAMES[m_SuballocationType]); json.WriteString("Size"); - json.WriteNumber(size); + json.WriteNumber(m_Size); + json.WriteString("Usage"); + json.WriteNumber(m_BufferImageUsage); - 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(hAllocator->GetAllocationCallbacks())), - m_FreeSuballocationsBySize(VmaStlAllocator(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(const auto& subAlloc : m_Suballocations) - { - // 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. - VMA_VALIDATE(subAlloc.size >= VMA_DEBUG_MARGIN); - } - 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 calculated 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(const auto& suballoc : m_Suballocations) - { - 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 - - for(const auto& suballoc : m_Suballocations) - { - if(suballoc.type == VMA_SUBALLOCATION_TYPE_FREE) - { - PrintDetailedMap_UnusedRange(json, suballoc.offset, suballoc.size); - } - else - { - PrintDetailedMap_Allocation(json, suballoc.offset, suballoc.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); - VMA_ASSERT(allocType != VMA_SUBALLOCATION_TYPE_FREE); - VMA_ASSERT(pAllocationRequest != VMA_NULL); - VMA_HEAVY_ASSERT(Validate()); - - pAllocationRequest->type = VmaAllocationRequestType::Normal; - - // 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. - - bool found = false; - VmaAllocationRequest tmpAllocRequest = {}; - tmpAllocRequest.type = VmaAllocationRequestType::Normal; - 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)) - { - if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_FIRST_FIT_BIT) - { - *pAllocationRequest = tmpAllocRequest; - pAllocationRequest->item = suballocIt; - break; - } - if(!found || tmpAllocRequest.CalcCost() < pAllocationRequest->CalcCost()) - { - *pAllocationRequest = tmpAllocRequest; - pAllocationRequest->item = suballocIt; - found = true; - } - } - } - } - - return found; - } - - return false; -} - -bool VmaBlockMetadata_Generic::MakeRequestedAllocationsLost( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VmaAllocationRequest* pAllocationRequest) -{ - VMA_ASSERT(pAllocationRequest && pAllocationRequest->type == VmaAllocationRequestType::Normal); - - 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(auto& suballoc : m_Suballocations) - { - 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_UNKNOWN; - } - if(!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) - { - VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); - return VK_ERROR_UNKNOWN; - } - } - } - - return VK_SUCCESS; -} - -void VmaBlockMetadata_Generic::Alloc( - const VmaAllocationRequest& request, - VmaSuballocationType type, - VkDeviceSize allocSize, - VmaAllocation hAllocation) -{ - VMA_ASSERT(request.type == VmaAllocationRequestType::Normal); - 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::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 && bufferImageGranularity != allocAlignment) - { - 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(allocSize % bufferImageGranularity || *pOffset % bufferImageGranularity) - { - 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 && bufferImageGranularity != allocAlignment) - { - 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(allocSize % bufferImageGranularity || *pOffset % bufferImageGranularity) - { - 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(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(const auto& suballoc : m_Suballocations) - { - const VmaSuballocationType suballocType = suballoc.type; - if(suballocType != VMA_SUBALLOCATION_TYPE_FREE) - { - minAlignment = VMA_MIN(minAlignment, suballoc.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(hAllocator->GetAllocationCallbacks())), - m_Suballocations1(VmaStlAllocator(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 - would 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()); - return upperAddress ? - CreateAllocationRequest_UpperAddress( - currentFrameIndex, frameInUseCount, bufferImageGranularity, - allocSize, allocAlignment, allocType, canMakeOtherLost, strategy, pAllocationRequest) : - CreateAllocationRequest_LowerAddress( - currentFrameIndex, frameInUseCount, bufferImageGranularity, - allocSize, allocAlignment, allocType, canMakeOtherLost, strategy, pAllocationRequest); -} - -bool VmaBlockMetadata_Linear::CreateAllocationRequest_UpperAddress( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, - VkDeviceSize allocSize, - VkDeviceSize allocAlignment, - VmaSuballocationType allocType, - bool canMakeOtherLost, - uint32_t strategy, - VmaAllocationRequest* pAllocationRequest) -{ - const VkDeviceSize size = GetSize(); - SuballocationVectorType& suballocations1st = AccessSuballocations1st(); - SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); - - 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(resultOffset < VMA_DEBUG_MARGIN) - { - return false; - } - 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 && bufferImageGranularity != allocAlignment && !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; - pAllocationRequest->type = VmaAllocationRequestType::UpperAddress; - return true; - } - - return false; -} - -bool VmaBlockMetadata_Linear::CreateAllocationRequest_LowerAddress( - uint32_t currentFrameIndex, - uint32_t frameInUseCount, - VkDeviceSize bufferImageGranularity, - VkDeviceSize allocSize, - VkDeviceSize allocAlignment, - VmaSuballocationType allocType, - bool canMakeOtherLost, - uint32_t strategy, - VmaAllocationRequest* pAllocationRequest) -{ - const VkDeviceSize size = GetSize(); - SuballocationVectorType& suballocations1st = AccessSuballocations1st(); - SuballocationVectorType& suballocations2nd = AccessSuballocations2nd(); - - 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 && bufferImageGranularity != allocAlignment && !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((allocSize % bufferImageGranularity || resultOffset % bufferImageGranularity) && 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, customData unused. - pAllocationRequest->type = VmaAllocationRequestType::EndOf1st; - 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 && bufferImageGranularity != allocAlignment && !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(allocSize % bufferImageGranularity || resultOffset % bufferImageGranularity) - { - 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; - } - } - - // Special case: There is not enough room at the end for this allocation, even after making all from the 1st lost. - if(index1st == suballocations1st.size() && - resultOffset + allocSize + VMA_DEBUG_MARGIN > size) - { - // TODO: This is a known bug that it's not yet implemented and the allocation is failing. - VMA_DEBUG_LOG("Unsupported special case in custom pool with linear allocation algorithm used as ring buffer with allocations that can be lost."); - } - } - - // 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(allocSize % bufferImageGranularity || resultOffset % bufferImageGranularity) - { - 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->type = VmaAllocationRequestType::EndOf2nd; - // pAllocationRequest->item, customData 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); - - // We always start from 1st. - SuballocationVectorType* suballocations = &AccessSuballocations1st(); - size_t index = m_1stNullItemsBeginCount; - size_t madeLostCount = 0; - while(madeLostCount < pAllocationRequest->itemsToMakeLostCount) - { - if(index == suballocations->size()) - { - index = 0; - // If we get to the end of 1st, we wrap around to beginning of 2nd of 1st. - if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER) - { - suballocations = &AccessSuballocations2nd(); - } - // else: m_2ndVectorMode == SECOND_VECTOR_EMPTY: - // suballocations continues pointing at AccessSuballocations1st(). - VMA_ASSERT(!suballocations->empty()); - } - VmaSuballocation& suballoc = (*suballocations)[index]; - 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; - if(suballocations == &AccessSuballocations1st()) - { - ++m_1stNullItemsMiddleCount; - } - else - { - ++m_2ndNullItemsCount; - } - ++madeLostCount; - } - else - { - return false; - } - } - ++index; - } - - CleanupAfterFree(); - //VMA_HEAVY_ASSERT(Validate()); // Already called by CleanupAfterFree(). - - 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; - m_SumFreeSize += suballoc.size; - ++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_UNKNOWN; - } - if(!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) - { - VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); - return VK_ERROR_UNKNOWN; - } - } - } - - 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_UNKNOWN; - } - if(!VmaValidateMagicValue(pBlockData, suballoc.offset + suballoc.size)) - { - VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER VALIDATED ALLOCATION!"); - return VK_ERROR_UNKNOWN; - } - } - } - - return VK_SUCCESS; -} - -void VmaBlockMetadata_Linear::Alloc( - const VmaAllocationRequest& request, - VmaSuballocationType type, - VkDeviceSize allocSize, - VmaAllocation hAllocation) -{ - const VmaSuballocation newSuballoc = { request.offset, allocSize, hAllocation, type }; - - switch(request.type) - { - case VmaAllocationRequestType::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; - } - break; - case VmaAllocationRequestType::EndOf1st: - { - SuballocationVectorType& suballocations1st = AccessSuballocations1st(); - - VMA_ASSERT(suballocations1st.empty() || - 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); - } - break; - case VmaAllocationRequestType::EndOf2nd: - { - SuballocationVectorType& suballocations1st = AccessSuballocations1st(); - // New allocation at the end of 2-part ring buffer, so before first allocation from 1st vector. - VMA_ASSERT(!suballocations1st.empty() && - 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); - } - break; - default: - 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 = VmaBinaryFindSorted( - suballocations1st.begin() + m_1stNullItemsBeginCount, - suballocations1st.end(), - refSuballoc, - VmaSuballocationOffsetLess()); - 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 ? - VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetLess()) : - VmaBinaryFindSorted(suballocations2nd.begin(), suballocations2nd.end(), refSuballoc, VmaSuballocationOffsetGreater()); - 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(); - } - - // Find more null items at the beginning of 2nd vector. - while(m_2ndNullItemsCount > 0 && - suballocations2nd[0].hAllocation == VK_NULL_HANDLE) - { - --m_2ndNullItemsCount; - VmaVectorRemove(suballocations2nd, 0); - } - - 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."); - - // 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->type = VmaAllocationRequestType::Normal; - 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, - VmaAllocation hAllocation) -{ - VMA_ASSERT(request.type == VmaAllocationRequestType::Normal); - - 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, - VmaPool hParentPool, - uint32_t newMemoryTypeIndex, - VkDeviceMemory newMemory, - VkDeviceSize newSize, - uint32_t id, - uint32_t algorithm) -{ - VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); - - m_hParentPool = hParentPool; - 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)) + if (m_pUserData != VMA_NULL) { - VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED BEFORE FREED ALLOCATION!"); + json.WriteString("CustomData"); + json.BeginString(); + json.ContinueString_Pointer(m_pUserData); + json.EndString(); } - else if(!VmaValidateMagicValue(pData, allocOffset + allocSize)) + if (m_pName != VMA_NULL) { - VMA_ASSERT(0 && "MEMORY CORRUPTION DETECTED AFTER FREED ALLOCATION!"); + json.WriteString("Name"); + json.WriteString(m_pName); } - - Unmap(hAllocator, 1); - - return VK_SUCCESS; -} - -VkResult VmaDeviceMemoryBlock::BindBufferMemory( - const VmaAllocator hAllocator, - const VmaAllocation hAllocation, - VkDeviceSize allocationLocalOffset, - VkBuffer hBuffer, - const void* pNext) -{ - VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && - hAllocation->GetBlock() == this); - VMA_ASSERT(allocationLocalOffset < hAllocation->GetSize() && - "Invalid allocationLocalOffset. Did you forget that this offset is relative to the beginning of the allocation, not the whole memory block?"); - const VkDeviceSize memoryOffset = hAllocation->GetOffset() + allocationLocalOffset; - // 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->BindVulkanBuffer(m_hMemory, memoryOffset, hBuffer, pNext); -} - -VkResult VmaDeviceMemoryBlock::BindImageMemory( - const VmaAllocator hAllocator, - const VmaAllocation hAllocation, - VkDeviceSize allocationLocalOffset, - VkImage hImage, - const void* pNext) -{ - VMA_ASSERT(hAllocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_BLOCK && - hAllocation->GetBlock() == this); - VMA_ASSERT(allocationLocalOffset < hAllocation->GetSize() && - "Invalid allocationLocalOffset. Did you forget that this offset is relative to the beginning of the allocation, not the whole memory block?"); - const VkDeviceSize memoryOffset = hAllocation->GetOffset() + allocationLocalOffset; - // 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->BindVulkanImage(m_hMemory, memoryOffset, hImage, pNext); -} - -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(inoutInfo.usedBytes, inoutInfo.allocationCount) : 0; - inoutInfo.unusedRangeSizeAvg = (inoutInfo.unusedRangeCount > 0) ? - VmaRoundDiv(inoutInfo.unusedBytes, inoutInfo.unusedRangeCount) : 0; -} - -VmaPool_T::VmaPool_T( - VmaAllocator hAllocator, - const VmaPoolCreateInfo& createInfo, - VkDeviceSize preferredBlockSize) : - m_BlockVector( - hAllocator, - this, // hParentPool - 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, - createInfo.blockSize != 0, // explicitBlockSize - createInfo.flags & VMA_POOL_CREATE_ALGORITHM_MASK, // algorithm - createInfo.priority, - VMA_MAX(hAllocator->GetMemoryTypeMinAlignment(createInfo.memoryTypeIndex), createInfo.minAllocationAlignment), - createInfo.pMemoryAllocateNext), - m_Id(0), - m_Name(VMA_NULL) -{ } +#endif // VMA_STATS_STRING_ENABLED -VmaPool_T::~VmaPool_T() +void VmaAllocation_T::FreeName(VmaAllocator hAllocator) { - VMA_ASSERT(m_PrevPool == VMA_NULL && m_NextPool == VMA_NULL); -} - -void VmaPool_T::SetName(const char* pName) -{ - const VkAllocationCallbacks* allocs = m_BlockVector.GetAllocator()->GetAllocationCallbacks(); - VmaFreeString(allocs, m_Name); - - if(pName != VMA_NULL) + if(m_pName) { - m_Name = VmaCreateStringCopy(allocs, pName); - } - else - { - m_Name = VMA_NULL; + VmaFreeString(hAllocator->GetAllocationCallbacks(), m_pName); + m_pName = VMA_NULL; } } +#endif // _VMA_ALLOCATION_T_FUNCTIONS -#if VMA_STATS_STRING_ENABLED - -#endif // #if VMA_STATS_STRING_ENABLED - +#ifndef _VMA_BLOCK_VECTOR_FUNCTIONS VmaBlockVector::VmaBlockVector( VmaAllocator hAllocator, VmaPool hParentPool, @@ -11084,34 +12272,29 @@ VmaBlockVector::VmaBlockVector( size_t minBlockCount, size_t maxBlockCount, VkDeviceSize bufferImageGranularity, - uint32_t frameInUseCount, bool explicitBlockSize, uint32_t algorithm, float priority, VkDeviceSize minAllocationAlignment, - void* pMemoryAllocateNext) : - m_hAllocator(hAllocator), + void* pMemoryAllocateNext) + : m_hAllocator(hAllocator), m_hParentPool(hParentPool), m_MemoryTypeIndex(memoryTypeIndex), m_PreferredBlockSize(preferredBlockSize), m_MinBlockCount(minBlockCount), m_MaxBlockCount(maxBlockCount), m_BufferImageGranularity(bufferImageGranularity), - m_FrameInUseCount(frameInUseCount), m_ExplicitBlockSize(explicitBlockSize), m_Algorithm(algorithm), m_Priority(priority), m_MinAllocationAlignment(minAllocationAlignment), m_pMemoryAllocateNext(pMemoryAllocateNext), - m_HasEmptyBlock(false), m_Blocks(VmaStlAllocator(hAllocator->GetAllocationCallbacks())), - m_NextBlockId(0) -{ -} + m_NextBlockId(0) {} VmaBlockVector::~VmaBlockVector() { - for(size_t i = m_Blocks.size(); i--; ) + for (size_t i = m_Blocks.size(); i--; ) { m_Blocks[i]->Destroy(m_hAllocator); vma_delete(m_hAllocator, m_Blocks[i]); @@ -11120,10 +12303,10 @@ VmaBlockVector::~VmaBlockVector() VkResult VmaBlockVector::CreateMinBlocks() { - for(size_t i = 0; i < m_MinBlockCount; ++i) + for (size_t i = 0; i < m_MinBlockCount; ++i) { VkResult res = CreateBlock(m_PreferredBlockSize, VMA_NULL); - if(res != VK_SUCCESS) + if (res != VK_SUCCESS) { return res; } @@ -11131,25 +12314,31 @@ VkResult VmaBlockVector::CreateMinBlocks() return VK_SUCCESS; } -void VmaBlockVector::GetPoolStats(VmaPoolStats* pStats) +void VmaBlockVector::AddStatistics(VmaStatistics& inoutStats) { 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) + 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); + pBlock->m_pMetadata->AddStatistics(inoutStats); + } +} + +void VmaBlockVector::AddDetailedStatistics(VmaDetailedStatistics& inoutStats) +{ + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + const size_t blockCount = m_Blocks.size(); + 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->AddDetailedStatistics(inoutStats); } } @@ -11168,10 +12357,7 @@ bool VmaBlockVector::IsCorruptionDetectionEnabled() const (m_hAllocator->m_MemProps.memoryTypes[m_MemoryTypeIndex].propertyFlags & requiredMemFlags) == requiredMemFlags; } -static const uint32_t VMA_ALLOCATION_TRY_COUNT = 32; - VkResult VmaBlockVector::Allocate( - uint32_t currentFrameIndex, VkDeviceSize size, VkDeviceSize alignment, const VmaAllocationCreateInfo& createInfo, @@ -11184,7 +12370,7 @@ VkResult VmaBlockVector::Allocate( alignment = VMA_MAX(alignment, m_MinAllocationAlignment); - if(IsCorruptionDetectionEnabled()) + if (IsCorruptionDetectionEnabled()) { size = VmaAlignUp(size, sizeof(VMA_CORRUPTION_DETECTION_MAGIC_VALUE)); alignment = VmaAlignUp(alignment, sizeof(VMA_CORRUPTION_DETECTION_MAGIC_VALUE)); @@ -11192,33 +12378,26 @@ VkResult VmaBlockVector::Allocate( { VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); - for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + for (allocIndex = 0; allocIndex < allocationCount; ++allocIndex) { res = AllocatePage( - currentFrameIndex, size, alignment, createInfo, suballocType, pAllocations + allocIndex); - if(res != VK_SUCCESS) + if (res != VK_SUCCESS) { break; } } } - if(res != VK_SUCCESS) + if (res != VK_SUCCESS) { // Free all already created allocations. - const uint32_t heapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex); - while(allocIndex--) - { - VmaAllocation_T* const alloc = pAllocations[allocIndex]; - const VkDeviceSize allocSize = alloc->GetSize(); - Free(alloc); - m_hAllocator->m_Budget.RemoveAllocation(heapIndex, allocSize); - } + while (allocIndex--) + Free(pAllocations[allocIndex]); memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); } @@ -11226,7 +12405,6 @@ VkResult VmaBlockVector::Allocate( } VkResult VmaBlockVector::AllocatePage( - uint32_t currentFrameIndex, VkDeviceSize size, VkDeviceSize alignment, const VmaAllocationCreateInfo& createInfo, @@ -11234,377 +12412,203 @@ VkResult VmaBlockVector::AllocatePage( 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; VkDeviceSize freeMemory; { const uint32_t heapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex); VmaBudget heapBudget = {}; - m_hAllocator->GetBudget(&heapBudget, heapIndex, 1); + m_hAllocator->GetHeapBudgets(&heapBudget, heapIndex, 1); freeMemory = (heapBudget.usage < heapBudget.budget) ? (heapBudget.budget - heapBudget.usage) : 0; } - const bool canFallbackToDedicated = !IsCustomPool(); + const bool canFallbackToDedicated = !HasExplicitBlockSize() && + (createInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) == 0; const bool canCreateNewBlock = ((createInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) == 0) && (m_Blocks.size() < m_MaxBlockCount) && (freeMemory >= size || !canFallbackToDedicated); 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 && + 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) + if (size + 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 is the case - e.g. for custom pools with linear algorithm. - */ - if(!canMakeOtherLost || canCreateNewBlock) + // 1. Search existing allocations. Try to allocate. + if (m_Algorithm == VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) { - // 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()) { - // Use only last block. - if(!m_Blocks.empty()) + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks.back(); + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock( + pCurrBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) { - VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks.back(); - VMA_ASSERT(pCurrBlock); - VkResult res = AllocateFromBlock( - pCurrBlock, - currentFrameIndex, - size, - alignment, - allocFlagsCopy, - createInfo.pUserData, - suballocType, - strategy, - pAllocation); - if(res == VK_SUCCESS) - { - VMA_DEBUG_LOG(" Returned from last block #%u", pCurrBlock->GetId()); - return VK_SUCCESS; - } + VMA_DEBUG_LOG(" Returned from last block #%u", pCurrBlock->GetId()); + IncrementallySortBlocks(); + return VK_SUCCESS; } } - else + } + else + { + if (strategy != VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT) // MIN_MEMORY or default { - if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT) + const bool isHostVisible = + (m_hAllocator->m_MemProps.memoryTypes[m_MemoryTypeIndex].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0; + if(isHostVisible) + { + const bool isMappingAllowed = (createInfo.flags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0; + /* + For non-mappable allocations, check blocks that are not mapped first. + For mappable allocations, check blocks that are already mapped first. + This way, having many blocks, we will separate mappable and non-mappable allocations, + hopefully limiting the number of blocks that are mapped, which will help tools like RenderDoc. + */ + for(size_t mappingI = 0; mappingI < 2; ++mappingI) + { + // 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); + const bool isBlockMapped = pCurrBlock->GetMappedData() != VMA_NULL; + if((mappingI == 0) == (isMappingAllowed == isBlockMapped)) + { + VkResult res = AllocateFromBlock( + pCurrBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) + { + VMA_DEBUG_LOG(" Returned from existing block #%u", pCurrBlock->GetId()); + IncrementallySortBlocks(); + return VK_SUCCESS; + } + } + } + } + } + else { // Forward order in m_Blocks - prefer blocks with smallest amount of free space. - for(size_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex ) + for (size_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) { VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; VMA_ASSERT(pCurrBlock); VkResult res = AllocateFromBlock( - pCurrBlock, - currentFrameIndex, - size, - alignment, - allocFlagsCopy, - createInfo.pUserData, - suballocType, - strategy, - pAllocation); - if(res == VK_SUCCESS) - { - VMA_DEBUG_LOG(" Returned from existing block #%u", pCurrBlock->GetId()); - 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, - currentFrameIndex, - size, - alignment, - allocFlagsCopy, - createInfo.pUserData, - suballocType, - strategy, - pAllocation); - if(res == VK_SUCCESS) + pCurrBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) { VMA_DEBUG_LOG(" Returned from existing block #%u", pCurrBlock->GetId()); + IncrementallySortBlocks(); return VK_SUCCESS; } } } } - - // 2. Try to create new block. - if(canCreateNewBlock) + else // VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT { - // 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) + // Backward order in m_Blocks - prefer blocks with largest amount of free space. + for (size_t blockIndex = m_Blocks.size(); blockIndex--; ) { - // 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) + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock(pCurrBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) { - const VkDeviceSize smallerNewBlockSize = newBlockSize / 2; - if(smallerNewBlockSize > maxExistingBlockSize && smallerNewBlockSize >= size * 2) - { - newBlockSize = smallerNewBlockSize; - ++newBlockSizeShift; - } - else - { - break; - } - } - } - - size_t newBlockIndex = 0; - VkResult res = (newBlockSize <= freeMemory || !canFallbackToDedicated) ? - CreateBlock(newBlockSize, &newBlockIndex) : VK_ERROR_OUT_OF_DEVICE_MEMORY; - // 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 = (newBlockSize <= freeMemory || !canFallbackToDedicated) ? - CreateBlock(newBlockSize, &newBlockIndex) : VK_ERROR_OUT_OF_DEVICE_MEMORY; - } - else - { - break; - } - } - } - - if(res == VK_SUCCESS) - { - VmaDeviceMemoryBlock* const pBlock = m_Blocks[newBlockIndex]; - VMA_ASSERT(pBlock->m_pMetadata->GetSize() >= size); - - res = AllocateFromBlock( - pBlock, - currentFrameIndex, - size, - alignment, - allocFlagsCopy, - createInfo.pUserData, - suballocType, - strategy, - pAllocation); - if(res == VK_SUCCESS) - { - VMA_DEBUG_LOG(" Created new block #%u Size=%llu", pBlock->GetId(), newBlockSize); + VMA_DEBUG_LOG(" Returned from existing block #%u", pCurrBlock->GetId()); + IncrementallySortBlocks(); 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) + // 2. Try to create new block. + if (canCreateNewBlock) { - uint32_t tryIndex = 0; - for(; tryIndex < VMA_ALLOCATION_TRY_COUNT; ++tryIndex) + // 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) { - VmaDeviceMemoryBlock* pBestRequestBlock = VMA_NULL; - VmaAllocationRequest bestRequest = {}; - VkDeviceSize bestRequestCost = VK_WHOLE_SIZE; - - // 1. Search existing allocations. - if(strategy == VMA_ALLOCATION_CREATE_STRATEGY_BEST_FIT_BIT) + // 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) { - // Forward order in m_Blocks - prefer blocks with smallest amount of free space. - for(size_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex ) + const VkDeviceSize smallerNewBlockSize = newBlockSize / 2; + if (smallerNewBlockSize > maxExistingBlockSize && smallerNewBlockSize >= size * 2) { - 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; - } - } - } + newBlockSize = smallerNewBlockSize; + ++newBlockSizeShift; + } + else + { + 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; - } - } - } + size_t newBlockIndex = 0; + VkResult res = (newBlockSize <= freeMemory || !canFallbackToDedicated) ? + CreateBlock(newBlockSize, &newBlockIndex) : VK_ERROR_OUT_OF_DEVICE_MEMORY; + // 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 = (newBlockSize <= freeMemory || !canFallbackToDedicated) ? + CreateBlock(newBlockSize, &newBlockIndex) : VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + else + { + break; } } + } - if(pBestRequestBlock != VMA_NULL) + if (res == VK_SUCCESS) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[newBlockIndex]; + VMA_ASSERT(pBlock->m_pMetadata->GetSize() >= size); + + res = AllocateFromBlock( + pBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) { - 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)) - { - // Allocate from this pBlock. - *pAllocation = m_hAllocator->m_AllocationObjectAllocator.Allocate(currentFrameIndex, isUserDataString); - pBestRequestBlock->m_pMetadata->Alloc(bestRequest, suballocType, size, *pAllocation); - UpdateHasEmptyBlock(); - (*pAllocation)->InitBlockAllocation( - pBestRequestBlock, - bestRequest.offset, - alignment, - size, - m_MemoryTypeIndex, - suballocType, - mapped, - (createInfo.flags & VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT) != 0); - VMA_HEAVY_ASSERT(pBestRequestBlock->Validate()); - VMA_DEBUG_LOG(" Returned from existing block #%u", pBestRequestBlock->GetId()); - (*pAllocation)->SetUserData(m_hAllocator, createInfo.pUserData); - m_hAllocator->m_Budget.AddAllocation(m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex), size); - 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); - 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. + VMA_DEBUG_LOG(" Created new block #%u Size=%llu", pBlock->GetId(), newBlockSize); + IncrementallySortBlocks(); + return VK_SUCCESS; } else { - // Could not find place in any of the blocks - break outer loop. - break; + // Allocation from new block failed, possibly due to VMA_DEBUG_MARGIN or alignment. + return VK_ERROR_OUT_OF_DEVICE_MEMORY; } } - /* 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( - const VmaAllocation hAllocation) +void VmaBlockVector::Free(const VmaAllocation hAllocation) { VmaDeviceMemoryBlock* pBlockToDelete = VMA_NULL; @@ -11612,7 +12616,7 @@ void VmaBlockVector::Free( { const uint32_t heapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex); VmaBudget heapBudget = {}; - m_hAllocator->GetBudget(&heapBudget, heapIndex, 1); + m_hAllocator->GetHeapBudgets(&heapBudget, heapIndex, 1); budgetExceeded = heapBudget.usage >= heapBudget.budget; } @@ -11622,67 +12626,71 @@ void VmaBlockVector::Free( VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); - if(IsCorruptionDetectionEnabled()) + if (IsCorruptionDetectionEnabled()) { - VkResult res = pBlock->ValidateMagicValueAroundAllocation(m_hAllocator, hAllocation->GetOffset(), hAllocation->GetSize()); + VkResult res = pBlock->ValidateMagicValueAfterAllocation(m_hAllocator, hAllocation->GetOffset(), hAllocation->GetSize()); VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to validate magic value."); } - if(hAllocation->IsPersistentMap()) + if (hAllocation->IsPersistentMap()) { pBlock->Unmap(m_hAllocator, 1); } - pBlock->m_pMetadata->Free(hAllocation); + const bool hadEmptyBlockBeforeFree = HasEmptyBlock(); + pBlock->m_pMetadata->Free(hAllocation->GetAllocHandle()); + pBlock->PostFree(m_hAllocator); VMA_HEAVY_ASSERT(pBlock->Validate()); VMA_DEBUG_LOG(" Freed from MemoryTypeIndex=%u", m_MemoryTypeIndex); const bool canDeleteBlock = m_Blocks.size() > m_MinBlockCount; // pBlock became empty after this deallocation. - if(pBlock->m_pMetadata->IsEmpty()) + if (pBlock->m_pMetadata->IsEmpty()) { - // Already has empty block. We don't want to have two, so delete this one. - if((m_HasEmptyBlock || budgetExceeded) && canDeleteBlock) + // Already had empty block. We don't want to have two, so delete this one. + if ((hadEmptyBlockBeforeFree || budgetExceeded) && canDeleteBlock) { pBlockToDelete = pBlock; Remove(pBlock); } - // else: We now have an empty block - leave it. + // else: We now have one empty block - leave it. A hysteresis to avoid allocating whole block back and forth. } // pBlock didn't become empty, but we have another empty block - find and free that one. // (This is optional, heuristics.) - else if(m_HasEmptyBlock && canDeleteBlock) + else if (hadEmptyBlockBeforeFree && canDeleteBlock) { VmaDeviceMemoryBlock* pLastBlock = m_Blocks.back(); - if(pLastBlock->m_pMetadata->IsEmpty()) + if (pLastBlock->m_pMetadata->IsEmpty()) { pBlockToDelete = pLastBlock; m_Blocks.pop_back(); } } - UpdateHasEmptyBlock(); IncrementallySortBlocks(); } // Destruction of a free block. Deferred until this point, outside of mutex // lock, for performance reason. - if(pBlockToDelete != VMA_NULL) + if (pBlockToDelete != VMA_NULL) { VMA_DEBUG_LOG(" Deleted empty block #%u", pBlockToDelete->GetId()); pBlockToDelete->Destroy(m_hAllocator); vma_delete(m_hAllocator, pBlockToDelete); } + + m_hAllocator->m_Budget.RemoveAllocation(m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex), hAllocation->GetSize()); + m_hAllocator->m_AllocationObjectAllocator.Free(hAllocation); } VkDeviceSize VmaBlockVector::CalcMaxBlockSize() const { VkDeviceSize result = 0; - for(size_t i = m_Blocks.size(); i--; ) + for (size_t i = m_Blocks.size(); i--; ) { result = VMA_MAX(result, m_Blocks[i]->m_pMetadata->GetSize()); - if(result >= m_PreferredBlockSize) + if (result >= m_PreferredBlockSize) { break; } @@ -11692,9 +12700,9 @@ VkDeviceSize VmaBlockVector::CalcMaxBlockSize() const void VmaBlockVector::Remove(VmaDeviceMemoryBlock* pBlock) { - for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + for (uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) { - if(m_Blocks[blockIndex] == pBlock) + if (m_Blocks[blockIndex] == pBlock) { VmaVectorRemove(m_Blocks, blockIndex); return; @@ -11705,12 +12713,14 @@ void VmaBlockVector::Remove(VmaDeviceMemoryBlock* pBlock) void VmaBlockVector::IncrementallySortBlocks() { - if(m_Algorithm != VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) + if (!m_IncrementalSort) + return; + 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) + for (size_t i = 1; i < m_Blocks.size(); ++i) { - if(m_Blocks[i - 1]->m_pMetadata->GetSumFreeSize() > m_Blocks[i]->m_pMetadata->GetSumFreeSize()) + if (m_Blocks[i - 1]->m_pMetadata->GetSumFreeSize() > m_Blocks[i]->m_pMetadata->GetSumFreeSize()) { VMA_SWAP(m_Blocks[i - 1], m_Blocks[i]); return; @@ -11719,9 +12729,17 @@ void VmaBlockVector::IncrementallySortBlocks() } } +void VmaBlockVector::SortByFreeSize() +{ + VMA_SORT(m_Blocks.begin(), m_Blocks.end(), + [](VmaDeviceMemoryBlock* b1, VmaDeviceMemoryBlock* b2) -> bool + { + return b1->m_pMetadata->GetSumFreeSize() < b2->m_pMetadata->GetSumFreeSize(); + }); +} + VkResult VmaBlockVector::AllocateFromBlock( VmaDeviceMemoryBlock* pBlock, - uint32_t currentFrameIndex, VkDeviceSize size, VkDeviceSize alignment, VmaAllocationCreateFlags allocFlags, @@ -11730,65 +12748,75 @@ VkResult VmaBlockVector::AllocateFromBlock( 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, + if (pBlock->m_pMetadata->CreateAllocationRequest( 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; - } - } - - *pAllocation = m_hAllocator->m_AllocationObjectAllocator.Allocate(currentFrameIndex, isUserDataString); - pBlock->m_pMetadata->Alloc(currRequest, suballocType, size, *pAllocation); - UpdateHasEmptyBlock(); - (*pAllocation)->InitBlockAllocation( - pBlock, - currRequest.offset, - alignment, - size, - m_MemoryTypeIndex, - suballocType, - mapped, - (allocFlags & VMA_ALLOCATION_CREATE_CAN_BECOME_LOST_BIT) != 0); - VMA_HEAVY_ASSERT(pBlock->Validate()); - (*pAllocation)->SetUserData(m_hAllocator, pUserData); - m_hAllocator->m_Budget.AddAllocation(m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex), size); - 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); - VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); - } - return VK_SUCCESS; + return CommitAllocationRequest(currRequest, pBlock, alignment, allocFlags, pUserData, suballocType, pAllocation); } return VK_ERROR_OUT_OF_DEVICE_MEMORY; } +VkResult VmaBlockVector::CommitAllocationRequest( + VmaAllocationRequest& allocRequest, + VmaDeviceMemoryBlock* pBlock, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation) +{ + const bool mapped = (allocFlags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0; + const bool isUserDataString = (allocFlags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0; + const bool isMappingAllowed = (allocFlags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0; + + pBlock->PostAlloc(); + // Allocate from pCurrBlock. + if (mapped) + { + VkResult res = pBlock->Map(m_hAllocator, 1, VMA_NULL); + if (res != VK_SUCCESS) + { + return res; + } + } + + *pAllocation = m_hAllocator->m_AllocationObjectAllocator.Allocate(isMappingAllowed); + pBlock->m_pMetadata->Alloc(allocRequest, suballocType, *pAllocation); + (*pAllocation)->InitBlockAllocation( + pBlock, + allocRequest.allocHandle, + alignment, + allocRequest.size, // Not size, as actual allocation size may be larger than requested! + m_MemoryTypeIndex, + suballocType, + mapped); + VMA_HEAVY_ASSERT(pBlock->Validate()); + if (isUserDataString) + (*pAllocation)->SetName(m_hAllocator, (const char*)pUserData); + else + (*pAllocation)->SetUserData(m_hAllocator, pUserData); + m_hAllocator->m_Budget.AddAllocation(m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex), allocRequest.size); + if (VMA_DEBUG_INITIALIZE_ALLOCATIONS) + { + m_hAllocator->FillAllocation(*pAllocation, VMA_ALLOCATION_FILL_PATTERN_CREATED); + } + if (IsCorruptionDetectionEnabled()) + { + VkResult res = pBlock->WriteMagicValueAfterAllocation(m_hAllocator, (*pAllocation)->GetOffset(), allocRequest.size); + VMA_ASSERT(res == VK_SUCCESS && "Couldn't map block memory to write magic value."); + } + return VK_SUCCESS; +} + VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIndex) { VkMemoryAllocateInfo allocInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO }; @@ -11799,35 +12827,36 @@ VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIn #if VMA_BUFFER_DEVICE_ADDRESS // Every standalone block can potentially contain a buffer with VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT - always enable the feature. VkMemoryAllocateFlagsInfoKHR allocFlagsInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO_KHR }; - if(m_hAllocator->m_UseKhrBufferDeviceAddress) + if (m_hAllocator->m_UseKhrBufferDeviceAddress) { allocFlagsInfo.flags = VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT_KHR; VmaPnextChainPushFront(&allocInfo, &allocFlagsInfo); } -#endif // #if VMA_BUFFER_DEVICE_ADDRESS +#endif // VMA_BUFFER_DEVICE_ADDRESS #if VMA_MEMORY_PRIORITY VkMemoryPriorityAllocateInfoEXT priorityInfo = { VK_STRUCTURE_TYPE_MEMORY_PRIORITY_ALLOCATE_INFO_EXT }; - if(m_hAllocator->m_UseExtMemoryPriority) + if (m_hAllocator->m_UseExtMemoryPriority) { + VMA_ASSERT(m_Priority >= 0.f && m_Priority <= 1.f); priorityInfo.priority = m_Priority; VmaPnextChainPushFront(&allocInfo, &priorityInfo); } -#endif // #if VMA_MEMORY_PRIORITY +#endif // VMA_MEMORY_PRIORITY #if VMA_EXTERNAL_MEMORY // Attach VkExportMemoryAllocateInfoKHR if necessary. VkExportMemoryAllocateInfoKHR exportMemoryAllocInfo = { VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO_KHR }; exportMemoryAllocInfo.handleTypes = m_hAllocator->GetExternalMemoryHandleTypeFlags(m_MemoryTypeIndex); - if(exportMemoryAllocInfo.handleTypes != 0) + if (exportMemoryAllocInfo.handleTypes != 0) { VmaPnextChainPushFront(&allocInfo, &exportMemoryAllocInfo); } -#endif // #if VMA_EXTERNAL_MEMORY +#endif // VMA_EXTERNAL_MEMORY VkDeviceMemory mem = VK_NULL_HANDLE; VkResult res = m_hAllocator->AllocateVulkanMemory(&allocInfo, &mem); - if(res < 0) + if (res < 0) { return res; } @@ -11843,10 +12872,11 @@ VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIn mem, allocInfo.allocationSize, m_NextBlockId++, - m_Algorithm); + m_Algorithm, + m_BufferImageGranularity); m_Blocks.push_back(pBlock); - if(pNewBlockIndex != VMA_NULL) + if (pNewBlockIndex != VMA_NULL) { *pNewBlockIndex = m_Blocks.size() - 1; } @@ -11854,532 +12884,12 @@ VkResult VmaBlockVector::CreateBlock(VkDeviceSize blockSize, size_t* pNewBlockIn return VK_SUCCESS; } -void VmaBlockVector::ApplyDefragmentationMovesCpu( - class VmaBlockVectorDefragmentationContext* pDefragCtx, - const VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves) +bool VmaBlockVector::HasEmptyBlock() { - 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(blockCount, BlockInfo(), VmaStlAllocator(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 = { 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(dstBlockInfo.pMappedData) + move.dstOffset, - reinterpret_cast(srcBlockInfo.pMappedData) + move.srcOffset, - static_cast(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, - VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves, - VkCommandBuffer commandBuffer) -{ - const size_t blockCount = m_Blocks.size(); - - pDefragCtx->blockContexts.resize(blockCount); - memset(pDefragCtx->blockContexts.data(), 0, blockCount * sizeof(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]; - - //if(move.type == VMA_ALLOCATION_TYPE_UNKNOWN) - { - // Old school move still require us to map the whole block - 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; - VmaFillGpuDefragmentationBufferCreateInfo(bufCreateInfo); - - 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) - { - 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) -{ - 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 - { - break; - } - } - } - UpdateHasEmptyBlock(); -} - -void VmaBlockVector::UpdateHasEmptyBlock() -{ - m_HasEmptyBlock = false; - for(size_t index = 0, count = m_Blocks.size(); index < count; ++index) + for (size_t index = 0, count = m_Blocks.size(); index < count; ++index) { VmaDeviceMemoryBlock* const pBlock = m_Blocks[index]; - if(pBlock->m_pMetadata->IsEmpty()) - { - m_HasEmptyBlock = true; - break; - } - } -} - -#if VMA_STATS_STRING_ENABLED - -void VmaBlockVector::PrintDetailedMap(class VmaJsonWriter& json) -{ - VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); - - json.BeginObject(); - - if(IsCustomPool()) - { - const char* poolName = m_hParentPool->GetName(); - if(poolName != VMA_NULL && poolName[0] != '\0') - { - json.WriteString("Name"); - json.WriteString(poolName); - } - - 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, VmaDefragmentationFlags flags, - 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 canDefragmentOnCpu = maxCpuBytesToMove > 0 && maxCpuAllocationsToMove > 0 && - isHostVisible; - const bool canDefragmentOnGpu = maxGpuBytesToMove > 0 && maxGpuAllocationsToMove > 0 && - !IsCorruptionDetectionEnabled() && - ((1u << m_MemoryTypeIndex) & m_hAllocator->GetGpuDefragmentationMemoryTypeBits()) != 0; - - // 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) - { - if(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL) - { - if(!m_Mutex.TryLockWrite()) - { - pCtx->res = VK_ERROR_INITIALIZATION_FAILED; - return; - } - } - else - { - m_Mutex.LockWrite(); - pCtx->mutexLocked = true; - } - } - - pCtx->Begin(overlappingMoveSupported, flags); - - // Defragment. - - const VkDeviceSize maxBytesToMove = defragmentOnGpu ? maxGpuBytesToMove : maxCpuBytesToMove; - const uint32_t maxAllocationsToMove = defragmentOnGpu ? maxGpuAllocationsToMove : maxCpuAllocationsToMove; - pCtx->res = pCtx->GetAlgorithm()->Defragment(pCtx->defragmentationMoves, maxBytesToMove, maxAllocationsToMove, flags); - - // 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(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL) - { - if(m_hAllocator->m_UseMutex) - m_Mutex.UnlockWrite(); - - if(pCtx->res >= VK_SUCCESS && !pCtx->defragmentationMoves.empty()) - pCtx->res = VK_NOT_READY; - - return; - } - - if(pCtx->res >= VK_SUCCESS) - { - if(defragmentOnGpu) - { - ApplyDefragmentationMovesGpu(pCtx, pCtx->defragmentationMoves, commandBuffer); - } - else - { - ApplyDefragmentationMovesCpu(pCtx, pCtx->defragmentationMoves); - } - } - } -} - -void VmaBlockVector::DefragmentationEnd( - class VmaBlockVectorDefragmentationContext* pCtx, - uint32_t flags, - VmaDefragmentationStats* pStats) -{ - if(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL && m_hAllocator->m_UseMutex) - { - VMA_ASSERT(pCtx->mutexLocked == false); - - // Incremental defragmentation doesn't hold the lock, so when we enter here we don't actually have any - // lock protecting us. Since we mutate state here, we have to take the lock out now - m_Mutex.LockWrite(); - pCtx->mutexLocked = true; - } - - // If the mutex isn't locked we didn't do any work and there is nothing to delete. - if(pCtx->mutexLocked || !m_hAllocator->m_UseMutex) - { - // 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(); - } -} - -uint32_t VmaBlockVector::ProcessDefragmentations( - class VmaBlockVectorDefragmentationContext *pCtx, - VmaDefragmentationPassMoveInfo* pMove, uint32_t maxMoves) -{ - VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); - - const uint32_t moveCount = VMA_MIN(uint32_t(pCtx->defragmentationMoves.size()) - pCtx->defragmentationMovesProcessed, maxMoves); - - for(uint32_t i = 0; i < moveCount; ++ i) - { - VmaDefragmentationMove& move = pCtx->defragmentationMoves[pCtx->defragmentationMovesProcessed + i]; - - pMove->allocation = move.hAllocation; - pMove->memory = move.pDstBlock->GetDeviceMemory(); - pMove->offset = move.dstOffset; - - ++ pMove; - } - - pCtx->defragmentationMovesProcessed += moveCount; - - return moveCount; -} - -void VmaBlockVector::CommitDefragmentations( - class VmaBlockVectorDefragmentationContext *pCtx, - VmaDefragmentationStats* pStats) -{ - VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); - - for(uint32_t i = pCtx->defragmentationMovesCommitted; i < pCtx->defragmentationMovesProcessed; ++ i) - { - const VmaDefragmentationMove &move = pCtx->defragmentationMoves[i]; - - move.pSrcBlock->m_pMetadata->FreeAtOffset(move.srcOffset); - move.hAllocation->ChangeBlockAllocation(m_hAllocator, move.pDstBlock, move.dstOffset); - } - - pCtx->defragmentationMovesCommitted = pCtx->defragmentationMovesProcessed; - FreeEmptyBlocks(pStats); -} - -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)) + if (pBlock->m_pMetadata->IsEmpty()) { return true; } @@ -12387,38 +12897,44 @@ bool VmaBlockVector::IsBufferImageGranularityConflictPossible() const return false; } -void VmaBlockVector::MakePoolAllocationsLost( - uint32_t currentFrameIndex, - size_t* pLostAllocationCount) +#if VMA_STATS_STRING_ENABLED +void VmaBlockVector::PrintDetailedMap(class VmaJsonWriter& json) { - VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); - size_t lostAllocationCount = 0; - for(uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + + json.BeginObject(); + for (size_t i = 0; i < m_Blocks.size(); ++i) { - VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; - VMA_ASSERT(pBlock); - lostAllocationCount += pBlock->m_pMetadata->MakeAllocationsLost(currentFrameIndex, m_FrameInUseCount); - } - if(pLostAllocationCount != VMA_NULL) - { - *pLostAllocationCount = lostAllocationCount; + json.BeginString(); + json.ContinueString(m_Blocks[i]->GetId()); + json.EndString(); + + json.BeginObject(); + json.WriteString("MapRefCount"); + json.WriteNumber(m_Blocks[i]->GetMapRefCount()); + + m_Blocks[i]->m_pMetadata->PrintDetailedMap(json); + json.EndObject(); } + json.EndObject(); } +#endif // VMA_STATS_STRING_ENABLED VkResult VmaBlockVector::CheckCorruption() { - if(!IsCorruptionDetectionEnabled()) + 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) + 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) + if (res != VK_SUCCESS) { return res; } @@ -12426,1755 +12942,1029 @@ VkResult VmaBlockVector::CheckCorruption() return VK_SUCCESS; } -void VmaBlockVector::AddStats(VmaStats* pStats) -{ - const uint32_t memTypeIndex = m_MemoryTypeIndex; - const uint32_t memHeapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(memTypeIndex); +#endif // _VMA_BLOCK_VECTOR_FUNCTIONS - 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( +#ifndef _VMA_DEFRAGMENTATION_CONTEXT_FUNCTIONS +VmaDefragmentationContext_T::VmaDefragmentationContext_T( 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(hAllocator->GetAllocationCallbacks())) + const VmaDefragmentationInfo& info) + : m_MaxPassBytes(info.maxBytesPerPass == 0 ? VK_WHOLE_SIZE : info.maxBytesPerPass), + m_MaxPassAllocations(info.maxAllocationsPerPass == 0 ? UINT32_MAX : info.maxAllocationsPerPass), + m_MoveAllocator(hAllocator->GetAllocationCallbacks()), + m_Moves(m_MoveAllocator) { - // Create block info for each block. - const size_t blockCount = m_pBlockVector->m_Blocks.size(); - for(size_t blockIndex = 0; blockIndex < blockCount; ++blockIndex) + m_Algorithm = info.flags & VMA_DEFRAGMENTATION_FLAG_ALGORITHM_MASK; + + if (info.pool != VMA_NULL) { - 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); + m_BlockVectorCount = 1; + m_PoolBlockVector = &info.pool->m_BlockVector; + m_pBlockVectors = &m_PoolBlockVector; + m_PoolBlockVector->SetIncrementalSort(false); + m_PoolBlockVector->SortByFreeSize(); } - - // 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--; ) + else { - 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) + m_BlockVectorCount = hAllocator->GetMemoryTypeCount(); + m_PoolBlockVector = VMA_NULL; + m_pBlockVectors = hAllocator->m_pBlockVectors; + for (uint32_t i = 0; i < m_BlockVectorCount; ++i) { - AllocationInfo allocInfo = AllocationInfo(hAlloc, pChanged); - (*it)->m_Allocations.push_back(allocInfo); + VmaBlockVector* vector = m_pBlockVectors[i]; + if (vector != VMA_NULL) + { + vector->SetIncrementalSort(false); + vector->SortByFreeSize(); + } + } + } + + switch (m_Algorithm) + { + case 0: // Default algorithm + m_Algorithm = VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT; + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT: + { + m_AlgorithmState = vma_new_array(hAllocator, StateBalanced, m_BlockVectorCount); + break; + } + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT: + { + if (hAllocator->GetBufferImageGranularity() > 1) + { + m_AlgorithmState = vma_new_array(hAllocator, StateExtensive, m_BlockVectorCount); + } + break; + } + } +} + +VmaDefragmentationContext_T::~VmaDefragmentationContext_T() +{ + if (m_PoolBlockVector != VMA_NULL) + { + m_PoolBlockVector->SetIncrementalSort(true); + } + else + { + for (uint32_t i = 0; i < m_BlockVectorCount; ++i) + { + VmaBlockVector* vector = m_pBlockVectors[i]; + if (vector != VMA_NULL) + vector->SetIncrementalSort(true); + } + } + + if (m_AlgorithmState) + { + switch (m_Algorithm) + { + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT: + vma_delete_array(m_MoveAllocator.m_pCallbacks, reinterpret_cast(m_AlgorithmState), m_BlockVectorCount); + break; + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT: + vma_delete_array(m_MoveAllocator.m_pCallbacks, reinterpret_cast(m_AlgorithmState), m_BlockVectorCount); + break; + default: + VMA_ASSERT(0); + } + } +} + +VkResult VmaDefragmentationContext_T::DefragmentPassBegin(VmaDefragmentationPassMoveInfo& moveInfo) +{ + if (m_PoolBlockVector != VMA_NULL) + { + VmaMutexLockWrite lock(m_PoolBlockVector->GetMutex(), m_PoolBlockVector->GetAllocator()->m_UseMutex); + + if (m_PoolBlockVector->GetBlockCount() > 1) + ComputeDefragmentation(*m_PoolBlockVector, 0); + else if (m_PoolBlockVector->GetBlockCount() == 1) + ReallocWithinBlock(*m_PoolBlockVector, m_PoolBlockVector->GetBlock(0)); + } + else + { + for (uint32_t i = 0; i < m_BlockVectorCount; ++i) + { + if (m_pBlockVectors[i] != VMA_NULL) + { + VmaMutexLockWrite lock(m_pBlockVectors[i]->GetMutex(), m_pBlockVectors[i]->GetAllocator()->m_UseMutex); + + if (m_pBlockVectors[i]->GetBlockCount() > 1) + { + if (ComputeDefragmentation(*m_pBlockVectors[i], i)) + break; + } + else if (m_pBlockVectors[i]->GetBlockCount() == 1) + { + if (ReallocWithinBlock(*m_pBlockVectors[i], m_pBlockVectors[i]->GetBlock(0))) + break; + } + } + } + } + + moveInfo.moveCount = static_cast(m_Moves.size()); + if (moveInfo.moveCount > 0) + { + moveInfo.pMoves = m_Moves.data(); + return VK_INCOMPLETE; + } + + moveInfo.pMoves = VMA_NULL; + return VK_SUCCESS; +} + +VkResult VmaDefragmentationContext_T::DefragmentPassEnd(VmaDefragmentationPassMoveInfo& moveInfo) +{ + VMA_ASSERT(moveInfo.moveCount > 0 ? moveInfo.pMoves != VMA_NULL : true); + + VkResult result = VK_SUCCESS; + VmaStlAllocator blockAllocator(m_MoveAllocator.m_pCallbacks); + VmaVector> immovableBlocks(blockAllocator); + VmaVector> mappedBlocks(blockAllocator); + + VmaAllocator allocator = VMA_NULL; + for (uint32_t i = 0; i < moveInfo.moveCount; ++i) + { + VmaDefragmentationMove& move = moveInfo.pMoves[i]; + size_t prevCount = 0, currentCount = 0; + VkDeviceSize freedBlockSize = 0; + + uint32_t vectorIndex; + VmaBlockVector* vector; + if (m_PoolBlockVector != VMA_NULL) + { + vectorIndex = 0; + vector = m_PoolBlockVector; } else { + vectorIndex = move.srcAllocation->GetMemoryTypeIndex(); + vector = m_pBlockVectors[vectorIndex]; + VMA_ASSERT(vector != VMA_NULL); + } + + switch (move.operation) + { + case VMA_DEFRAGMENTATION_MOVE_OPERATION_COPY: + { + uint8_t mapCount = move.srcAllocation->SwapBlockAllocation(vector->m_hAllocator, move.dstTmpAllocation); + if (mapCount > 0) + { + allocator = vector->m_hAllocator; + VmaDeviceMemoryBlock* newMapBlock = move.srcAllocation->GetBlock(); + bool notPresent = true; + for (FragmentedBlock& block : mappedBlocks) + { + if (block.block == newMapBlock) + { + notPresent = false; + block.data += mapCount; + break; + } + } + if (notPresent) + mappedBlocks.push_back({ mapCount, newMapBlock }); + } + + // Scope for locks, Free have it's own lock + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + prevCount = vector->GetBlockCount(); + freedBlockSize = move.dstTmpAllocation->GetBlock()->m_pMetadata->GetSize(); + } + vector->Free(move.dstTmpAllocation); + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + currentCount = vector->GetBlockCount(); + } + + result = VK_INCOMPLETE; + break; + } + case VMA_DEFRAGMENTATION_MOVE_OPERATION_IGNORE: + { + m_PassStats.bytesMoved -= move.srcAllocation->GetSize(); + --m_PassStats.allocationsMoved; + vector->Free(move.dstTmpAllocation); + + VmaDeviceMemoryBlock* newBlock = move.srcAllocation->GetBlock(); + bool notPresent = true; + for (const FragmentedBlock& block : immovableBlocks) + { + if (block.block == newBlock) + { + notPresent = false; + break; + } + } + if (notPresent) + immovableBlocks.push_back({ vectorIndex, newBlock }); + break; + } + case VMA_DEFRAGMENTATION_MOVE_OPERATION_DESTROY: + { + m_PassStats.bytesMoved -= move.srcAllocation->GetSize(); + --m_PassStats.allocationsMoved; + // Scope for locks, Free have it's own lock + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + prevCount = vector->GetBlockCount(); + freedBlockSize = move.srcAllocation->GetBlock()->m_pMetadata->GetSize(); + } + vector->Free(move.srcAllocation); + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + currentCount = vector->GetBlockCount(); + } + freedBlockSize *= prevCount - currentCount; + + VkDeviceSize dstBlockSize; + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + dstBlockSize = move.dstTmpAllocation->GetBlock()->m_pMetadata->GetSize(); + } + vector->Free(move.dstTmpAllocation); + { + VmaMutexLockRead lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + freedBlockSize += dstBlockSize * (currentCount - vector->GetBlockCount()); + currentCount = vector->GetBlockCount(); + } + + result = VK_INCOMPLETE; + break; + } + default: VMA_ASSERT(0); } - ++m_AllocationCount; - } -} - -VkResult VmaDefragmentationAlgorithm_Generic::DefragmentRound( - VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - bool freeOldAllocations) -{ - 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) + if (prevCount > currentCount) { - 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; - } + size_t freedBlocks = prevCount - currentCount; + m_PassStats.deviceMemoryBlocksFreed += static_cast(freedBlocks); + m_PassStats.bytesFreed += freedBlockSize; } - 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) + switch (m_Algorithm) { - 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)) + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT: + { + if (m_AlgorithmState != VMA_NULL) { - VMA_ASSERT(dstAllocRequest.itemsToMakeLostCount == 0); - - // Reached limit on number of allocations or bytes to move. - if((m_AllocationsMoved + 1 > maxAllocationsToMove) || - (m_BytesMoved + size > maxBytesToMove)) + // Avoid unnecessary tries to allocate when new free block is avaiable + StateExtensive& state = reinterpret_cast(m_AlgorithmState)[vectorIndex]; + if (state.firstFreeBlock != SIZE_MAX) { - return VK_SUCCESS; + const size_t diff = prevCount - currentCount; + if (state.firstFreeBlock >= diff) + { + state.firstFreeBlock -= diff; + if (state.firstFreeBlock != 0) + state.firstFreeBlock -= vector->GetBlock(state.firstFreeBlock - 1)->m_pMetadata->IsEmpty(); + } + else + state.firstFreeBlock = 0; } + } + } + } + } + moveInfo.moveCount = 0; + moveInfo.pMoves = VMA_NULL; + m_Moves.clear(); - VmaDefragmentationMove move = {}; - move.srcBlockIndex = pSrcBlockInfo->m_OriginalBlockIndex; - move.dstBlockIndex = pDstBlockInfo->m_OriginalBlockIndex; - move.srcOffset = srcOffset; - move.dstOffset = dstAllocRequest.offset; - move.size = size; - move.hAllocation = allocInfo.m_hAllocation; - move.pSrcBlock = pSrcBlockInfo->m_pBlock; - move.pDstBlock = pDstBlockInfo->m_pBlock; + // Update stats + m_GlobalStats.allocationsMoved += m_PassStats.allocationsMoved; + m_GlobalStats.bytesFreed += m_PassStats.bytesFreed; + m_GlobalStats.bytesMoved += m_PassStats.bytesMoved; + m_GlobalStats.deviceMemoryBlocksFreed += m_PassStats.deviceMemoryBlocksFreed; + m_PassStats = { 0 }; - moves.push_back(move); - - pDstBlockInfo->m_pBlock->m_pMetadata->Alloc( - dstAllocRequest, - suballocType, - size, - allocInfo.m_hAllocation); - - if(freeOldAllocations) + // Move blocks with immovable allocations according to algorithm + if (immovableBlocks.size() > 0) + { + switch (m_Algorithm) + { + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT: + { + if (m_AlgorithmState != VMA_NULL) + { + bool swapped = false; + // Move to the start of free blocks range + for (const FragmentedBlock& block : immovableBlocks) { - pSrcBlockInfo->m_pBlock->m_pMetadata->FreeAtOffset(srcOffset); - allocInfo.m_hAllocation->ChangeBlockAllocation(m_hAllocator, pDstBlockInfo->m_pBlock, dstAllocRequest.offset); + StateExtensive& state = reinterpret_cast(m_AlgorithmState)[block.data]; + if (state.operation != StateExtensive::Operation::Cleanup) + { + VmaBlockVector* vector = m_pBlockVectors[block.data]; + VmaMutexLockWrite lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + + for (size_t i = 0, count = vector->GetBlockCount() - m_ImmovableBlockCount; i < count; ++i) + { + if (vector->GetBlock(i) == block.block) + { + VMA_SWAP(vector->m_Blocks[i], vector->m_Blocks[vector->GetBlockCount() - ++m_ImmovableBlockCount]); + if (state.firstFreeBlock != SIZE_MAX) + { + if (i + 1 < state.firstFreeBlock) + { + if (state.firstFreeBlock > 1) + VMA_SWAP(vector->m_Blocks[i], vector->m_Blocks[--state.firstFreeBlock]); + else + --state.firstFreeBlock; + } + } + swapped = true; + break; + } + } + } } - - if(allocInfo.m_pChanged != VMA_NULL) - { - *allocInfo.m_pChanged = VK_TRUE; - } - - ++m_AllocationsMoved; - m_BytesMoved += size; - - VmaVectorRemove(pSrcBlockInfo->m_Allocations, srcAllocIndex); - + if (swapped) + result = VK_INCOMPLETE; break; } } - - // If not processed, this allocInfo remains in pBlockInfo->m_Allocations for next round. - - if(srcAllocIndex > 0) + default: { - --srcAllocIndex; - } - else - { - if(srcBlockIndex > 0) + // Move to the begining + for (const FragmentedBlock& block : immovableBlocks) { - --srcBlockIndex; - srcAllocIndex = SIZE_MAX; - } - else - { - return VK_SUCCESS; - } - } - } -} + VmaBlockVector* vector = m_pBlockVectors[block.data]; + VmaMutexLockWrite lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); -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 >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - VmaDefragmentationFlags flags) -{ - 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) + for (size_t i = m_ImmovableBlockCount; i < vector->GetBlockCount(); ++i) { - AllocationInfo allocInfo = AllocationInfo(it->hAllocation, VMA_NULL); - pBlockInfo->m_Allocations.push_back(allocInfo); + if (vector->GetBlock(i) == block.block) + { + VMA_SWAP(vector->m_Blocks[i], vector->m_Blocks[m_ImmovableBlockCount++]); + break; + } } } + break; + } } - - 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) + // Bulk-map destination blocks + for (const FragmentedBlock& block : mappedBlocks) { - result = DefragmentRound(moves, maxBytesToMove, maxAllocationsToMove, !(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL)); + VkResult res = block.block->Map(allocator, block.data, VMA_NULL); + VMA_ASSERT(res == VK_SUCCESS); } - return result; } -bool VmaDefragmentationAlgorithm_Generic::MoveMakesSense( - size_t dstBlockIndex, VkDeviceSize dstOffset, - size_t srcBlockIndex, VkDeviceSize srcOffset) +bool VmaDefragmentationContext_T::ComputeDefragmentation(VmaBlockVector& vector, size_t index) { - if(dstBlockIndex < srcBlockIndex) + switch (m_Algorithm) { - return true; + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT: + return ComputeDefragmentation_Fast(vector); + default: + VMA_ASSERT(0); + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_BALANCED_BIT: + return ComputeDefragmentation_Balanced(vector, index, true); + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FULL_BIT: + return ComputeDefragmentation_Full(vector); + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT: + return ComputeDefragmentation_Extensive(vector, index); } - if(dstBlockIndex > srcBlockIndex) +} + +VmaDefragmentationContext_T::MoveAllocationData VmaDefragmentationContext_T::GetMoveData( + VmaAllocHandle handle, VmaBlockMetadata* metadata) +{ + MoveAllocationData moveData; + moveData.move.srcAllocation = (VmaAllocation)metadata->GetAllocationUserData(handle); + moveData.size = moveData.move.srcAllocation->GetSize(); + moveData.alignment = moveData.move.srcAllocation->GetAlignment(); + moveData.type = moveData.move.srcAllocation->GetSuballocationType(); + moveData.flags = 0; + + if (moveData.move.srcAllocation->IsPersistentMap()) + moveData.flags |= VMA_ALLOCATION_CREATE_MAPPED_BIT; + if (moveData.move.srcAllocation->IsMappingAllowed()) + moveData.flags |= VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT; + + return moveData; +} + +VmaDefragmentationContext_T::CounterStatus VmaDefragmentationContext_T::CheckCounters(VkDeviceSize bytes) +{ + // Ignore allocation if will exceed max size for copy + if (m_PassStats.bytesMoved + bytes > m_MaxPassBytes) { - return false; + if (++m_IgnoredAllocs < MAX_ALLOCS_TO_IGNORE) + return CounterStatus::Ignore; + else + return CounterStatus::End; } - if(dstOffset < srcOffset) + return CounterStatus::Pass; +} + +bool VmaDefragmentationContext_T::IncrementCounters(VkDeviceSize bytes) +{ + m_PassStats.bytesMoved += bytes; + // Early return when max found + if (++m_PassStats.allocationsMoved >= m_MaxPassAllocations || m_PassStats.bytesMoved >= m_MaxPassBytes) { + VMA_ASSERT(m_PassStats.allocationsMoved == m_MaxPassAllocations || + m_PassStats.bytesMoved == m_MaxPassBytes && "Exceeded maximal pass threshold!"); 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(hAllocator->GetAllocationCallbacks())) +bool VmaDefragmentationContext_T::ReallocWithinBlock(VmaBlockVector& vector, VmaDeviceMemoryBlock* block) { - VMA_ASSERT(VMA_DEBUG_MARGIN == 0); + VmaBlockMetadata* metadata = block->m_pMetadata; -} - -VmaDefragmentationAlgorithm_Fast::~VmaDefragmentationAlgorithm_Fast() -{ -} - -VkResult VmaDefragmentationAlgorithm_Fast::Defragment( - VmaVector< VmaDefragmentationMove, VmaStlAllocator >& moves, - VkDeviceSize maxBytesToMove, - uint32_t maxAllocationsToMove, - VmaDefragmentationFlags flags) -{ - VMA_ASSERT(m_AllAllocations || m_pBlockVector->CalcAllocationCount() == m_AllocationCount); - - const size_t blockCount = m_pBlockVector->GetBlockCount(); - if(blockCount == 0 || maxBytesToMove == 0 || maxAllocationsToMove == 0) + for (VmaAllocHandle handle = metadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = metadata->GetNextAllocation(handle)) { - 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(); ) + MoveAllocationData moveData = GetMoveData(handle, metadata); + // Ignore newly created allocations by defragmentation algorithm + if (moveData.move.srcAllocation->GetUserData() == this) + continue; + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) { - VmaAllocation_T* const pAlloc = srcSuballocIt->hAllocation; - const VkDeviceSize srcAllocAlignment = pAlloc->GetAlignment(); - const VkDeviceSize srcAllocSize = srcSuballocIt->size; - if(m_AllocationsMoved == maxAllocationsToMove || - m_BytesMoved + srcAllocSize > maxBytesToMove) + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; + } + + VkDeviceSize offset = moveData.move.srcAllocation->GetOffset(); + if (offset != 0 && metadata->GetSumFreeSize() >= moveData.size) + { + VmaAllocationRequest request = {}; + if (metadata->CreateAllocationRequest( + moveData.size, + moveData.alignment, + false, + moveData.type, + VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT, + &request)) { - end = true; + if (metadata->GetAllocationOffset(request.allocHandle) < offset) + { + if (vector.CommitAllocationRequest( + request, + block, + moveData.alignment, + moveData.flags, + this, + moveData.type, + &moveData.move.dstTmpAllocation) == VK_SUCCESS) + { + m_Moves.push_back(moveData.move); + if (IncrementCounters(moveData.size)) + return true; + } + } + } + } + } + return false; +} + +bool VmaDefragmentationContext_T::AllocInOtherBlock(size_t start, size_t end, MoveAllocationData& data, VmaBlockVector& vector) +{ + for (; start < end; ++start) + { + VmaDeviceMemoryBlock* dstBlock = vector.GetBlock(start); + if (dstBlock->m_pMetadata->GetSumFreeSize() >= data.size) + { + if (vector.AllocateFromBlock(dstBlock, + data.size, + data.alignment, + data.flags, + this, + data.type, + 0, + &data.move.dstTmpAllocation) == VK_SUCCESS) + { + m_Moves.push_back(data.move); + if (IncrementCounters(data.size)) + return true; break; } - const VkDeviceSize srcAllocOffset = srcSuballocIt->offset; + } + } + return false; +} - VmaDefragmentationMove move = {}; - // Try to place it in one of free spaces from the database. - size_t freeSpaceInfoIndex; - VkDeviceSize dstAllocOffset; - if(freeSpaceDb.Fetch(srcAllocAlignment, srcAllocSize, - freeSpaceInfoIndex, dstAllocOffset)) +bool VmaDefragmentationContext_T::ComputeDefragmentation_Fast(VmaBlockVector& vector) +{ + // Move only between blocks + + // Go through allocations in last blocks and try to fit them inside first ones + for (size_t i = vector.GetBlockCount() - 1; i > m_ImmovableBlockCount; --i) + { + VmaBlockMetadata* metadata = vector.GetBlock(i)->m_pMetadata; + + for (VmaAllocHandle handle = metadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = metadata->GetNextAllocation(handle)) + { + MoveAllocationData moveData = GetMoveData(handle, metadata); + // Ignore newly created allocations by defragmentation algorithm + if (moveData.move.srcAllocation->GetUserData() == this) + continue; + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) { - size_t freeSpaceOrigBlockIndex = m_BlockInfos[freeSpaceInfoIndex].origBlockIndex; - VmaDeviceMemoryBlock* pFreeSpaceBlock = m_pBlockVector->GetBlock(freeSpaceOrigBlockIndex); - VmaBlockMetadata_Generic* pFreeSpaceMetadata = (VmaBlockMetadata_Generic*)pFreeSpaceBlock->m_pMetadata; - - // 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); - - move.srcBlockIndex = srcOrigBlockIndex; - move.dstBlockIndex = freeSpaceOrigBlockIndex; - move.srcOffset = srcAllocOffset; - move.dstOffset = dstAllocOffset; - move.size = 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); - - move.srcBlockIndex = srcOrigBlockIndex; - move.dstBlockIndex = freeSpaceOrigBlockIndex; - move.srcOffset = srcAllocOffset; - move.dstOffset = dstAllocOffset; - move.size = srcAllocSize; - - moves.push_back(move); - } + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; } - else + + // Check all previous blocks for free space + if (AllocInOtherBlock(0, i, moveData, vector)) + return true; + } + } + return false; +} + +bool VmaDefragmentationContext_T::ComputeDefragmentation_Balanced(VmaBlockVector& vector, size_t index, bool update) +{ + // Go over every allocation and try to fit it in previous blocks at lowest offsets, + // if not possible: realloc within single block to minimize offset (exclude offset == 0), + // but only if there are noticable gaps between them (some heuristic, ex. average size of allocation in block) + VMA_ASSERT(m_AlgorithmState != VMA_NULL); + + StateBalanced& vectorState = reinterpret_cast(m_AlgorithmState)[index]; + if (update && vectorState.avgAllocSize == UINT64_MAX) + UpdateVectorStatistics(vector, vectorState); + + const size_t startMoveCount = m_Moves.size(); + VkDeviceSize minimalFreeRegion = vectorState.avgFreeSize / 2; + for (size_t i = vector.GetBlockCount() - 1; i > m_ImmovableBlockCount; --i) + { + VmaDeviceMemoryBlock* block = vector.GetBlock(i); + VmaBlockMetadata* metadata = block->m_pMetadata; + VkDeviceSize prevFreeRegionSize = 0; + + for (VmaAllocHandle handle = metadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = metadata->GetNextAllocation(handle)) + { + MoveAllocationData moveData = GetMoveData(handle, metadata); + // Ignore newly created allocations by defragmentation algorithm + if (moveData.move.srcAllocation->GetUserData() == this) + continue; + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) { - dstAllocOffset = VmaAlignUp(dstOffset, srcAllocAlignment); + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; + } - // If the allocation doesn't fit before the end of dstBlock, forward to next block. - while(dstBlockInfoIndex < srcBlockInfoIndex && - dstAllocOffset + srcAllocSize > dstBlockSize) + // Check all previous blocks for free space + const size_t prevMoveCount = m_Moves.size(); + if (AllocInOtherBlock(0, i, moveData, vector)) + return true; + + VkDeviceSize nextFreeRegionSize = metadata->GetNextFreeRegionSize(handle); + // If no room found then realloc within block for lower offset + VkDeviceSize offset = moveData.move.srcAllocation->GetOffset(); + if (prevMoveCount == m_Moves.size() && offset != 0 && metadata->GetSumFreeSize() >= moveData.size) + { + // Check if realloc will make sense + if (prevFreeRegionSize >= minimalFreeRegion || + nextFreeRegionSize >= minimalFreeRegion || + moveData.size <= vectorState.avgFreeSize || + moveData.size <= vectorState.avgAllocSize) { - // 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) + VmaAllocationRequest request = {}; + if (metadata->CreateAllocationRequest( + moveData.size, + moveData.alignment, + false, + moveData.type, + VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT, + &request)) { - // 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; - - move.srcBlockIndex = srcOrigBlockIndex; - move.dstBlockIndex = dstOrigBlockIndex; - move.srcOffset = srcAllocOffset; - move.dstOffset = dstAllocOffset; - move.size = 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); - - move.srcBlockIndex = srcOrigBlockIndex; - move.dstBlockIndex = dstOrigBlockIndex; - move.srcOffset = srcAllocOffset; - move.dstOffset = dstAllocOffset; - move.size = 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) : - res(VK_SUCCESS), - mutexLocked(false), - blockContexts(VmaStlAllocator(hAllocator->GetAllocationCallbacks())), - defragmentationMoves(VmaStlAllocator(hAllocator->GetAllocationCallbacks())), - defragmentationMovesProcessed(0), - defragmentationMovesCommitted(0), - hasDefragmentationPlan(0), - m_hAllocator(hAllocator), - m_hCustomPool(hCustomPool), - m_pBlockVector(pBlockVector), - m_CurrFrameIndex(currFrameIndex), - m_pAlgorithm(VMA_NULL), - m_Allocations(VmaStlAllocator(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, VmaDefragmentationFlags flags) -{ - 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. - - The defragmentation is not incremental - */ - if(VMA_DEBUG_MARGIN == 0 && - allAllocations && - !m_pBlockVector->IsBufferImageGranularityConflictPossible() && - !(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL)) - { - 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(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_Flags, 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_Flags, m_pStats); - vma_delete(m_hAllocator, pBlockVectorCtx); - } - } -} - -void VmaDefragmentationContext_T::AddPools(uint32_t poolCount, const 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_CustomPoolContexts.push_back(pBlockVectorDefragCtx); - } - - pBlockVectorDefragCtx->AddAll(); - } - } -} - -void VmaDefragmentationContext_T::AddAllocations( - uint32_t allocationCount, - const 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->GetBlock()->GetParentPool(); - // 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) + if (metadata->GetAllocationOffset(request.allocHandle) < offset) { - pBlockVectorDefragCtx = m_CustomPoolContexts[i]; - break; + if (vector.CommitAllocationRequest( + request, + block, + moveData.alignment, + moveData.flags, + this, + moveData.type, + &moveData.move.dstTmpAllocation) == VK_SUCCESS) + { + m_Moves.push_back(moveData.move); + if (IncrementCounters(moveData.size)) + return true; + } } } - if(!pBlockVectorDefragCtx) + } + } + prevFreeRegionSize = nextFreeRegionSize; + } + } + + // No moves perfomed, update statistics to current vector state + if (startMoveCount == m_Moves.size() && !update) + { + vectorState.avgAllocSize = UINT64_MAX; + return ComputeDefragmentation_Balanced(vector, index, false); + } + return false; +} + +bool VmaDefragmentationContext_T::ComputeDefragmentation_Full(VmaBlockVector& vector) +{ + // Go over every allocation and try to fit it in previous blocks at lowest offsets, + // if not possible: realloc within single block to minimize offset (exclude offset == 0) + + for (size_t i = vector.GetBlockCount() - 1; i > m_ImmovableBlockCount; --i) + { + VmaDeviceMemoryBlock* block = vector.GetBlock(i); + VmaBlockMetadata* metadata = block->m_pMetadata; + + for (VmaAllocHandle handle = metadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = metadata->GetNextAllocation(handle)) + { + MoveAllocationData moveData = GetMoveData(handle, metadata); + // Ignore newly created allocations by defragmentation algorithm + if (moveData.move.srcAllocation->GetUserData() == this) + continue; + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) + { + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; + } + + // Check all previous blocks for free space + const size_t prevMoveCount = m_Moves.size(); + if (AllocInOtherBlock(0, i, moveData, vector)) + return true; + + // If no room found then realloc within block for lower offset + VkDeviceSize offset = moveData.move.srcAllocation->GetOffset(); + if (prevMoveCount == m_Moves.size() && offset != 0 && metadata->GetSumFreeSize() >= moveData.size) + { + VmaAllocationRequest request = {}; + if (metadata->CreateAllocationRequest( + moveData.size, + moveData.alignment, + false, + moveData.type, + VMA_ALLOCATION_CREATE_STRATEGY_MIN_OFFSET_BIT, + &request)) + { + if (metadata->GetAllocationOffset(request.allocHandle) < offset) { - pBlockVectorDefragCtx = vma_new(m_hAllocator, VmaBlockVectorDefragmentationContext)( - m_hAllocator, - hAllocPool, - &hAllocPool->m_BlockVector, - m_CurrFrameIndex); - m_CustomPoolContexts.push_back(pBlockVectorDefragCtx); + if (vector.CommitAllocationRequest( + request, + block, + moveData.alignment, + moveData.flags, + this, + moveData.type, + &moveData.move.dstTmpAllocation) == VK_SUCCESS) + { + m_Moves.push_back(moveData.move); + if (IncrementCounters(moveData.size)) + return true; + } } } } - // This allocation belongs to default pool. - else + } + } + return false; +} + +bool VmaDefragmentationContext_T::ComputeDefragmentation_Extensive(VmaBlockVector& vector, size_t index) +{ + // First free single block, then populate it to the brim, then free another block, and so on + + // Fallback to previous algorithm since without granularity conflicts it can achieve max packing + if (vector.m_BufferImageGranularity == 1) + return ComputeDefragmentation_Full(vector); + + VMA_ASSERT(m_AlgorithmState != VMA_NULL); + + StateExtensive& vectorState = reinterpret_cast(m_AlgorithmState)[index]; + + bool texturePresent = false, bufferPresent = false, otherPresent = false; + switch (vectorState.operation) + { + case StateExtensive::Operation::Done: // Vector defragmented + return false; + case StateExtensive::Operation::FindFreeBlockBuffer: + case StateExtensive::Operation::FindFreeBlockTexture: + case StateExtensive::Operation::FindFreeBlockAll: + { + // No more blocks to free, just perform fast realloc and move to cleanup + if (vectorState.firstFreeBlock == 0) + { + vectorState.operation = StateExtensive::Operation::Cleanup; + return ComputeDefragmentation_Fast(vector); + } + + // No free blocks, have to clear last one + size_t last = (vectorState.firstFreeBlock == SIZE_MAX ? vector.GetBlockCount() : vectorState.firstFreeBlock) - 1; + VmaBlockMetadata* freeMetadata = vector.GetBlock(last)->m_pMetadata; + + const size_t prevMoveCount = m_Moves.size(); + for (VmaAllocHandle handle = freeMetadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = freeMetadata->GetNextAllocation(handle)) + { + MoveAllocationData moveData = GetMoveData(handle, freeMetadata); + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) { - const uint32_t memTypeIndex = hAlloc->GetMemoryTypeIndex(); - pBlockVectorDefragCtx = m_DefaultPoolContexts[memTypeIndex]; - if(!pBlockVectorDefragCtx) + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; + } + + // Check all previous blocks for free space + if (AllocInOtherBlock(0, last, moveData, vector)) + { + // Full clear performed already + if (prevMoveCount != m_Moves.size() && freeMetadata->GetNextAllocation(handle) == VK_NULL_HANDLE) + reinterpret_cast(m_AlgorithmState)[index] = last; + return true; + } + } + + if (prevMoveCount == m_Moves.size()) + { + // Cannot perform full clear, have to move data in other blocks around + if (last != 0) + { + for (size_t i = last - 1; i; --i) { - pBlockVectorDefragCtx = vma_new(m_hAllocator, VmaBlockVectorDefragmentationContext)( - m_hAllocator, - VMA_NULL, // hCustomPool - m_hAllocator->m_pBlockVectors[memTypeIndex], - m_CurrFrameIndex); - m_DefaultPoolContexts[memTypeIndex] = pBlockVectorDefragCtx; + if (ReallocWithinBlock(vector, vector.GetBlock(i))) + return true; } } - if(pBlockVectorDefragCtx) + if (prevMoveCount == m_Moves.size()) { - VkBool32* const pChanged = (pAllocationsChanged != VMA_NULL) ? - &pAllocationsChanged[allocIndex] : VMA_NULL; - pBlockVectorDefragCtx->AddAllocation(hAlloc, pChanged); + // No possible reallocs within blocks, try to move them around fast + return ComputeDefragmentation_Fast(vector); } } - } -} - -VkResult VmaDefragmentationContext_T::Defragment( - VkDeviceSize maxCpuBytesToMove, uint32_t maxCpuAllocationsToMove, - VkDeviceSize maxGpuBytesToMove, uint32_t maxGpuAllocationsToMove, - VkCommandBuffer commandBuffer, VmaDefragmentationStats* pStats, VmaDefragmentationFlags flags) -{ - if(pStats) - { - memset(pStats, 0, sizeof(VmaDefragmentationStats)); - } - - if(flags & VMA_DEFRAGMENTATION_FLAG_INCREMENTAL) - { - // For incremental defragmetnations, we just earmark how much we can move - // The real meat is in the defragmentation steps - m_MaxCpuBytesToMove = maxCpuBytesToMove; - m_MaxCpuAllocationsToMove = maxCpuAllocationsToMove; - - m_MaxGpuBytesToMove = maxGpuBytesToMove; - m_MaxGpuAllocationsToMove = maxGpuAllocationsToMove; - - if(m_MaxCpuBytesToMove == 0 && m_MaxCpuAllocationsToMove == 0 && - m_MaxGpuBytesToMove == 0 && m_MaxGpuAllocationsToMove == 0) - return VK_SUCCESS; - - return VK_NOT_READY; - } - - 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, flags, - 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, flags, - maxCpuBytesToMove, maxCpuAllocationsToMove, - maxGpuBytesToMove, maxGpuAllocationsToMove, - commandBuffer); - if(pBlockVectorCtx->res != VK_SUCCESS) - { - res = pBlockVectorCtx->res; - } - } - - return res; -} - -VkResult VmaDefragmentationContext_T::DefragmentPassBegin(VmaDefragmentationPassInfo* pInfo) -{ - VmaDefragmentationPassMoveInfo* pCurrentMove = pInfo->pMoves; - uint32_t movesLeft = pInfo->moveCount; - - // Process default pools. - for(uint32_t memTypeIndex = 0; - memTypeIndex < m_hAllocator->GetMemoryTypeCount(); - ++memTypeIndex) - { - VmaBlockVectorDefragmentationContext *pBlockVectorCtx = m_DefaultPoolContexts[memTypeIndex]; - if(pBlockVectorCtx) - { - VMA_ASSERT(pBlockVectorCtx->GetBlockVector()); - - if(!pBlockVectorCtx->hasDefragmentationPlan) - { - pBlockVectorCtx->GetBlockVector()->Defragment( - pBlockVectorCtx, - m_pStats, m_Flags, - m_MaxCpuBytesToMove, m_MaxCpuAllocationsToMove, - m_MaxGpuBytesToMove, m_MaxGpuAllocationsToMove, - VK_NULL_HANDLE); - - if(pBlockVectorCtx->res < VK_SUCCESS) - continue; - - pBlockVectorCtx->hasDefragmentationPlan = true; - } - - const uint32_t processed = pBlockVectorCtx->GetBlockVector()->ProcessDefragmentations( - pBlockVectorCtx, - pCurrentMove, movesLeft); - - movesLeft -= processed; - pCurrentMove += processed; - } - } - - // Process custom pools. - for(size_t customCtxIndex = 0, customCtxCount = m_CustomPoolContexts.size(); - customCtxIndex < customCtxCount; - ++customCtxIndex) - { - VmaBlockVectorDefragmentationContext *pBlockVectorCtx = m_CustomPoolContexts[customCtxIndex]; - VMA_ASSERT(pBlockVectorCtx && pBlockVectorCtx->GetBlockVector()); - - if(!pBlockVectorCtx->hasDefragmentationPlan) - { - pBlockVectorCtx->GetBlockVector()->Defragment( - pBlockVectorCtx, - m_pStats, m_Flags, - m_MaxCpuBytesToMove, m_MaxCpuAllocationsToMove, - m_MaxGpuBytesToMove, m_MaxGpuAllocationsToMove, - VK_NULL_HANDLE); - - if(pBlockVectorCtx->res < VK_SUCCESS) - continue; - - pBlockVectorCtx->hasDefragmentationPlan = true; - } - - const uint32_t processed = pBlockVectorCtx->GetBlockVector()->ProcessDefragmentations( - pBlockVectorCtx, - pCurrentMove, movesLeft); - - movesLeft -= processed; - pCurrentMove += processed; - } - - pInfo->moveCount = pInfo->moveCount - movesLeft; - - return VK_SUCCESS; -} -VkResult VmaDefragmentationContext_T::DefragmentPassEnd() -{ - VkResult res = VK_SUCCESS; - - // Process default pools. - for(uint32_t memTypeIndex = 0; - memTypeIndex < m_hAllocator->GetMemoryTypeCount(); - ++memTypeIndex) - { - VmaBlockVectorDefragmentationContext *pBlockVectorCtx = m_DefaultPoolContexts[memTypeIndex]; - if(pBlockVectorCtx) - { - VMA_ASSERT(pBlockVectorCtx->GetBlockVector()); - - if(!pBlockVectorCtx->hasDefragmentationPlan) - { - res = VK_NOT_READY; - continue; - } - - pBlockVectorCtx->GetBlockVector()->CommitDefragmentations( - pBlockVectorCtx, m_pStats); - - if(pBlockVectorCtx->defragmentationMoves.size() != pBlockVectorCtx->defragmentationMovesCommitted) - res = VK_NOT_READY; - } - } - - // Process custom pools. - for(size_t customCtxIndex = 0, customCtxCount = m_CustomPoolContexts.size(); - customCtxIndex < customCtxCount; - ++customCtxIndex) - { - VmaBlockVectorDefragmentationContext *pBlockVectorCtx = m_CustomPoolContexts[customCtxIndex]; - VMA_ASSERT(pBlockVectorCtx && pBlockVectorCtx->GetBlockVector()); - - if(!pBlockVectorCtx->hasDefragmentationPlan) - { - res = VK_NOT_READY; - continue; - } - - pBlockVectorCtx->GetBlockVector()->CommitDefragmentations( - pBlockVectorCtx, m_pStats); - - if(pBlockVectorCtx->defragmentationMoves.size() != pBlockVectorCtx->defragmentationMovesCommitted) - res = VK_NOT_READY; - } - - return res; -} - -//////////////////////////////////////////////////////////////////////////////// -// VmaRecorder - -#if VMA_RECORDING_ENABLED - -VmaRecorder::VmaRecorder() : - m_UseMutex(true), - m_Flags(0), - m_File(VMA_NULL), - m_RecordingStartTime(std::chrono::high_resolution_clock::now()) -{ -} - -VkResult VmaRecorder::Init(const VmaRecordSettings& settings, bool useMutex) -{ - m_UseMutex = useMutex; - m_Flags = settings.flags; - -#if defined(_WIN32) - // Open file for writing. - errno_t err = fopen_s(&m_File, settings.pFilePath, "wb"); - - if(err != 0) - { - return VK_ERROR_INITIALIZATION_FAILED; - } -#else - // Open file for writing. - m_File = fopen(settings.pFilePath, "wb"); - - if(m_File == 0) - { - return VK_ERROR_INITIALIZATION_FAILED; - } -#endif - - // Write header. - fprintf(m_File, "%s\n", "Vulkan Memory Allocator,Calls recording"); - fprintf(m_File, "%s\n", "1,8"); - - 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::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(); -} - -void VmaRecorder::RecordSetPoolName(uint32_t frameIndex, - VmaPool pool, - const char* name) -{ - CallParams callParams; - GetBasicParams(callParams); - - VmaMutexLock lock(m_FileMutex, m_UseMutex); - fprintf(m_File, "%u,%.3f,%u,vmaSetPoolName,%p,%s\n", callParams.threadId, callParams.time, frameIndex, - pool, name != VMA_NULL ? name : ""); - 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 { - // If VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT is not specified, convert the string's memory address to a string and store it. - snprintf(m_PtrStr, 17, "%p", pUserData); - m_Str = m_PtrStr; + switch (vectorState.operation) + { + case StateExtensive::Operation::FindFreeBlockBuffer: + vectorState.operation = StateExtensive::Operation::MoveBuffers; + break; + default: + VMA_ASSERT(0); + case StateExtensive::Operation::FindFreeBlockTexture: + vectorState.operation = StateExtensive::Operation::MoveTextures; + break; + case StateExtensive::Operation::FindFreeBlockAll: + vectorState.operation = StateExtensive::Operation::MoveAll; + break; + } + vectorState.firstFreeBlock = last; + // Nothing done, block found without reallocations, can perform another reallocs in same pass + return ComputeDefragmentation_Extensive(vector, index); } + break; + } + case StateExtensive::Operation::MoveTextures: + { + if (MoveDataToFreeBlocks(VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL, vector, + vectorState.firstFreeBlock, texturePresent, bufferPresent, otherPresent)) + { + if (texturePresent) + { + vectorState.operation = StateExtensive::Operation::FindFreeBlockTexture; + return ComputeDefragmentation_Extensive(vector, index); + } + + if (!bufferPresent && !otherPresent) + { + vectorState.operation = StateExtensive::Operation::Cleanup; + break; + } + + // No more textures to move, check buffers + vectorState.operation = StateExtensive::Operation::MoveBuffers; + bufferPresent = false; + otherPresent = false; + } + else + break; + } + case StateExtensive::Operation::MoveBuffers: + { + if (MoveDataToFreeBlocks(VMA_SUBALLOCATION_TYPE_BUFFER, vector, + vectorState.firstFreeBlock, texturePresent, bufferPresent, otherPresent)) + { + if (bufferPresent) + { + vectorState.operation = StateExtensive::Operation::FindFreeBlockBuffer; + return ComputeDefragmentation_Extensive(vector, index); + } + + if (!otherPresent) + { + vectorState.operation = StateExtensive::Operation::Cleanup; + break; + } + + // No more buffers to move, check all others + vectorState.operation = StateExtensive::Operation::MoveAll; + otherPresent = false; + } + else + break; + } + case StateExtensive::Operation::MoveAll: + { + if (MoveDataToFreeBlocks(VMA_SUBALLOCATION_TYPE_FREE, vector, + vectorState.firstFreeBlock, texturePresent, bufferPresent, otherPresent)) + { + if (otherPresent) + { + vectorState.operation = StateExtensive::Operation::FindFreeBlockBuffer; + return ComputeDefragmentation_Extensive(vector, index); + } + // Everything moved + vectorState.operation = StateExtensive::Operation::Cleanup; + } + break; + } + case StateExtensive::Operation::Cleanup: + // Cleanup is handled below so that other operations may reuse the cleanup code. This case is here to prevent the unhandled enum value warning (C4062). + break; + } + + if (vectorState.operation == StateExtensive::Operation::Cleanup) + { + // All other work done, pack data in blocks even tighter if possible + const size_t prevMoveCount = m_Moves.size(); + for (size_t i = 0; i < vector.GetBlockCount(); ++i) + { + if (ReallocWithinBlock(vector, vector.GetBlock(i))) + return true; + } + + if (prevMoveCount == m_Moves.size()) + vectorState.operation = StateExtensive::Operation::Done; + } + return false; +} + +void VmaDefragmentationContext_T::UpdateVectorStatistics(VmaBlockVector& vector, StateBalanced& state) +{ + size_t allocCount = 0; + size_t freeCount = 0; + state.avgFreeSize = 0; + state.avgAllocSize = 0; + + for (size_t i = 0; i < vector.GetBlockCount(); ++i) + { + VmaBlockMetadata* metadata = vector.GetBlock(i)->m_pMetadata; + + allocCount += metadata->GetAllocationCount(); + freeCount += metadata->GetFreeRegionsCount(); + state.avgFreeSize += metadata->GetSumFreeSize(); + state.avgAllocSize += metadata->GetSize(); + } + + state.avgAllocSize = (state.avgAllocSize - state.avgFreeSize) / allocCount; + state.avgFreeSize /= freeCount; +} + +bool VmaDefragmentationContext_T::MoveDataToFreeBlocks(VmaSuballocationType currentType, + VmaBlockVector& vector, size_t firstFreeBlock, + bool& texturePresent, bool& bufferPresent, bool& otherPresent) +{ + const size_t prevMoveCount = m_Moves.size(); + for (size_t i = firstFreeBlock ; i;) + { + VmaDeviceMemoryBlock* block = vector.GetBlock(--i); + VmaBlockMetadata* metadata = block->m_pMetadata; + + for (VmaAllocHandle handle = metadata->GetAllocationListBegin(); + handle != VK_NULL_HANDLE; + handle = metadata->GetNextAllocation(handle)) + { + MoveAllocationData moveData = GetMoveData(handle, metadata); + // Ignore newly created allocations by defragmentation algorithm + if (moveData.move.srcAllocation->GetUserData() == this) + continue; + switch (CheckCounters(moveData.move.srcAllocation->GetSize())) + { + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + default: + VMA_ASSERT(0); + case CounterStatus::Pass: + break; + } + + // Move only single type of resources at once + if (!VmaIsBufferImageGranularityConflict(moveData.type, currentType)) + { + // Try to fit allocation into free blocks + if (AllocInOtherBlock(firstFreeBlock, vector.GetBlockCount(), moveData, vector)) + return false; + } + + if (!VmaIsBufferImageGranularityConflict(moveData.type, VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL)) + texturePresent = true; + else if (!VmaIsBufferImageGranularityConflict(moveData.type, VMA_SUBALLOCATION_TYPE_BUFFER)) + bufferPresent = true; + else + otherPresent = true; + } + } + return prevMoveCount == m_Moves.size(); +} +#endif // _VMA_DEFRAGMENTATION_CONTEXT_FUNCTIONS + +#ifndef _VMA_POOL_T_FUNCTIONS +VmaPool_T::VmaPool_T( + VmaAllocator hAllocator, + const VmaPoolCreateInfo& createInfo, + VkDeviceSize preferredBlockSize) + : m_BlockVector( + hAllocator, + this, // hParentPool + 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.blockSize != 0, // explicitBlockSize + createInfo.flags & VMA_POOL_CREATE_ALGORITHM_MASK, // algorithm + createInfo.priority, + VMA_MAX(hAllocator->GetMemoryTypeMinAlignment(createInfo.memoryTypeIndex), createInfo.minAllocationAlignment), + createInfo.pMemoryAllocateNext), + m_Id(0), + m_Name(VMA_NULL) {} + +VmaPool_T::~VmaPool_T() +{ + VMA_ASSERT(m_PrevPool == VMA_NULL && m_NextPool == VMA_NULL); +} + +void VmaPool_T::SetName(const char* pName) +{ + const VkAllocationCallbacks* allocs = m_BlockVector.GetAllocator()->GetAllocationCallbacks(); + VmaFreeString(allocs, m_Name); + + if (pName != VMA_NULL) + { + m_Name = VmaCreateStringCopy(allocs, pName); } else { - m_Str = ""; + m_Name = VMA_NULL; } } +#endif // _VMA_POOL_T_FUNCTIONS -void VmaRecorder::WriteConfiguration( - const VkPhysicalDeviceProperties& devProps, - const VkPhysicalDeviceMemoryProperties& memProps, - uint32_t vulkanApiVersion, - bool dedicatedAllocationExtensionEnabled, - bool bindMemory2ExtensionEnabled, - bool memoryBudgetExtensionEnabled, - bool deviceCoherentMemoryExtensionEnabled) -{ - fprintf(m_File, "Config,Begin\n"); - - fprintf(m_File, "VulkanApiVersion,%u,%u\n", VK_VERSION_MAJOR(vulkanApiVersion), VK_VERSION_MINOR(vulkanApiVersion)); - - 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, "Extension,VK_KHR_bind_memory2,%u\n", bindMemory2ExtensionEnabled ? 1 : 0); - fprintf(m_File, "Extension,VK_EXT_memory_budget,%u\n", memoryBudgetExtensionEnabled ? 1 : 0); - fprintf(m_File, "Extension,VK_AMD_device_coherent_memory,%u\n", deviceCoherentMemoryExtensionEnabled ? 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_MIN_ALIGNMENT,%llu\n", (VkDeviceSize)VMA_MIN_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) -{ - #if defined(_WIN32) - outParams.threadId = GetCurrentThreadId(); - #else - // Use C++11 features to get thread id and convert it to uint32_t. - // There is room for optimization since sstream is quite slow. - // Is there a better way to convert std::this_thread::get_id() to uint32_t? - std::thread::id thread_id = std::this_thread::get_id(); - std::stringstream thread_id_to_string_converter; - thread_id_to_string_converter << thread_id; - std::string thread_id_as_string = thread_id_to_string_converter.str(); - outParams.threadId = static_cast(std::stoi(thread_id_as_string.c_str())); - #endif - - auto current_time = std::chrono::high_resolution_clock::now(); - - outParams.time = std::chrono::duration(current_time - m_RecordingStartTime).count(); -} - -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 - -//////////////////////////////////////////////////////////////////////////////// -// VmaAllocationObjectAllocator - -VmaAllocationObjectAllocator::VmaAllocationObjectAllocator(const VkAllocationCallbacks* pAllocationCallbacks) : - m_Allocator(pAllocationCallbacks, 1024) -{ -} - -template VmaAllocation VmaAllocationObjectAllocator::Allocate(Types&&... args) -{ - VmaMutexLock mutexLock(m_Mutex); - return m_Allocator.Alloc(std::forward(args)...); -} - -void VmaAllocationObjectAllocator::Free(VmaAllocation hAlloc) -{ - VmaMutexLock mutexLock(m_Mutex); - m_Allocator.Free(hAlloc); -} - -//////////////////////////////////////////////////////////////////////////////// -// VmaAllocator_T - +#ifndef _VMA_ALLOCATOR_T_FUNCTIONS VmaAllocator_T::VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo) : m_UseMutex((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT) == 0), m_VulkanApiVersion(pCreateInfo->vulkanApiVersion != 0 ? pCreateInfo->vulkanApiVersion : VK_API_VERSION_1_0), @@ -14194,13 +13984,9 @@ VmaAllocator_T::VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo) : m_DeviceMemoryCount(0), m_PreferredLargeHeapBlockSize(0), m_PhysicalDevice(pCreateInfo->physicalDevice), - m_CurrentFrameIndex(0), m_GpuDefragmentationMemoryTypeBits(UINT32_MAX), m_NextPoolId(0), m_GlobalMemoryTypeBits(UINT32_MAX) -#if VMA_RECORDING_ENABLED - ,m_pRecorder(VMA_NULL) -#endif { if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) { @@ -14321,24 +14107,26 @@ VmaAllocator_T::VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo) : for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) { - const VkDeviceSize preferredBlockSize = CalcPreferredBlockSize(memTypeIndex); - - m_pBlockVectors[memTypeIndex] = vma_new(this, VmaBlockVector)( - this, - VK_NULL_HANDLE, // hParentPool - memTypeIndex, - preferredBlockSize, - 0, - SIZE_MAX, - GetBufferImageGranularity(), - pCreateInfo->frameInUseCount, - false, // explicitBlockSize - false, // linearAlgorithm - 0.5f, // priority (0.5 is the default per Vulkan spec) - GetMemoryTypeMinAlignment(memTypeIndex), // minAllocationAlignment - VMA_NULL); // // pMemoryAllocateNext - // No need to call m_pBlockVectors[memTypeIndex][blockVectorTypeIndex]->CreateMinBlocks here, - // becase minBlockCount is 0. + // Create only supported types + if((m_GlobalMemoryTypeBits & (1u << memTypeIndex)) != 0) + { + const VkDeviceSize preferredBlockSize = CalcPreferredBlockSize(memTypeIndex); + m_pBlockVectors[memTypeIndex] = vma_new(this, VmaBlockVector)( + this, + VK_NULL_HANDLE, // hParentPool + memTypeIndex, + preferredBlockSize, + 0, + SIZE_MAX, + GetBufferImageGranularity(), + false, // explicitBlockSize + 0, // algorithm + 0.5f, // priority (0.5 is the default per Vulkan spec) + GetMemoryTypeMinAlignment(memTypeIndex), // minAllocationAlignment + VMA_NULL); // // pMemoryAllocateNext + // No need to call m_pBlockVectors[memTypeIndex][blockVectorTypeIndex]->CreateMinBlocks here, + // becase minBlockCount is 0. + } } } @@ -14346,31 +14134,6 @@ 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_VulkanApiVersion, - m_UseKhrDedicatedAllocation, - m_UseKhrBindMemory2, - m_UseExtMemoryBudget, - m_UseAmdDeviceCoherentMemory); - 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 - } - #if VMA_MEMORY_BUDGET if(m_UseExtMemoryBudget) { @@ -14383,23 +14146,10 @@ VkResult VmaAllocator_T::Init(const VmaAllocatorCreateInfo* pCreateInfo) 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.IsEmpty()); for(size_t memTypeIndex = GetMemoryTypeCount(); memTypeIndex--; ) { - if(!m_DedicatedAllocations[memTypeIndex].IsEmpty()) - { - VMA_ASSERT(0 && "Unfreed dedicated allocations found."); - } - vma_delete(this, m_pBlockVectors[memTypeIndex]); } } @@ -14427,6 +14177,8 @@ void VmaAllocator_T::ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunc void VmaAllocator_T::ImportVulkanFunctions_Static() { // Vulkan 1.0 + m_VulkanFunctions.vkGetInstanceProcAddr = (PFN_vkGetInstanceProcAddr)vkGetInstanceProcAddr; + m_VulkanFunctions.vkGetDeviceProcAddr = (PFN_vkGetDeviceProcAddr)vkGetDeviceProcAddr; m_VulkanFunctions.vkGetPhysicalDeviceProperties = (PFN_vkGetPhysicalDeviceProperties)vkGetPhysicalDeviceProperties; m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties = (PFN_vkGetPhysicalDeviceMemoryProperties)vkGetPhysicalDeviceMemoryProperties; m_VulkanFunctions.vkAllocateMemory = (PFN_vkAllocateMemory)vkAllocateMemory; @@ -14456,9 +14208,17 @@ void VmaAllocator_T::ImportVulkanFunctions_Static() m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties2KHR = (PFN_vkGetPhysicalDeviceMemoryProperties2)vkGetPhysicalDeviceMemoryProperties2; } #endif + +#if VMA_VULKAN_VERSION >= 1003000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 3, 0)) + { + m_VulkanFunctions.vkGetDeviceBufferMemoryRequirements = (PFN_vkGetDeviceBufferMemoryRequirements)vkGetDeviceBufferMemoryRequirements; + m_VulkanFunctions.vkGetDeviceImageMemoryRequirements = (PFN_vkGetDeviceImageMemoryRequirements)vkGetDeviceImageMemoryRequirements; + } +#endif } -#endif // #if VMA_STATIC_VULKAN_FUNCTIONS == 1 +#endif // VMA_STATIC_VULKAN_FUNCTIONS == 1 void VmaAllocator_T::ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVulkanFunctions) { @@ -14467,6 +14227,8 @@ void VmaAllocator_T::ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVul #define VMA_COPY_IF_NOT_NULL(funcName) \ if(pVulkanFunctions->funcName != VMA_NULL) m_VulkanFunctions.funcName = pVulkanFunctions->funcName; + VMA_COPY_IF_NOT_NULL(vkGetInstanceProcAddr); + VMA_COPY_IF_NOT_NULL(vkGetDeviceProcAddr); VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceProperties); VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceMemoryProperties); VMA_COPY_IF_NOT_NULL(vkAllocateMemory); @@ -14499,6 +14261,11 @@ void VmaAllocator_T::ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVul VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceMemoryProperties2KHR); #endif +#if VMA_VULKAN_VERSION >= 1003000 + VMA_COPY_IF_NOT_NULL(vkGetDeviceBufferMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkGetDeviceImageMemoryRequirements); +#endif + #undef VMA_COPY_IF_NOT_NULL } @@ -14506,14 +14273,19 @@ void VmaAllocator_T::ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVul void VmaAllocator_T::ImportVulkanFunctions_Dynamic() { + VMA_ASSERT(m_VulkanFunctions.vkGetInstanceProcAddr && m_VulkanFunctions.vkGetDeviceProcAddr && + "To use VMA_DYNAMIC_VULKAN_FUNCTIONS in new versions of VMA you now have to pass " + "VmaVulkanFunctions::vkGetInstanceProcAddr and vkGetDeviceProcAddr as VmaAllocatorCreateInfo::pVulkanFunctions. " + "Other members can be null."); + #define VMA_FETCH_INSTANCE_FUNC(memberName, functionPointerType, functionNameString) \ if(m_VulkanFunctions.memberName == VMA_NULL) \ m_VulkanFunctions.memberName = \ - (functionPointerType)vkGetInstanceProcAddr(m_hInstance, functionNameString); + (functionPointerType)m_VulkanFunctions.vkGetInstanceProcAddr(m_hInstance, functionNameString); #define VMA_FETCH_DEVICE_FUNC(memberName, functionPointerType, functionNameString) \ if(m_VulkanFunctions.memberName == VMA_NULL) \ m_VulkanFunctions.memberName = \ - (functionPointerType)vkGetDeviceProcAddr(m_hDevice, functionNameString); + (functionPointerType)m_VulkanFunctions.vkGetDeviceProcAddr(m_hDevice, functionNameString); VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceProperties, PFN_vkGetPhysicalDeviceProperties, "vkGetPhysicalDeviceProperties"); VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties, PFN_vkGetPhysicalDeviceMemoryProperties, "vkGetPhysicalDeviceMemoryProperties"); @@ -14567,11 +14339,19 @@ void VmaAllocator_T::ImportVulkanFunctions_Dynamic() } #endif // #if VMA_MEMORY_BUDGET +#if VMA_VULKAN_VERSION >= 1003000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 3, 0)) + { + VMA_FETCH_DEVICE_FUNC(vkGetDeviceBufferMemoryRequirements, PFN_vkGetDeviceBufferMemoryRequirements, "vkGetDeviceBufferMemoryRequirements"); + VMA_FETCH_DEVICE_FUNC(vkGetDeviceImageMemoryRequirements, PFN_vkGetDeviceImageMemoryRequirements, "vkGetDeviceImageMemoryRequirements"); + } +#endif + #undef VMA_FETCH_DEVICE_FUNC #undef VMA_FETCH_INSTANCE_FUNC } -#endif // #if VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 +#endif // VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 void VmaAllocator_T::ValidateVulkanFunctions() { @@ -14615,6 +14395,14 @@ void VmaAllocator_T::ValidateVulkanFunctions() VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties2KHR != VMA_NULL); } #endif + +#if VMA_VULKAN_VERSION >= 1003000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 3, 0)) + { + VMA_ASSERT(m_VulkanFunctions.vkGetDeviceBufferMemoryRequirements != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetDeviceImageMemoryRequirements != VMA_NULL); + } +#endif } VkDeviceSize VmaAllocator_T::CalcPreferredBlockSize(uint32_t memTypeIndex) @@ -14626,15 +14414,18 @@ VkDeviceSize VmaAllocator_T::CalcPreferredBlockSize(uint32_t memTypeIndex) } VkResult VmaAllocator_T::AllocateMemoryOfType( + VmaPool pool, VkDeviceSize size, VkDeviceSize alignment, - bool dedicatedAllocation, + bool dedicatedPreferred, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, const VmaAllocationCreateInfo& createInfo, uint32_t memTypeIndex, VmaSuballocationType suballocType, + VmaDedicatedAllocationList& dedicatedAllocations, + VmaBlockVector& blockVector, size_t allocationCount, VmaAllocation* pAllocations) { @@ -14642,64 +14433,88 @@ VkResult VmaAllocator_T::AllocateMemoryOfType( VMA_DEBUG_LOG(" AllocateMemory: MemoryTypeIndex=%u, AllocationCount=%zu, Size=%llu", memTypeIndex, allocationCount, 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; - } - // If memory is lazily allocated, it should be always dedicated. - if(finalCreateInfo.usage == VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED) - { - finalCreateInfo.flags |= VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_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; - } + VkResult res = CalcMemTypeParams( + finalCreateInfo, + memTypeIndex, + size, + allocationCount); + if(res != VK_SUCCESS) + return res; 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_WITHIN_BUDGET_BIT) != 0, - (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, - (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, - finalCreateInfo.pUserData, - finalCreateInfo.priority, - dedicatedBuffer, - dedicatedBufferUsage, - dedicatedImage, - allocationCount, - pAllocations); - } + return AllocateDedicatedMemory( + pool, + size, + suballocType, + dedicatedAllocations, + memTypeIndex, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, + (finalCreateInfo.flags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_CAN_ALIAS_BIT) != 0, + finalCreateInfo.pUserData, + finalCreateInfo.priority, + dedicatedBuffer, + dedicatedImage, + dedicatedBufferImageUsage, + allocationCount, + pAllocations, + blockVector.GetAllocationNextPtr()); } else { - VkResult res = blockVector->Allocate( - m_CurrentFrameIndex.load(), + const bool canAllocateDedicated = + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) == 0 && + (pool == VK_NULL_HANDLE || !blockVector.HasExplicitBlockSize()); + + if(canAllocateDedicated) + { + // Heuristics: Allocate dedicated memory if requested size if greater than half of preferred block size. + if(size > blockVector.GetPreferredBlockSize() / 2) + { + dedicatedPreferred = true; + } + // Protection against creating each allocation as dedicated when we reach or exceed heap size/budget, + // which can quickly deplete maxMemoryAllocationCount: Don't prefer dedicated allocations when above + // 3/4 of the maximum allocation count. + if(m_DeviceMemoryCount.load() > m_PhysicalDeviceProperties.limits.maxMemoryAllocationCount * 3 / 4) + { + dedicatedPreferred = false; + } + + if(dedicatedPreferred) + { + res = AllocateDedicatedMemory( + pool, + size, + suballocType, + dedicatedAllocations, + memTypeIndex, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, + (finalCreateInfo.flags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_CAN_ALIAS_BIT) != 0, + finalCreateInfo.pUserData, + finalCreateInfo.priority, + dedicatedBuffer, + dedicatedImage, + dedicatedBufferImageUsage, + allocationCount, + pAllocations, + blockVector.GetAllocationNextPtr()); + if(res == VK_SUCCESS) + { + // Succeeded: AllocateDedicatedMemory function already filld pMemory, nothing more to do here. + VMA_DEBUG_LOG(" Allocated as DedicatedMemory"); + return VK_SUCCESS; + } + } + } + + res = blockVector.Allocate( size, alignment, finalCreateInfo, @@ -14707,99 +14522,86 @@ VkResult VmaAllocator_T::AllocateMemoryOfType( 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; - } - - // Protection against creating each allocation as dedicated when we reach or exceed heap size/budget, - // which can quickly deplete maxMemoryAllocationCount: Don't try dedicated allocations when above - // 3/4 of the maximum allocation count. - if(m_DeviceMemoryCount.load() > m_PhysicalDeviceProperties.limits.maxMemoryAllocationCount * 3 / 4) - { - return VK_ERROR_OUT_OF_DEVICE_MEMORY; - } - - res = AllocateDedicatedMemory( - size, - suballocType, - memTypeIndex, - (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT) != 0, - (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, - (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, - finalCreateInfo.pUserData, - finalCreateInfo.priority, - dedicatedBuffer, - dedicatedBufferUsage, - 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 + + // Try dedicated memory. + if(canAllocateDedicated && !dedicatedPreferred) { - // Everything failed: Return error code. - VMA_DEBUG_LOG(" vkAllocateMemory FAILED"); - return res; + res = AllocateDedicatedMemory( + pool, + size, + suballocType, + dedicatedAllocations, + memTypeIndex, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT) != 0, + (finalCreateInfo.flags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0, + (finalCreateInfo.flags & VMA_ALLOCATION_CREATE_CAN_ALIAS_BIT) != 0, + finalCreateInfo.pUserData, + finalCreateInfo.priority, + dedicatedBuffer, + dedicatedImage, + dedicatedBufferImageUsage, + allocationCount, + pAllocations, + blockVector.GetAllocationNextPtr()); + if(res == VK_SUCCESS) + { + // Succeeded: AllocateDedicatedMemory function already filld pMemory, nothing more to do here. + VMA_DEBUG_LOG(" Allocated as DedicatedMemory"); + return VK_SUCCESS; + } } + // Everything failed: Return error code. + VMA_DEBUG_LOG(" vkAllocateMemory FAILED"); + return res; } } VkResult VmaAllocator_T::AllocateDedicatedMemory( + VmaPool pool, VkDeviceSize size, VmaSuballocationType suballocType, + VmaDedicatedAllocationList& dedicatedAllocations, uint32_t memTypeIndex, - bool withinBudget, bool map, bool isUserDataString, + bool isMappingAllowed, + bool canAliasMemory, void* pUserData, float priority, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, size_t allocationCount, - VmaAllocation* pAllocations) + VmaAllocation* pAllocations, + const void* pNextChain) { VMA_ASSERT(allocationCount > 0 && pAllocations); - if(withinBudget) - { - const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); - VmaBudget heapBudget = {}; - GetBudget(&heapBudget, heapIndex, 1); - if(heapBudget.usage + size * allocationCount > heapBudget.budget) - { - return VK_ERROR_OUT_OF_DEVICE_MEMORY; - } - } - VkMemoryAllocateInfo allocInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO }; allocInfo.memoryTypeIndex = memTypeIndex; allocInfo.allocationSize = size; + allocInfo.pNext = pNextChain; #if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 VkMemoryDedicatedAllocateInfoKHR dedicatedAllocInfo = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO_KHR }; - if(m_UseKhrDedicatedAllocation || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + if(!canAliasMemory) { - if(dedicatedBuffer != VK_NULL_HANDLE) + if(m_UseKhrDedicatedAllocation || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) { - VMA_ASSERT(dedicatedImage == VK_NULL_HANDLE); - dedicatedAllocInfo.buffer = dedicatedBuffer; - VmaPnextChainPushFront(&allocInfo, &dedicatedAllocInfo); - } - else if(dedicatedImage != VK_NULL_HANDLE) - { - dedicatedAllocInfo.image = dedicatedImage; - VmaPnextChainPushFront(&allocInfo, &dedicatedAllocInfo); + if(dedicatedBuffer != VK_NULL_HANDLE) + { + VMA_ASSERT(dedicatedImage == VK_NULL_HANDLE); + dedicatedAllocInfo.buffer = dedicatedBuffer; + VmaPnextChainPushFront(&allocInfo, &dedicatedAllocInfo); + } + else if(dedicatedImage != VK_NULL_HANDLE) + { + dedicatedAllocInfo.image = dedicatedImage; + VmaPnextChainPushFront(&allocInfo, &dedicatedAllocInfo); + } } } #endif // #if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 @@ -14811,8 +14613,8 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( bool canContainBufferWithDeviceAddress = true; if(dedicatedBuffer != VK_NULL_HANDLE) { - canContainBufferWithDeviceAddress = dedicatedBufferUsage == UINT32_MAX || // Usage flags unknown - (dedicatedBufferUsage & VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_EXT) != 0; + canContainBufferWithDeviceAddress = dedicatedBufferImageUsage == UINT32_MAX || // Usage flags unknown + (dedicatedBufferImageUsage & VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_EXT) != 0; } else if(dedicatedImage != VK_NULL_HANDLE) { @@ -14830,6 +14632,7 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( VkMemoryPriorityAllocateInfoEXT priorityInfo = { VK_STRUCTURE_TYPE_MEMORY_PRIORITY_ALLOCATE_INFO_EXT }; if(m_UseExtMemoryPriority) { + VMA_ASSERT(priority >= 0.f && priority <= 1.f); priorityInfo.priority = priority; VmaPnextChainPushFront(&allocInfo, &priorityInfo); } @@ -14850,12 +14653,14 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) { res = AllocateDedicatedMemoryPage( + pool, size, suballocType, memTypeIndex, allocInfo, map, isUserDataString, + isMappingAllowed, pUserData, pAllocations + allocIndex); if(res != VK_SUCCESS) @@ -14866,16 +14671,10 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( if(res == VK_SUCCESS) { - // Register them in m_DedicatedAllocations. + for (allocIndex = 0; allocIndex < allocationCount; ++allocIndex) { - VmaMutexLockWrite lock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); - DedicatedAllocationLinkedList& dedicatedAllocations = m_DedicatedAllocations[memTypeIndex]; - for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) - { - dedicatedAllocations.PushBack(pAllocations[allocIndex]); - } + dedicatedAllocations.Register(pAllocations[allocIndex]); } - VMA_DEBUG_LOG(" Allocated DedicatedMemory Count=%zu, MemoryTypeIndex=#%u", allocationCount, memTypeIndex); } else @@ -14898,7 +14697,6 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( FreeVulkanMemory(memTypeIndex, currAlloc->GetSize(), hMemory); m_Budget.RemoveAllocation(MemoryTypeIndexToHeapIndex(memTypeIndex), currAlloc->GetSize()); - currAlloc->SetUserData(this, VMA_NULL); m_AllocationObjectAllocator.Free(currAlloc); } @@ -14909,12 +14707,14 @@ VkResult VmaAllocator_T::AllocateDedicatedMemory( } VkResult VmaAllocator_T::AllocateDedicatedMemoryPage( + VmaPool pool, VkDeviceSize size, VmaSuballocationType suballocType, uint32_t memTypeIndex, const VkMemoryAllocateInfo& allocInfo, bool map, bool isUserDataString, + bool isMappingAllowed, void* pUserData, VmaAllocation* pAllocation) { @@ -14944,9 +14744,12 @@ VkResult VmaAllocator_T::AllocateDedicatedMemoryPage( } } - *pAllocation = m_AllocationObjectAllocator.Allocate(m_CurrentFrameIndex.load(), isUserDataString); - (*pAllocation)->InitDedicatedAllocation(memTypeIndex, hMemory, suballocType, pMappedData, size); - (*pAllocation)->SetUserData(this, pUserData); + *pAllocation = m_AllocationObjectAllocator.Allocate(isMappingAllowed); + (*pAllocation)->InitDedicatedAllocation(pool, memTypeIndex, hMemory, suballocType, pMappedData, size); + if (isUserDataString) + (*pAllocation)->SetName(this, (const char*)pUserData); + else + (*pAllocation)->SetUserData(this, pUserData); m_Budget.AddAllocation(MemoryTypeIndexToHeapIndex(memTypeIndex), size); if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) { @@ -15020,13 +14823,165 @@ void VmaAllocator_T::GetImageMemoryRequirements( } } +VkResult VmaAllocator_T::FindMemoryTypeIndex( + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkFlags bufImgUsage, + uint32_t* pMemoryTypeIndex) const +{ + memoryTypeBits &= GetGlobalMemoryTypeBits(); + + if(pAllocationCreateInfo->memoryTypeBits != 0) + { + memoryTypeBits &= pAllocationCreateInfo->memoryTypeBits; + } + + VkMemoryPropertyFlags requiredFlags = 0, preferredFlags = 0, notPreferredFlags = 0; + if(!FindMemoryPreferences( + IsIntegratedGpu(), + *pAllocationCreateInfo, + bufImgUsage, + requiredFlags, preferredFlags, notPreferredFlags)) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + *pMemoryTypeIndex = UINT32_MAX; + uint32_t minCost = UINT32_MAX; + for(uint32_t memTypeIndex = 0, memTypeBit = 1; + memTypeIndex < GetMemoryTypeCount(); + ++memTypeIndex, memTypeBit <<= 1) + { + // This memory type is acceptable according to memoryTypeBits bitmask. + if((memTypeBit & memoryTypeBits) != 0) + { + const VkMemoryPropertyFlags currFlags = + 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 = VMA_COUNT_BITS_SET(preferredFlags & ~currFlags) + + VMA_COUNT_BITS_SET(currFlags & notPreferredFlags); + // 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 VmaAllocator_T::CalcMemTypeParams( + VmaAllocationCreateInfo& inoutCreateInfo, + uint32_t memTypeIndex, + VkDeviceSize size, + size_t allocationCount) +{ + // If memory type is not HOST_VISIBLE, disable MAPPED. + if((inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0 && + (m_MemProps.memoryTypes[memTypeIndex].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) + { + inoutCreateInfo.flags &= ~VMA_ALLOCATION_CREATE_MAPPED_BIT; + } + + if((inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0 && + (inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT) != 0) + { + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); + VmaBudget heapBudget = {}; + GetHeapBudgets(&heapBudget, heapIndex, 1); + if(heapBudget.usage + size * allocationCount > heapBudget.budget) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + } + return VK_SUCCESS; +} + +VkResult VmaAllocator_T::CalcAllocationParams( + VmaAllocationCreateInfo& inoutCreateInfo, + bool dedicatedRequired, + bool dedicatedPreferred) +{ + VMA_ASSERT((inoutCreateInfo.flags & + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != + (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT) && + "Specifying both flags VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT and VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT is incorrect."); + VMA_ASSERT((((inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT) == 0 || + (inoutCreateInfo.flags & (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0)) && + "Specifying VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT requires also VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT."); + if(inoutCreateInfo.usage == VMA_MEMORY_USAGE_AUTO || inoutCreateInfo.usage == VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE || inoutCreateInfo.usage == VMA_MEMORY_USAGE_AUTO_PREFER_HOST) + { + if((inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0) + { + VMA_ASSERT((inoutCreateInfo.flags & (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) != 0 && + "When using VMA_ALLOCATION_CREATE_MAPPED_BIT and usage = VMA_MEMORY_USAGE_AUTO*, you must also specify VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT."); + } + } + + // If memory is lazily allocated, it should be always dedicated. + if(dedicatedRequired || + inoutCreateInfo.usage == VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED) + { + inoutCreateInfo.flags |= VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT; + } + + if(inoutCreateInfo.pool != VK_NULL_HANDLE) + { + if(inoutCreateInfo.pool->m_BlockVector.HasExplicitBlockSize() && + (inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0) + { + VMA_ASSERT(0 && "Specifying VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT while current custom pool doesn't support dedicated allocations."); + return VK_ERROR_FEATURE_NOT_PRESENT; + } + inoutCreateInfo.priority = inoutCreateInfo.pool->m_BlockVector.GetPriority(); + } + + if((inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0 && + (inoutCreateInfo.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_FEATURE_NOT_PRESENT; + } + + if(VMA_DEBUG_ALWAYS_DEDICATED_MEMORY && + (inoutCreateInfo.flags & VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT) != 0) + { + inoutCreateInfo.flags |= VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT; + } + + // Non-auto USAGE values imply HOST_ACCESS flags. + // And so does VMA_MEMORY_USAGE_UNKNOWN because it is used with custom pools. + // Which specific flag is used doesn't matter. They change things only when used with VMA_MEMORY_USAGE_AUTO*. + // Otherwise they just protect from assert on mapping. + if(inoutCreateInfo.usage != VMA_MEMORY_USAGE_AUTO && + inoutCreateInfo.usage != VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE && + inoutCreateInfo.usage != VMA_MEMORY_USAGE_AUTO_PREFER_HOST) + { + if((inoutCreateInfo.flags & (VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT)) == 0) + { + inoutCreateInfo.flags |= VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT; + } + } + + return VK_SUCCESS; +} + VkResult VmaAllocator_T::AllocateMemory( const VkMemoryRequirements& vkMemReq, bool requiresDedicatedAllocation, bool prefersDedicatedAllocation, VkBuffer dedicatedBuffer, - VkBufferUsageFlags dedicatedBufferUsage, VkImage dedicatedImage, + VkFlags dedicatedBufferImageUsage, const VmaAllocationCreateInfo& createInfo, VmaSuballocationType suballocType, size_t allocationCount, @@ -15040,54 +14995,28 @@ VkResult VmaAllocator_T::AllocateMemory( { return VK_ERROR_INITIALIZATION_FAILED; } - 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) - { - VmaAllocationCreateInfo createInfoForPool = createInfo; - // If memory type is not HOST_VISIBLE, disable MAPPED. - if((createInfoForPool.flags & VMA_ALLOCATION_CREATE_MAPPED_BIT) != 0 && - (m_MemProps.memoryTypes[createInfo.pool->m_BlockVector.GetMemoryTypeIndex()].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) == 0) - { - createInfoForPool.flags &= ~VMA_ALLOCATION_CREATE_MAPPED_BIT; - } + VmaAllocationCreateInfo createInfoFinal = createInfo; + VkResult res = CalcAllocationParams(createInfoFinal, requiresDedicatedAllocation, prefersDedicatedAllocation); + if(res != VK_SUCCESS) + return res; - return createInfo.pool->m_BlockVector.Allocate( - m_CurrentFrameIndex.load(), + if(createInfoFinal.pool != VK_NULL_HANDLE) + { + VmaBlockVector& blockVector = createInfoFinal.pool->m_BlockVector; + return AllocateMemoryOfType( + createInfoFinal.pool, vkMemReq.size, vkMemReq.alignment, - createInfoForPool, + prefersDedicatedAllocation, + dedicatedBuffer, + dedicatedImage, + dedicatedBufferImageUsage, + createInfoFinal, + blockVector.GetMemoryTypeIndex(), suballocType, + createInfoFinal.pool->m_DedicatedAllocations, + blockVector, allocationCount, pAllocations); } @@ -15096,68 +15025,42 @@ VkResult VmaAllocator_T::AllocateMemory( // 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) + res = FindMemoryTypeIndex(memoryTypeBits, &createInfoFinal, dedicatedBufferImageUsage, &memTypeIndex); + // Can't find any single memory type matching requirements. res is VK_ERROR_FEATURE_NOT_PRESENT. + if(res != VK_SUCCESS) + return res; + do { + VmaBlockVector* blockVector = m_pBlockVectors[memTypeIndex]; + VMA_ASSERT(blockVector && "Trying to use unsupported memory type!"); res = AllocateMemoryOfType( + VK_NULL_HANDLE, vkMemReq.size, vkMemReq.alignment, requiresDedicatedAllocation || prefersDedicatedAllocation, dedicatedBuffer, - dedicatedBufferUsage, dedicatedImage, - createInfo, + dedicatedBufferImageUsage, + createInfoFinal, memTypeIndex, suballocType, + m_DedicatedAllocations[memTypeIndex], + *blockVector, allocationCount, pAllocations); - // Succeeded on first try. + // Allocation succeeded 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) - { - res = AllocateMemoryOfType( - vkMemReq.size, - vkMemReq.alignment, - requiresDedicatedAllocation || prefersDedicatedAllocation, - dedicatedBuffer, - dedicatedBufferUsage, - 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; + return VK_SUCCESS; + + // Remove old memTypeIndex from list of possibilities. + memoryTypeBits &= ~(1u << memTypeIndex); + // Find alternative memTypeIndex. + res = FindMemoryTypeIndex(memoryTypeBits, &createInfoFinal, dedicatedBufferImageUsage, &memTypeIndex); + } while(res == VK_SUCCESS); + + // No other matching memory type index could be found. + // Not returning res, which is VK_ERROR_FEATURE_NOT_PRESENT, because we already failed to allocate once. + return VK_ERROR_OUT_OF_DEVICE_MEMORY; } } @@ -15173,62 +15076,57 @@ void VmaAllocator_T::FreeMemory( if(allocation != VK_NULL_HANDLE) { - if(TouchAllocation(allocation)) + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS) { - 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->GetBlock()->GetParentPool(); - 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); - } + FillAllocation(allocation, VMA_ALLOCATION_FILL_PATTERN_DESTROYED); } - // Do this regardless of whether the allocation is lost. Lost allocations still account to Budget.AllocationBytes. - m_Budget.RemoveAllocation(MemoryTypeIndexToHeapIndex(allocation->GetMemoryTypeIndex()), allocation->GetSize()); - allocation->SetUserData(this, VMA_NULL); - m_AllocationObjectAllocator.Free(allocation); + allocation->FreeName(this); + + switch(allocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaBlockVector* pBlockVector = VMA_NULL; + VmaPool hPool = allocation->GetParentPool(); + if(hPool != VK_NULL_HANDLE) + { + pBlockVector = &hPool->m_BlockVector; + } + else + { + const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + pBlockVector = m_pBlockVectors[memTypeIndex]; + VMA_ASSERT(pBlockVector && "Trying to free memory of unsupported type!"); + } + pBlockVector->Free(allocation); + } + break; + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + FreeDedicatedMemory(allocation); + break; + default: + VMA_ASSERT(0); + } } } } -void VmaAllocator_T::CalculateStats(VmaStats* pStats) +void VmaAllocator_T::CalculateStatistics(VmaTotalStatistics* 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]); + VmaClearDetailedStatistics(pStats->total); + for(uint32_t i = 0; i < VK_MAX_MEMORY_TYPES; ++i) + VmaClearDetailedStatistics(pStats->memoryType[i]); + for(uint32_t i = 0; i < VK_MAX_MEMORY_HEAPS; ++i) + VmaClearDetailedStatistics(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); + if (pBlockVector != VMA_NULL) + pBlockVector->AddDetailedStatistics(pStats->memoryType[memTypeIndex]); } // Process custom pools. @@ -15236,36 +15134,37 @@ void VmaAllocator_T::CalculateStats(VmaStats* pStats) VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); for(VmaPool pool = m_Pools.Front(); pool != VMA_NULL; pool = m_Pools.GetNext(pool)) { - pool->m_BlockVector.AddStats(pStats); + VmaBlockVector& blockVector = pool->m_BlockVector; + const uint32_t memTypeIndex = blockVector.GetMemoryTypeIndex(); + blockVector.AddDetailedStatistics(pStats->memoryType[memTypeIndex]); + pool->m_DedicatedAllocations.AddDetailedStatistics(pStats->memoryType[memTypeIndex]); } } // Process dedicated allocations. for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) { - const uint32_t memHeapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); - VmaMutexLockRead dedicatedAllocationsLock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); - DedicatedAllocationLinkedList& dedicatedAllocList = m_DedicatedAllocations[memTypeIndex]; - for(VmaAllocation alloc = dedicatedAllocList.Front(); - alloc != VMA_NULL; alloc = dedicatedAllocList.GetNext(alloc)) - { - VmaStatInfo allocationStatInfo; - alloc->DedicatedAllocCalcStatsInfo(allocationStatInfo); - VmaAddStatInfo(pStats->total, allocationStatInfo); - VmaAddStatInfo(pStats->memoryType[memTypeIndex], allocationStatInfo); - VmaAddStatInfo(pStats->memoryHeap[memHeapIndex], allocationStatInfo); - } + m_DedicatedAllocations[memTypeIndex].AddDetailedStatistics(pStats->memoryType[memTypeIndex]); } - // 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]); + // Sum from memory types to memory heaps. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + const uint32_t memHeapIndex = m_MemProps.memoryTypes[memTypeIndex].heapIndex; + VmaAddDetailedStatistics(pStats->memoryHeap[memHeapIndex], pStats->memoryType[memTypeIndex]); + } + + // Sum from memory heaps to total. + for(uint32_t memHeapIndex = 0; memHeapIndex < GetMemoryHeapCount(); ++memHeapIndex) + VmaAddDetailedStatistics(pStats->total, pStats->memoryHeap[memHeapIndex]); + + VMA_ASSERT(pStats->total.statistics.allocationCount == 0 || + pStats->total.allocationSizeMax >= pStats->total.allocationSizeMin); + VMA_ASSERT(pStats->total.unusedRangeCount == 0 || + pStats->total.unusedRangeSizeMax >= pStats->total.unusedRangeSizeMin); } -void VmaAllocator_T::GetBudget(VmaBudget* outBudget, uint32_t firstHeap, uint32_t heapCount) +void VmaAllocator_T::GetHeapBudgets(VmaBudget* outBudgets, uint32_t firstHeap, uint32_t heapCount) { #if VMA_MEMORY_BUDGET if(m_UseExtMemoryBudget) @@ -15273,225 +15172,63 @@ void VmaAllocator_T::GetBudget(VmaBudget* outBudget, uint32_t firstHeap, uint32_ if(m_Budget.m_OperationsSinceBudgetFetch < 30) { VmaMutexLockRead lockRead(m_Budget.m_BudgetMutex, m_UseMutex); - for(uint32_t i = 0; i < heapCount; ++i, ++outBudget) + for(uint32_t i = 0; i < heapCount; ++i, ++outBudgets) { const uint32_t heapIndex = firstHeap + i; - outBudget->blockBytes = m_Budget.m_BlockBytes[heapIndex]; - outBudget->allocationBytes = m_Budget.m_AllocationBytes[heapIndex]; + outBudgets->statistics.blockCount = m_Budget.m_BlockCount[heapIndex]; + outBudgets->statistics.allocationCount = m_Budget.m_AllocationCount[heapIndex]; + outBudgets->statistics.blockBytes = m_Budget.m_BlockBytes[heapIndex]; + outBudgets->statistics.allocationBytes = m_Budget.m_AllocationBytes[heapIndex]; - if(m_Budget.m_VulkanUsage[heapIndex] + outBudget->blockBytes > m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]) + if(m_Budget.m_VulkanUsage[heapIndex] + outBudgets->statistics.blockBytes > m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]) { - outBudget->usage = m_Budget.m_VulkanUsage[heapIndex] + - outBudget->blockBytes - m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]; + outBudgets->usage = m_Budget.m_VulkanUsage[heapIndex] + + outBudgets->statistics.blockBytes - m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]; } else { - outBudget->usage = 0; + outBudgets->usage = 0; } // Have to take MIN with heap size because explicit HeapSizeLimit is included in it. - outBudget->budget = VMA_MIN( + outBudgets->budget = VMA_MIN( m_Budget.m_VulkanBudget[heapIndex], m_MemProps.memoryHeaps[heapIndex].size); } } else { UpdateVulkanBudget(); // Outside of mutex lock - GetBudget(outBudget, firstHeap, heapCount); // Recursion + GetHeapBudgets(outBudgets, firstHeap, heapCount); // Recursion } } else #endif { - for(uint32_t i = 0; i < heapCount; ++i, ++outBudget) + for(uint32_t i = 0; i < heapCount; ++i, ++outBudgets) { const uint32_t heapIndex = firstHeap + i; - outBudget->blockBytes = m_Budget.m_BlockBytes[heapIndex]; - outBudget->allocationBytes = m_Budget.m_AllocationBytes[heapIndex]; + outBudgets->statistics.blockCount = m_Budget.m_BlockCount[heapIndex]; + outBudgets->statistics.allocationCount = m_Budget.m_AllocationCount[heapIndex]; + outBudgets->statistics.blockBytes = m_Budget.m_BlockBytes[heapIndex]; + outBudgets->statistics.allocationBytes = m_Budget.m_AllocationBytes[heapIndex]; - outBudget->usage = outBudget->blockBytes; - outBudget->budget = m_MemProps.memoryHeaps[heapIndex].size * 8 / 10; // 80% heuristics. + outBudgets->usage = outBudgets->statistics.blockBytes; + outBudgets->budget = m_MemProps.memoryHeaps[heapIndex].size * 8 / 10; // 80% heuristics. } } } -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, info.flags); - - 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; -} - -VkResult VmaAllocator_T::DefragmentationPassBegin( - VmaDefragmentationPassInfo* pInfo, - VmaDefragmentationContext context) -{ - return context->DefragmentPassBegin(pInfo); -} -VkResult VmaAllocator_T::DefragmentationPassEnd( - VmaDefragmentationContext context) -{ - return context->DefragmentPassEnd(); - -} - 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 are 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; - } + pAllocationInfo->memoryType = hAllocation->GetMemoryTypeIndex(); + pAllocationInfo->deviceMemory = hAllocation->GetMemory(); + pAllocationInfo->offset = hAllocation->GetOffset(); + pAllocationInfo->size = hAllocation->GetSize(); + pAllocationInfo->pMappedData = hAllocation->GetMappedData(); + pAllocationInfo->pUserData = hAllocation->GetUserData(); + pAllocationInfo->pName = hAllocation->GetName(); } VkResult VmaAllocator_T::CreatePool(const VmaPoolCreateInfo* pCreateInfo, VmaPool* pPool) @@ -15558,9 +15295,18 @@ void VmaAllocator_T::DestroyPool(VmaPool pool) vma_delete(this, pool); } -void VmaAllocator_T::GetPoolStats(VmaPool pool, VmaPoolStats* pPoolStats) +void VmaAllocator_T::GetPoolStatistics(VmaPool pool, VmaStatistics* pPoolStats) { - pool->m_BlockVector.GetPoolStats(pPoolStats); + VmaClearStatistics(*pPoolStats); + pool->m_BlockVector.AddStatistics(*pPoolStats); + pool->m_DedicatedAllocations.AddStatistics(*pPoolStats); +} + +void VmaAllocator_T::CalculatePoolStatistics(VmaPool pool, VmaDetailedStatistics* pPoolStats) +{ + VmaClearDetailedStatistics(*pPoolStats); + pool->m_BlockVector.AddDetailedStatistics(*pPoolStats); + pool->m_DedicatedAllocations.AddDetailedStatistics(*pPoolStats); } void VmaAllocator_T::SetCurrentFrameIndex(uint32_t frameIndex) @@ -15575,15 +15321,6 @@ void VmaAllocator_T::SetCurrentFrameIndex(uint32_t frameIndex) #endif // #if VMA_MEMORY_BUDGET } -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(); @@ -15596,10 +15333,9 @@ VkResult VmaAllocator_T::CheckCorruption(uint32_t memoryTypeBits) // Process default pools. for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) { - if(((1u << memTypeIndex) & memoryTypeBits) != 0) + VmaBlockVector* const pBlockVector = m_pBlockVectors[memTypeIndex]; + if(pBlockVector != VMA_NULL) { - VmaBlockVector* const pBlockVector = m_pBlockVectors[memTypeIndex]; - VMA_ASSERT(pBlockVector); VkResult localRes = pBlockVector->CheckCorruption(); switch(localRes) { @@ -15639,37 +15375,6 @@ VkResult VmaAllocator_T::CheckCorruption(uint32_t memoryTypeBits) return finalRes; } -void VmaAllocator_T::CreateLostAllocation(VmaAllocation* pAllocation) -{ - *pAllocation = m_AllocationObjectAllocator.Allocate(VMA_FRAME_INDEX_LOST, false); - (*pAllocation)->InitLost(); -} - -// An object that increments given atomic but decrements it back in the destructor unless Commit() is called. -template -struct AtomicTransactionalIncrement -{ -public: - typedef std::atomic AtomicT; - ~AtomicTransactionalIncrement() - { - if(m_Atomic) - --(*m_Atomic); - } - T Increment(AtomicT* atomic) - { - m_Atomic = atomic; - return m_Atomic->fetch_add(1); - } - void Commit() - { - m_Atomic = nullptr; - } - -private: - AtomicT* m_Atomic = nullptr; -}; - VkResult VmaAllocator_T::AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory) { AtomicTransactionalIncrement deviceMemoryCountIncrement; @@ -15705,6 +15410,7 @@ VkResult VmaAllocator_T::AllocateVulkanMemory(const VkMemoryAllocateInfo* pAlloc { m_Budget.m_BlockBytes[heapIndex] += pAllocateInfo->allocationSize; } + ++m_Budget.m_BlockCount[heapIndex]; // VULKAN CALL vkAllocateMemory. VkResult res = (*m_VulkanFunctions.vkAllocateMemory)(m_hDevice, pAllocateInfo, GetAllocationCallbacks(), pMemory); @@ -15725,6 +15431,7 @@ VkResult VmaAllocator_T::AllocateVulkanMemory(const VkMemoryAllocateInfo* pAlloc } else { + --m_Budget.m_BlockCount[heapIndex]; m_Budget.m_BlockBytes[heapIndex] -= pAllocateInfo->allocationSize; } @@ -15742,7 +15449,9 @@ void VmaAllocator_T::FreeVulkanMemory(uint32_t memoryType, VkDeviceSize size, Vk // VULKAN CALL vkFreeMemory. (*m_VulkanFunctions.vkFreeMemory)(m_hDevice, hMemory, GetAllocationCallbacks()); - m_Budget.m_BlockBytes[MemoryTypeIndexToHeapIndex(memoryType)] -= size; + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memoryType); + --m_Budget.m_BlockCount[heapIndex]; + m_Budget.m_BlockBytes[heapIndex] -= size; --m_DeviceMemoryCount; } @@ -15811,11 +15520,6 @@ VkResult VmaAllocator_T::BindVulkanImage( 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: @@ -15872,7 +15576,7 @@ VkResult VmaAllocator_T::BindBufferMemory( case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: { VmaDeviceMemoryBlock* const pBlock = hAllocation->GetBlock(); - VMA_ASSERT(pBlock && "Binding buffer to allocation that doesn't belong to any block. Is the allocation lost?"); + VMA_ASSERT(pBlock && "Binding buffer to allocation that doesn't belong to any block."); res = pBlock->BindBufferMemory(this, hAllocation, allocationLocalOffset, hBuffer, pNext); break; } @@ -15897,7 +15601,7 @@ VkResult VmaAllocator_T::BindImageMemory( 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?"); + VMA_ASSERT(pBlock && "Binding image to allocation that doesn't belong to any block."); res = pBlock->BindImageMemory(this, hAllocation, allocationLocalOffset, hImage, pNext); break; } @@ -15979,10 +15683,16 @@ void VmaAllocator_T::FreeDedicatedMemory(const VmaAllocation allocation) VMA_ASSERT(allocation && allocation->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + VmaPool parentPool = allocation->GetParentPool(); + if(parentPool == VK_NULL_HANDLE) { - VmaMutexLockWrite lock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); - DedicatedAllocationLinkedList& dedicatedAllocations = m_DedicatedAllocations[memTypeIndex]; - dedicatedAllocations.Remove(allocation); + // Default pool + m_DedicatedAllocations[memTypeIndex].Unregister(allocation); + } + else + { + // Custom pool + parentPool->m_DedicatedAllocations.Unregister(allocation); } VkDeviceMemory hMemory = allocation->GetMemory(); @@ -15999,6 +15709,9 @@ void VmaAllocator_T::FreeDedicatedMemory(const VmaAllocation allocation) FreeVulkanMemory(memTypeIndex, allocation->GetSize(), hMemory); + m_Budget.RemoveAllocation(MemoryTypeIndexToHeapIndex(allocation->GetMemoryTypeIndex()), allocation->GetSize()); + m_AllocationObjectAllocator.Free(allocation); + VMA_DEBUG_LOG(" Freed DedicatedMemory MemoryTypeIndex=%u", memTypeIndex); } @@ -16113,7 +15826,6 @@ bool VmaAllocator_T::GetFlushOrInvalidateRange( } #if VMA_MEMORY_BUDGET - void VmaAllocator_T::UpdateVulkanBudget() { VMA_ASSERT(m_UseExtMemoryBudget); @@ -16151,13 +15863,12 @@ void VmaAllocator_T::UpdateVulkanBudget() m_Budget.m_OperationsSinceBudgetFetch = 0; } } - -#endif // #if VMA_MEMORY_BUDGET +#endif // VMA_MEMORY_BUDGET void VmaAllocator_T::FillAllocation(const VmaAllocation hAllocation, uint8_t pattern) { if(VMA_DEBUG_INITIALIZE_ALLOCATIONS && - !hAllocation->CanBecomeLost() && + hAllocation->IsMappingAllowed() && (m_MemProps.memoryTypes[hAllocation->GetMemoryTypeIndex()].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0) { void* pData = VMA_NULL; @@ -16187,106 +15898,114 @@ uint32_t VmaAllocator_T::GetGpuDefragmentationMemoryTypeBits() } #if VMA_STATS_STRING_ENABLED - void VmaAllocator_T::PrintDetailedMap(VmaJsonWriter& json) { - bool dedicatedAllocationsStarted = false; - for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + json.WriteString("DefaultPools"); + json.BeginObject(); { - VmaMutexLockRead dedicatedAllocationsLock(m_DedicatedAllocationsMutex[memTypeIndex], m_UseMutex); - DedicatedAllocationLinkedList& dedicatedAllocList = m_DedicatedAllocations[memTypeIndex]; - if(!dedicatedAllocList.IsEmpty()) + for (uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) { - if(dedicatedAllocationsStarted == false) + VmaBlockVector* pBlockVector = m_pBlockVectors[memTypeIndex]; + VmaDedicatedAllocationList& dedicatedAllocList = m_DedicatedAllocations[memTypeIndex]; + if (pBlockVector != VMA_NULL) { - dedicatedAllocationsStarted = true; - json.WriteString("DedicatedAllocations"); - json.BeginObject(); - } - - json.BeginString("Type "); - json.ContinueString(memTypeIndex); - json.EndString(); - - json.BeginArray(); - - for(VmaAllocation alloc = dedicatedAllocList.Front(); - alloc != VMA_NULL; alloc = dedicatedAllocList.GetNext(alloc)) - { - json.BeginObject(true); - alloc->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(); + json.BeginObject(); + { + json.WriteString("PreferredBlockSize"); + json.WriteNumber(pBlockVector->GetPreferredBlockSize()); - m_pBlockVectors[memTypeIndex]->PrintDetailedMap(json); + json.WriteString("Blocks"); + pBlockVector->PrintDetailedMap(json); + + json.WriteString("DedicatedAllocations"); + dedicatedAllocList.BuildStatsString(json); + } + json.EndObject(); } } - if(allocationsStarted) - { - json.EndObject(); - } } + json.EndObject(); - // Custom pools + json.WriteString("CustomPools"); + json.BeginObject(); { VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); - if(!m_Pools.IsEmpty()) + if (!m_Pools.IsEmpty()) { - json.WriteString("Pools"); - json.BeginObject(); - for(VmaPool pool = m_Pools.Front(); pool != VMA_NULL; pool = m_Pools.GetNext(pool)) + for (uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) { - json.BeginString(); - json.ContinueString(pool->GetId()); - json.EndString(); + bool displayType = true; + size_t index = 0; + for (VmaPool pool = m_Pools.Front(); pool != VMA_NULL; pool = m_Pools.GetNext(pool)) + { + VmaBlockVector& blockVector = pool->m_BlockVector; + if (blockVector.GetMemoryTypeIndex() == memTypeIndex) + { + if (displayType) + { + json.BeginString("Type "); + json.ContinueString(memTypeIndex); + json.EndString(); + json.BeginArray(); + displayType = false; + } - pool->m_BlockVector.PrintDetailedMap(json); + json.BeginObject(); + { + json.WriteString("Name"); + json.BeginString(); + json.ContinueString_Size(index++); + if (pool->GetName()) + { + json.ContinueString(" - "); + json.ContinueString(pool->GetName()); + } + json.EndString(); + + json.WriteString("PreferredBlockSize"); + json.WriteNumber(blockVector.GetPreferredBlockSize()); + + json.WriteString("Blocks"); + blockVector.PrintDetailedMap(json); + + json.WriteString("DedicatedAllocations"); + pool->m_DedicatedAllocations.BuildStatsString(json); + } + json.EndObject(); + } + } + + if (!displayType) + json.EndArray(); } - json.EndObject(); } } + json.EndObject(); } +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_ALLOCATOR_T_FUNCTIONS -#endif // #if VMA_STATS_STRING_ENABLED - -//////////////////////////////////////////////////////////////////////////////// -// Public interface +#ifndef _VMA_PUBLIC_INTERFACE VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAllocator( const VmaAllocatorCreateInfo* pCreateInfo, VmaAllocator* pAllocator) { VMA_ASSERT(pCreateInfo && pAllocator); VMA_ASSERT(pCreateInfo->vulkanApiVersion == 0 || - (VK_VERSION_MAJOR(pCreateInfo->vulkanApiVersion) == 1 && VK_VERSION_MINOR(pCreateInfo->vulkanApiVersion) <= 2)); + (VK_VERSION_MAJOR(pCreateInfo->vulkanApiVersion) == 1 && VK_VERSION_MINOR(pCreateInfo->vulkanApiVersion) <= 3)); VMA_DEBUG_LOG("vmaCreateAllocator"); *pAllocator = vma_new(pCreateInfo->pAllocationCallbacks, VmaAllocator_T)(pCreateInfo); - return (*pAllocator)->Init(pCreateInfo); + VkResult result = (*pAllocator)->Init(pCreateInfo); + if(result < 0) + { + vma_delete(pCreateInfo->pAllocationCallbacks, *pAllocator); + *pAllocator = VK_NULL_HANDLE; + } + return result; } VMA_CALL_PRE void VMA_CALL_POST vmaDestroyAllocator( @@ -16295,7 +16014,7 @@ VMA_CALL_PRE void VMA_CALL_POST vmaDestroyAllocator( if(allocator != VK_NULL_HANDLE) { VMA_DEBUG_LOG("vmaDestroyAllocator"); - VkAllocationCallbacks allocationCallbacks = allocator->m_AllocationCallbacks; + VkAllocationCallbacks allocationCallbacks = allocator->m_AllocationCallbacks; // Have to copy the callbacks when destroying. vma_delete(&allocationCallbacks, allocator); } } @@ -16339,29 +16058,28 @@ VMA_CALL_PRE void VMA_CALL_POST vmaSetCurrentFrameIndex( uint32_t frameIndex) { VMA_ASSERT(allocator); - VMA_ASSERT(frameIndex != VMA_FRAME_INDEX_LOST); VMA_DEBUG_GLOBAL_MUTEX_LOCK allocator->SetCurrentFrameIndex(frameIndex); } -VMA_CALL_PRE void VMA_CALL_POST vmaCalculateStats( +VMA_CALL_PRE void VMA_CALL_POST vmaCalculateStatistics( VmaAllocator allocator, - VmaStats* pStats) + VmaTotalStatistics* pStats) { VMA_ASSERT(allocator && pStats); VMA_DEBUG_GLOBAL_MUTEX_LOCK - allocator->CalculateStats(pStats); + allocator->CalculateStatistics(pStats); } -VMA_CALL_PRE void VMA_CALL_POST vmaGetBudget( +VMA_CALL_PRE void VMA_CALL_POST vmaGetHeapBudgets( VmaAllocator allocator, - VmaBudget* pBudget) + VmaBudget* pBudgets) { - VMA_ASSERT(allocator && pBudget); + VMA_ASSERT(allocator && pBudgets); VMA_DEBUG_GLOBAL_MUTEX_LOCK - allocator->GetBudget(pBudget, 0, allocator->GetMemoryHeapCount()); + allocator->GetHeapBudgets(pBudgets, 0, allocator->GetMemoryHeapCount()); } #if VMA_STATS_STRING_ENABLED @@ -16374,137 +16092,183 @@ VMA_CALL_PRE void VMA_CALL_POST vmaBuildStatsString( VMA_ASSERT(allocator && ppStatsString); VMA_DEBUG_GLOBAL_MUTEX_LOCK - VmaStringBuilder sb(allocator); + VmaStringBuilder sb(allocator->GetAllocationCallbacks()); { + VmaBudget budgets[VK_MAX_MEMORY_HEAPS]; + allocator->GetHeapBudgets(budgets, 0, allocator->GetMemoryHeapCount()); + + VmaTotalStatistics stats; + allocator->CalculateStatistics(&stats); + VmaJsonWriter json(allocator->GetAllocationCallbacks(), sb); json.BeginObject(); - - VmaBudget budget[VK_MAX_MEMORY_HEAPS]; - allocator->GetBudget(budget, 0, allocator->GetMemoryHeapCount()); - - 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(); - - json.WriteString("Budget"); + json.WriteString("General"); json.BeginObject(); { - json.WriteString("BlockBytes"); - json.WriteNumber(budget[heapIndex].blockBytes); - json.WriteString("AllocationBytes"); - json.WriteNumber(budget[heapIndex].allocationBytes); - json.WriteString("Usage"); - json.WriteNumber(budget[heapIndex].usage); - json.WriteString("Budget"); - json.WriteNumber(budget[heapIndex].budget); + const VkPhysicalDeviceProperties& deviceProperties = allocator->m_PhysicalDeviceProperties; + const VkPhysicalDeviceMemoryProperties& memoryProperties = allocator->m_MemProps; + + json.WriteString("API"); + json.WriteString("Vulkan"); + + json.WriteString("apiVersion"); + json.BeginString(); + json.ContinueString(VK_API_VERSION_MAJOR(deviceProperties.apiVersion)); + json.ContinueString("."); + json.ContinueString(VK_API_VERSION_MINOR(deviceProperties.apiVersion)); + json.ContinueString("."); + json.ContinueString(VK_API_VERSION_PATCH(deviceProperties.apiVersion)); + json.EndString(); + + json.WriteString("GPU"); + json.WriteString(deviceProperties.deviceName); + json.WriteString("deviceType"); + json.WriteNumber(static_cast(deviceProperties.deviceType)); + + json.WriteString("maxMemoryAllocationCount"); + json.WriteNumber(deviceProperties.limits.maxMemoryAllocationCount); + json.WriteString("bufferImageGranularity"); + json.WriteNumber(deviceProperties.limits.bufferImageGranularity); + json.WriteString("nonCoherentAtomSize"); + json.WriteNumber(deviceProperties.limits.nonCoherentAtomSize); + + json.WriteString("memoryHeapCount"); + json.WriteNumber(memoryProperties.memoryHeapCount); + json.WriteString("memoryTypeCount"); + json.WriteNumber(memoryProperties.memoryTypeCount); } json.EndObject(); - - if(stats.memoryHeap[heapIndex].blockCount > 0) + } + { + json.WriteString("Total"); + VmaPrintDetailedStatistics(json, stats.total); + } + { + json.WriteString("MemoryInfo"); + json.BeginObject(); { - json.WriteString("Stats"); - VmaPrintStatInfo(json, stats.memoryHeap[heapIndex]); - } - - for(uint32_t typeIndex = 0; typeIndex < allocator->GetMemoryTypeCount(); ++typeIndex) - { - if(allocator->MemoryTypeIndexToHeapIndex(typeIndex) == heapIndex) + for (uint32_t heapIndex = 0; heapIndex < allocator->GetMemoryHeapCount(); ++heapIndex) { - json.BeginString("Type "); - json.ContinueString(typeIndex); + json.BeginString("Heap "); + json.ContinueString(heapIndex); json.EndString(); - json.BeginObject(); + { + const VkMemoryHeap& heapInfo = allocator->m_MemProps.memoryHeaps[heapIndex]; + json.WriteString("Flags"); + json.BeginArray(true); + { + if (heapInfo.flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) + json.WriteString("DEVICE_LOCAL"); + #if VMA_VULKAN_VERSION >= 1001000 + if (heapInfo.flags & VK_MEMORY_HEAP_MULTI_INSTANCE_BIT) + json.WriteString("MULTI_INSTANCE"); + #endif - 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"); - } -#if VMA_VULKAN_VERSION >= 1001000 - if((flags & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) - { - json.WriteString("PROTECTED"); - } -#endif // #if VMA_VULKAN_VERSION >= 1001000 -#if VK_AMD_device_coherent_memory - if((flags & VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY) != 0) - { - json.WriteString("DEVICE_COHERENT"); - } - if((flags & VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY) != 0) - { - json.WriteString("DEVICE_UNCACHED"); - } -#endif // #if VK_AMD_device_coherent_memory - json.EndArray(); + VkMemoryHeapFlags flags = heapInfo.flags & + ~(VK_MEMORY_HEAP_DEVICE_LOCAL_BIT + #if VMA_VULKAN_VERSION >= 1001000 + | VK_MEMORY_HEAP_MULTI_INSTANCE_BIT + #endif + ); + if (flags != 0) + json.WriteNumber(flags); + } + json.EndArray(); + + json.WriteString("Size"); + json.WriteNumber(heapInfo.size); + + json.WriteString("Budget"); + json.BeginObject(); + { + json.WriteString("BudgetBytes"); + json.WriteNumber(budgets[heapIndex].budget); + json.WriteString("UsageBytes"); + json.WriteNumber(budgets[heapIndex].usage); + } + json.EndObject(); - if(stats.memoryType[typeIndex].blockCount > 0) - { json.WriteString("Stats"); - VmaPrintStatInfo(json, stats.memoryType[typeIndex]); - } + VmaPrintDetailedStatistics(json, stats.memoryHeap[heapIndex]); + json.WriteString("MemoryPools"); + json.BeginObject(); + { + 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) + json.WriteString("DEVICE_LOCAL"); + if (flags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) + json.WriteString("HOST_VISIBLE"); + if (flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) + json.WriteString("HOST_COHERENT"); + if (flags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) + json.WriteString("HOST_CACHED"); + if (flags & VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT) + json.WriteString("LAZILY_ALLOCATED"); + #if VMA_VULKAN_VERSION >= 1001000 + if (flags & VK_MEMORY_PROPERTY_PROTECTED_BIT) + json.WriteString("PROTECTED"); + #endif + #if VK_AMD_device_coherent_memory + if (flags & VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY) + json.WriteString("DEVICE_COHERENT_AMD"); + if (flags & VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY) + json.WriteString("DEVICE_UNCACHED_AMD"); + #endif + + flags &= ~(VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT + #if VMA_VULKAN_VERSION >= 1001000 + | VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT + #endif + #if VK_AMD_device_coherent_memory + | VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY + | VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY + #endif + | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT + | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT + | VK_MEMORY_PROPERTY_HOST_CACHED_BIT); + if (flags != 0) + json.WriteNumber(flags); + } + json.EndArray(); + + json.WriteString("Stats"); + VmaPrintDetailedStatistics(json, stats.memoryType[typeIndex]); + } + json.EndObject(); + } + } + + } + json.EndObject(); + } json.EndObject(); } } - json.EndObject(); } - if(detailedMap == VK_TRUE) - { + + 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; + *ppStatsString = VmaCreateStringCopy(allocator->GetAllocationCallbacks(), sb.GetData(), sb.GetLength()); } VMA_CALL_PRE void VMA_CALL_POST vmaFreeStatsString( @@ -16514,12 +16278,11 @@ VMA_CALL_PRE void VMA_CALL_POST vmaFreeStatsString( if(pStatsString != VMA_NULL) { VMA_ASSERT(allocator); - size_t len = strlen(pStatsString); - vma_delete_array(allocator, pStatsString, len + 1); + VmaFreeString(allocator->GetAllocationCallbacks(), pStatsString); } } -#endif // #if VMA_STATS_STRING_ENABLED +#endif // VMA_STATS_STRING_ENABLED /* This function is not protected by any mutex because it just reads immutable data. @@ -16534,91 +16297,7 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndex( VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); - memoryTypeBits &= allocator->GetGlobalMemoryTypeBits(); - - if(pAllocationCreateInfo->memoryTypeBits != 0) - { - memoryTypeBits &= pAllocationCreateInfo->memoryTypeBits; - } - - uint32_t requiredFlags = pAllocationCreateInfo->requiredFlags; - uint32_t preferredFlags = pAllocationCreateInfo->preferredFlags; - uint32_t notPreferredFlags = 0; - - // 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_CACHED_BIT; - break; - case VMA_MEMORY_USAGE_CPU_COPY: - notPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; - break; - case VMA_MEMORY_USAGE_GPU_LAZILY_ALLOCATED: - requiredFlags |= VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT; - break; - default: - VMA_ASSERT(0); - break; - } - - // Avoid DEVICE_COHERENT unless explicitly requested. - if(((pAllocationCreateInfo->requiredFlags | pAllocationCreateInfo->preferredFlags) & - (VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY | VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD_COPY)) == 0) - { - notPreferredFlags |= VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY; - } - - *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) + - VmaCountBitsSet(currFlags & notPreferredFlags); - // 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; + return allocator->FindMemoryTypeIndex(memoryTypeBits, pAllocationCreateInfo, UINT32_MAX, pMemoryTypeIndex); } VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( @@ -16633,23 +16312,40 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( 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) + const VmaVulkanFunctions* funcs = &allocator->GetVulkanFunctions(); + VkResult res; + +#if VMA_VULKAN_VERSION >= 1003000 + if(funcs->vkGetDeviceBufferMemoryRequirements) { - VkMemoryRequirements memReq = {}; - allocator->GetVulkanFunctions().vkGetBufferMemoryRequirements( - hDev, hBuffer, &memReq); + // Can query straight from VkBufferCreateInfo :) + VkDeviceBufferMemoryRequirements devBufMemReq = {VK_STRUCTURE_TYPE_DEVICE_BUFFER_MEMORY_REQUIREMENTS}; + devBufMemReq.pCreateInfo = pBufferCreateInfo; - res = vmaFindMemoryTypeIndex( - allocator, - memReq.memoryTypeBits, - pAllocationCreateInfo, - pMemoryTypeIndex); + VkMemoryRequirements2 memReq = {VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2}; + (*funcs->vkGetDeviceBufferMemoryRequirements)(hDev, &devBufMemReq, &memReq); - allocator->GetVulkanFunctions().vkDestroyBuffer( - hDev, hBuffer, allocator->GetAllocationCallbacks()); + res = allocator->FindMemoryTypeIndex( + memReq.memoryRequirements.memoryTypeBits, pAllocationCreateInfo, pBufferCreateInfo->usage, pMemoryTypeIndex); + } + else +#endif // #if VMA_VULKAN_VERSION >= 1003000 + { + // Must create a dummy buffer to query :( + VkBuffer hBuffer = VK_NULL_HANDLE; + res = funcs->vkCreateBuffer( + hDev, pBufferCreateInfo, allocator->GetAllocationCallbacks(), &hBuffer); + if(res == VK_SUCCESS) + { + VkMemoryRequirements memReq = {}; + funcs->vkGetBufferMemoryRequirements(hDev, hBuffer, &memReq); + + res = allocator->FindMemoryTypeIndex( + memReq.memoryTypeBits, pAllocationCreateInfo, pBufferCreateInfo->usage, pMemoryTypeIndex); + + funcs->vkDestroyBuffer( + hDev, hBuffer, allocator->GetAllocationCallbacks()); + } } return res; } @@ -16666,23 +16362,42 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForImageInfo( 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) + const VmaVulkanFunctions* funcs = &allocator->GetVulkanFunctions(); + VkResult res; + +#if VMA_VULKAN_VERSION >= 1003000 + if(funcs->vkGetDeviceImageMemoryRequirements) { - VkMemoryRequirements memReq = {}; - allocator->GetVulkanFunctions().vkGetImageMemoryRequirements( - hDev, hImage, &memReq); + // Can query straight from VkImageCreateInfo :) + VkDeviceImageMemoryRequirements devImgMemReq = {VK_STRUCTURE_TYPE_DEVICE_IMAGE_MEMORY_REQUIREMENTS}; + devImgMemReq.pCreateInfo = pImageCreateInfo; + VMA_ASSERT(pImageCreateInfo->tiling != VK_IMAGE_TILING_DRM_FORMAT_MODIFIER_EXT_COPY && (pImageCreateInfo->flags & VK_IMAGE_CREATE_DISJOINT_BIT_COPY) == 0 && + "Cannot use this VkImageCreateInfo with vmaFindMemoryTypeIndexForImageInfo as I don't know what to pass as VkDeviceImageMemoryRequirements::planeAspect."); - res = vmaFindMemoryTypeIndex( - allocator, - memReq.memoryTypeBits, - pAllocationCreateInfo, - pMemoryTypeIndex); + VkMemoryRequirements2 memReq = {VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2}; + (*funcs->vkGetDeviceImageMemoryRequirements)(hDev, &devImgMemReq, &memReq); - allocator->GetVulkanFunctions().vkDestroyImage( - hDev, hImage, allocator->GetAllocationCallbacks()); + res = allocator->FindMemoryTypeIndex( + memReq.memoryRequirements.memoryTypeBits, pAllocationCreateInfo, pImageCreateInfo->usage, pMemoryTypeIndex); + } + else +#endif // #if VMA_VULKAN_VERSION >= 1003000 + { + // Must create a dummy image to query :( + VkImage hImage = VK_NULL_HANDLE; + res = funcs->vkCreateImage( + hDev, pImageCreateInfo, allocator->GetAllocationCallbacks(), &hImage); + if(res == VK_SUCCESS) + { + VkMemoryRequirements memReq = {}; + funcs->vkGetImageMemoryRequirements(hDev, hImage, &memReq); + + res = allocator->FindMemoryTypeIndex( + memReq.memoryTypeBits, pAllocationCreateInfo, pImageCreateInfo->usage, pMemoryTypeIndex); + + funcs->vkDestroyImage( + hDev, hImage, allocator->GetAllocationCallbacks()); + } } return res; } @@ -16698,16 +16413,7 @@ VMA_CALL_PRE VkResult VMA_CALL_POST 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; + return allocator->CreatePool(pCreateInfo, pPool); } VMA_CALL_PRE void VMA_CALL_POST vmaDestroyPool( @@ -16725,45 +16431,31 @@ VMA_CALL_PRE void VMA_CALL_POST 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); } -VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStats( +VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStatistics( VmaAllocator allocator, VmaPool pool, - VmaPoolStats* pPoolStats) + VmaStatistics* pPoolStats) { VMA_ASSERT(allocator && pool && pPoolStats); VMA_DEBUG_GLOBAL_MUTEX_LOCK - allocator->GetPoolStats(pool, pPoolStats); + allocator->GetPoolStatistics(pool, pPoolStats); } -VMA_CALL_PRE void VMA_CALL_POST vmaMakePoolAllocationsLost( +VMA_CALL_PRE void VMA_CALL_POST vmaCalculatePoolStatistics( VmaAllocator allocator, VmaPool pool, - size_t* pLostAllocationCount) + VmaDetailedStatistics* pPoolStats) { - VMA_ASSERT(allocator && pool); + VMA_ASSERT(allocator && pool && pPoolStats); 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); + allocator->CalculatePoolStatistics(pool, pPoolStats); } VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckPoolCorruption(VmaAllocator allocator, VmaPool pool) @@ -16803,13 +16495,6 @@ VMA_CALL_PRE void VMA_CALL_POST vmaSetPoolName( VMA_DEBUG_GLOBAL_MUTEX_LOCK pool->SetName(pName); - -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - allocator->GetRecorder()->RecordSetPoolName(allocator->GetCurrentFrameIndex(), pool, pName); - } -#endif } VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemory( @@ -16830,24 +16515,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemory( false, // requiresDedicatedAllocation false, // prefersDedicatedAllocation VK_NULL_HANDLE, // dedicatedBuffer - UINT32_MAX, // dedicatedBufferUsage VK_NULL_HANDLE, // dedicatedImage + UINT32_MAX, // dedicatedBufferImageUsage *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); @@ -16880,25 +16554,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryPages( false, // requiresDedicatedAllocation false, // prefersDedicatedAllocation VK_NULL_HANDLE, // dedicatedBuffer - UINT32_MAX, // dedicatedBufferUsage VK_NULL_HANDLE, // dedicatedImage + UINT32_MAX, // dedicatedBufferImageUsage *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) @@ -16935,26 +16597,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForBuffer( requiresDedicatedAllocation, prefersDedicatedAllocation, buffer, // dedicatedBuffer - UINT32_MAX, // dedicatedBufferUsage VK_NULL_HANDLE, // dedicatedImage + UINT32_MAX, // dedicatedBufferImageUsage *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); @@ -16987,26 +16636,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForImage( requiresDedicatedAllocation, prefersDedicatedAllocation, VK_NULL_HANDLE, // dedicatedBuffer - UINT32_MAX, // dedicatedBufferUsage image, // dedicatedImage + UINT32_MAX, // dedicatedBufferImageUsage *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); @@ -17030,15 +16666,6 @@ VMA_CALL_PRE void VMA_CALL_POST 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); @@ -17060,16 +16687,6 @@ VMA_CALL_PRE void VMA_CALL_POST 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); } @@ -17082,38 +16699,9 @@ VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationInfo( 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); } -VMA_CALL_PRE VkBool32 VMA_CALL_POST 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); -} - VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationUserData( VmaAllocator allocator, VmaAllocation allocation, @@ -17124,36 +16712,24 @@ VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationUserData( 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 } -VMA_CALL_PRE void VMA_CALL_POST vmaCreateLostAllocation( - VmaAllocator allocator, - VmaAllocation* pAllocation) +VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationName( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const char* VMA_NULLABLE pName) { - VMA_ASSERT(allocator && pAllocation); + allocation->SetName(allocator, pName); +} - VMA_DEBUG_GLOBAL_MUTEX_LOCK; - - allocator->CreateLostAllocation(pAllocation); - -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - allocator->GetRecorder()->RecordCreateLostAllocation( - allocator->GetCurrentFrameIndex(), - *pAllocation); - } -#endif +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationMemoryProperties( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkMemoryPropertyFlags* VMA_NOT_NULL pFlags) +{ + VMA_ASSERT(allocator && allocation && pFlags); + const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + *pFlags = allocator->m_MemProps.memoryTypes[memTypeIndex].propertyFlags; } VMA_CALL_PRE VkResult VMA_CALL_POST vmaMapMemory( @@ -17165,18 +16741,7 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaMapMemory( 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; + return allocator->Map(allocation, ppData); } VMA_CALL_PRE void VMA_CALL_POST vmaUnmapMemory( @@ -17187,19 +16752,14 @@ VMA_CALL_PRE void VMA_CALL_POST vmaUnmapMemory( VMA_DEBUG_GLOBAL_MUTEX_LOCK -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - allocator->GetRecorder()->RecordUnmapMemory( - allocator->GetCurrentFrameIndex(), - allocation); - } -#endif - allocator->Unmap(allocation); } -VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize offset, + VkDeviceSize size) { VMA_ASSERT(allocator && allocation); @@ -17209,19 +16769,14 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocation(VmaAllocator allocator, V const VkResult res = 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 - return res; } -VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocation(VmaAllocator allocator, VmaAllocation allocation, VkDeviceSize offset, VkDeviceSize size) +VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize offset, + VkDeviceSize size) { VMA_ASSERT(allocator && allocation); @@ -17231,15 +16786,6 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocation(VmaAllocator allocat const VkResult res = 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 - return res; } @@ -17265,13 +16811,6 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocations( const VkResult res = allocator->FlushOrInvalidateAllocations(allocationCount, allocations, offsets, sizes, VMA_CACHE_FLUSH); -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - //TODO - } -#endif - return res; } @@ -17297,17 +16836,12 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocations( const VkResult res = allocator->FlushOrInvalidateAllocations(allocationCount, allocations, offsets, sizes, VMA_CACHE_INVALIDATE); -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - //TODO - } -#endif - return res; } -VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption(VmaAllocator allocator, uint32_t memoryTypeBits) +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption( + VmaAllocator allocator, + uint32_t memoryTypeBits) { VMA_ASSERT(allocator); @@ -17318,139 +16852,70 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption(VmaAllocator allocator, u return allocator->CheckCorruption(memoryTypeBits); } -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragment( +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentation( VmaAllocator allocator, - const 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; -} - -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragmentationBegin( - VmaAllocator allocator, - const VmaDefragmentationInfo2* pInfo, - VmaDefragmentationStats* pStats, - VmaDefragmentationContext *pContext) + const VmaDefragmentationInfo* pInfo, + VmaDefragmentationContext* pContext) { VMA_ASSERT(allocator && pInfo && pContext); - // Degenerate case: Nothing to defragment. - if(pInfo->allocationCount == 0 && pInfo->poolCount == 0) + VMA_DEBUG_LOG("vmaBeginDefragmentation"); + + if (pInfo->pool != VMA_NULL) { - return VK_SUCCESS; + // Check if run on supported algorithms + if (pInfo->pool->m_BlockVector.GetAlgorithm() & VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) + return VK_ERROR_FEATURE_NOT_PRESENT; } - 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; + *pContext = vma_new(allocator, VmaDefragmentationContext_T)(allocator, *pInfo); + return VK_SUCCESS; } -VMA_CALL_PRE VkResult VMA_CALL_POST vmaDefragmentationEnd( +VMA_CALL_PRE void VMA_CALL_POST vmaEndDefragmentation( VmaAllocator allocator, - VmaDefragmentationContext context) + VmaDefragmentationContext context, + VmaDefragmentationStats* pStats) { - VMA_ASSERT(allocator); + VMA_ASSERT(allocator && context); - VMA_DEBUG_LOG("vmaDefragmentationEnd"); + VMA_DEBUG_LOG("vmaEndDefragmentation"); - if(context != VK_NULL_HANDLE) - { - VMA_DEBUG_GLOBAL_MUTEX_LOCK + 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; - } + if (pStats) + context->GetStats(*pStats); + vma_delete(allocator, context); } VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentationPass( - VmaAllocator allocator, - VmaDefragmentationContext context, - VmaDefragmentationPassInfo* pInfo - ) + VmaAllocator VMA_NOT_NULL allocator, + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo) { - VMA_ASSERT(allocator); - VMA_ASSERT(pInfo); + VMA_ASSERT(context && pPassInfo); VMA_DEBUG_LOG("vmaBeginDefragmentationPass"); VMA_DEBUG_GLOBAL_MUTEX_LOCK - if(context == VK_NULL_HANDLE) - { - pInfo->moveCount = 0; - return VK_SUCCESS; - } - - return allocator->DefragmentationPassBegin(pInfo, context); + return context->DefragmentPassBegin(*pPassInfo); } + VMA_CALL_PRE VkResult VMA_CALL_POST vmaEndDefragmentationPass( - VmaAllocator allocator, - VmaDefragmentationContext context) + VmaAllocator VMA_NOT_NULL allocator, + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo) { - VMA_ASSERT(allocator); + VMA_ASSERT(context && pPassInfo); VMA_DEBUG_LOG("vmaEndDefragmentationPass"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK - if(context == VK_NULL_HANDLE) - return VK_SUCCESS; - - return allocator->DefragmentationPassEnd(context); + return context->DefragmentPassEnd(*pPassInfo); } VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory( @@ -17562,24 +17027,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBuffer( requiresDedicatedAllocation, prefersDedicatedAllocation, *pBuffer, // dedicatedBuffer - pBufferCreateInfo->usage, // dedicatedBufferUsage VK_NULL_HANDLE, // dedicatedImage + pBufferCreateInfo->usage, // dedicatedBufferImageUsage *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. @@ -17668,20 +17122,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBufferWithAlignment( requiresDedicatedAllocation, prefersDedicatedAllocation, *pBuffer, // dedicatedBuffer - pBufferCreateInfo->usage, // dedicatedBufferUsage VK_NULL_HANDLE, // dedicatedImage + pBufferCreateInfo->usage, // dedicatedBufferImageUsage *pAllocationCreateInfo, VMA_SUBALLOCATION_TYPE_BUFFER, 1, // allocationCount pAllocation); -#if VMA_RECORDING_ENABLED - if(allocator->GetRecorder() != VMA_NULL) - { - VMA_ASSERT(0 && "Not implemented."); - } -#endif - if(res >= 0) { // 3. Bind buffer with memory. @@ -17717,6 +17164,50 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBufferWithAlignment( return res; } +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingBuffer( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pBuffer) +{ + VMA_ASSERT(allocator && pBufferCreateInfo && pBuffer && allocation); + + VMA_DEBUG_LOG("vmaCreateAliasingBuffer"); + + *pBuffer = VK_NULL_HANDLE; + + if (pBufferCreateInfo->size == 0) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + if ((pBufferCreateInfo->usage & VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_COPY) != 0 && + !allocator->m_UseKhrBufferDeviceAddress) + { + VMA_ASSERT(0 && "Creating a buffer with VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT is not valid if VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT was not used."); + return VK_ERROR_INITIALIZATION_FAILED; + } + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + // 1. Create VkBuffer. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateBuffer)( + allocator->m_hDevice, + pBufferCreateInfo, + allocator->GetAllocationCallbacks(), + pBuffer); + if (res >= 0) + { + // 2. Bind buffer with memory. + res = allocator->BindBufferMemory(allocation, 0, *pBuffer, VMA_NULL); + if (res >= 0) + { + return VK_SUCCESS; + } + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + } + return res; +} + VMA_CALL_PRE void VMA_CALL_POST vmaDestroyBuffer( VmaAllocator allocator, VkBuffer buffer, @@ -17733,15 +17224,6 @@ VMA_CALL_PRE void VMA_CALL_POST 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()); @@ -17805,24 +17287,13 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateImage( requiresDedicatedAllocation, prefersDedicatedAllocation, VK_NULL_HANDLE, // dedicatedBuffer - UINT32_MAX, // dedicatedBufferUsage *pImage, // dedicatedImage + pImageCreateInfo->usage, // dedicatedBufferImageUsage *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. @@ -17858,10 +17329,52 @@ VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateImage( return res; } +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingImage( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const VkImageCreateInfo* VMA_NOT_NULL pImageCreateInfo, + VkImage VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pImage) +{ + VMA_ASSERT(allocator && pImageCreateInfo && pImage && allocation); + + *pImage = VK_NULL_HANDLE; + + VMA_DEBUG_LOG("vmaCreateImage"); + + if (pImageCreateInfo->extent.width == 0 || + pImageCreateInfo->extent.height == 0 || + pImageCreateInfo->extent.depth == 0 || + pImageCreateInfo->mipLevels == 0 || + pImageCreateInfo->arrayLayers == 0) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + // 1. Create VkImage. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateImage)( + allocator->m_hDevice, + pImageCreateInfo, + allocator->GetAllocationCallbacks(), + pImage); + if (res >= 0) + { + // 2. Bind image with memory. + res = allocator->BindImageMemory(allocation, 0, *pImage, VMA_NULL); + if (res >= 0) + { + return VK_SUCCESS; + } + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, *pImage, allocator->GetAllocationCallbacks()); + } + return res; +} + VMA_CALL_PRE void VMA_CALL_POST vmaDestroyImage( - VmaAllocator allocator, - VkImage image, - VmaAllocation allocation) + VmaAllocator VMA_NOT_NULL allocator, + VkImage VMA_NULLABLE_NON_DISPATCHABLE image, + VmaAllocation VMA_NULLABLE allocation) { VMA_ASSERT(allocator); @@ -17874,15 +17387,6 @@ VMA_CALL_PRE void VMA_CALL_POST 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()); @@ -17895,7 +17399,134 @@ VMA_CALL_PRE void VMA_CALL_POST vmaDestroyImage( } } -#endif // #ifdef VMA_IMPLEMENTATION +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateVirtualBlock( + const VmaVirtualBlockCreateInfo* VMA_NOT_NULL pCreateInfo, + VmaVirtualBlock VMA_NULLABLE * VMA_NOT_NULL pVirtualBlock) +{ + VMA_ASSERT(pCreateInfo && pVirtualBlock); + VMA_ASSERT(pCreateInfo->size > 0); + VMA_DEBUG_LOG("vmaCreateVirtualBlock"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + *pVirtualBlock = vma_new(pCreateInfo->pAllocationCallbacks, VmaVirtualBlock_T)(*pCreateInfo); + VkResult res = (*pVirtualBlock)->Init(); + if(res < 0) + { + vma_delete(pCreateInfo->pAllocationCallbacks, *pVirtualBlock); + *pVirtualBlock = VK_NULL_HANDLE; + } + return res; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyVirtualBlock(VmaVirtualBlock VMA_NULLABLE virtualBlock) +{ + if(virtualBlock != VK_NULL_HANDLE) + { + VMA_DEBUG_LOG("vmaDestroyVirtualBlock"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + VkAllocationCallbacks allocationCallbacks = virtualBlock->m_AllocationCallbacks; // Have to copy the callbacks when destroying. + vma_delete(&allocationCallbacks, virtualBlock); + } +} + +VMA_CALL_PRE VkBool32 VMA_CALL_POST vmaIsVirtualBlockEmpty(VmaVirtualBlock VMA_NOT_NULL virtualBlock) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE); + VMA_DEBUG_LOG("vmaIsVirtualBlockEmpty"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + return virtualBlock->IsEmpty() ? VK_TRUE : VK_FALSE; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetVirtualAllocationInfo(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaVirtualAllocation VMA_NOT_NULL_NON_DISPATCHABLE allocation, VmaVirtualAllocationInfo* VMA_NOT_NULL pVirtualAllocInfo) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE && pVirtualAllocInfo != VMA_NULL); + VMA_DEBUG_LOG("vmaGetVirtualAllocationInfo"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->GetAllocationInfo(allocation, *pVirtualAllocInfo); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaVirtualAllocate(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + const VmaVirtualAllocationCreateInfo* VMA_NOT_NULL pCreateInfo, VmaVirtualAllocation VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pAllocation, + VkDeviceSize* VMA_NULLABLE pOffset) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE && pCreateInfo != VMA_NULL && pAllocation != VMA_NULL); + VMA_DEBUG_LOG("vmaVirtualAllocate"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + return virtualBlock->Allocate(*pCreateInfo, *pAllocation, pOffset); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaVirtualFree(VmaVirtualBlock VMA_NOT_NULL virtualBlock, VmaVirtualAllocation VMA_NULLABLE_NON_DISPATCHABLE allocation) +{ + if(allocation != VK_NULL_HANDLE) + { + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE); + VMA_DEBUG_LOG("vmaVirtualFree"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->Free(allocation); + } +} + +VMA_CALL_PRE void VMA_CALL_POST vmaClearVirtualBlock(VmaVirtualBlock VMA_NOT_NULL virtualBlock) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE); + VMA_DEBUG_LOG("vmaClearVirtualBlock"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->Clear(); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaSetVirtualAllocationUserData(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaVirtualAllocation VMA_NOT_NULL_NON_DISPATCHABLE allocation, void* VMA_NULLABLE pUserData) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE); + VMA_DEBUG_LOG("vmaSetVirtualAllocationUserData"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->SetAllocationUserData(allocation, pUserData); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetVirtualBlockStatistics(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaStatistics* VMA_NOT_NULL pStats) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE && pStats != VMA_NULL); + VMA_DEBUG_LOG("vmaGetVirtualBlockStatistics"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->GetStatistics(*pStats); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaCalculateVirtualBlockStatistics(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + VmaDetailedStatistics* VMA_NOT_NULL pStats) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE && pStats != VMA_NULL); + VMA_DEBUG_LOG("vmaCalculateVirtualBlockStatistics"); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + virtualBlock->CalculateDetailedStatistics(*pStats); +} + +#if VMA_STATS_STRING_ENABLED + +VMA_CALL_PRE void VMA_CALL_POST vmaBuildVirtualBlockStatsString(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + char* VMA_NULLABLE * VMA_NOT_NULL ppStatsString, VkBool32 detailedMap) +{ + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE && ppStatsString != VMA_NULL); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + const VkAllocationCallbacks* allocationCallbacks = virtualBlock->GetAllocationCallbacks(); + VmaStringBuilder sb(allocationCallbacks); + virtualBlock->BuildStatsString(detailedMap != VK_FALSE, sb); + *ppStatsString = VmaCreateStringCopy(allocationCallbacks, sb.GetData(), sb.GetLength()); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaFreeVirtualBlockStatsString(VmaVirtualBlock VMA_NOT_NULL virtualBlock, + char* VMA_NULLABLE pStatsString) +{ + if(pStatsString != VMA_NULL) + { + VMA_ASSERT(virtualBlock != VK_NULL_HANDLE); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + VmaFreeString(virtualBlock->GetAllocationCallbacks(), pStatsString); + } +} +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_PUBLIC_INTERFACE +#endif // VMA_IMPLEMENTATION /** \page quick_start Quick start @@ -17925,24 +17556,16 @@ To do it properly: 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 ``, which in turn +This library includes header ``, which in turn includes `` 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. -You may need to configure the way you import Vulkan functions. - -- By default, VMA assumes you you link statically with Vulkan API. If this is not the case, - `#define VMA_STATIC_VULKAN_FUNCTIONS 0` before `#include` of the VMA implementation and use another way. -- You can `#define VMA_DYNAMIC_VULKAN_FUNCTIONS 1` and make sure `vkGetInstanceProcAddr` and `vkGetDeviceProcAddr` globals are defined. - All the remaining Vulkan functions will be fetched automatically. -- Finally, you can provide your own pointers to all Vulkan functions needed by VMA using structure member - VmaAllocatorCreateInfo::pVulkanFunctions, if you fetched them in some custom way e.g. using some loader like [Volk](https://github.com/zeux/volk). +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. +Some features of C++14 used. STL containers, RTTI, or C++ exceptions are not used. \section quick_start_initialization Initialization @@ -17953,23 +17576,44 @@ At program startup: -# Fill VmaAllocatorCreateInfo structure and create #VmaAllocator object by calling vmaCreateAllocator(). -\code -VmaAllocatorCreateInfo allocatorInfo = {}; -allocatorInfo.vulkanApiVersion = VK_API_VERSION_1_2; -allocatorInfo.physicalDevice = physicalDevice; -allocatorInfo.device = device; -allocatorInfo.instance = instance; - -VmaAllocator allocator; -vmaCreateAllocator(&allocatorInfo, &allocator); -\endcode - Only members `physicalDevice`, `device`, `instance` are required. However, you should inform the library which Vulkan version do you use by setting VmaAllocatorCreateInfo::vulkanApiVersion and which extensions did you enable by setting VmaAllocatorCreateInfo::flags (like #VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT for VK_KHR_buffer_device_address). Otherwise, VMA would use only features of Vulkan 1.0 core with no extensions. +You may need to configure importing Vulkan functions. There are 3 ways to do this: + +-# **If you link with Vulkan static library** (e.g. "vulkan-1.lib" on Windows): + - You don't need to do anything. + - VMA will use these, as macro `VMA_STATIC_VULKAN_FUNCTIONS` is defined to 1 by default. +-# **If you want VMA to fetch pointers to Vulkan functions dynamically** using `vkGetInstanceProcAddr`, + `vkGetDeviceProcAddr` (this is the option presented in the example below): + - Define `VMA_STATIC_VULKAN_FUNCTIONS` to 0, `VMA_DYNAMIC_VULKAN_FUNCTIONS` to 1. + - Provide pointers to these two functions via VmaVulkanFunctions::vkGetInstanceProcAddr, + VmaVulkanFunctions::vkGetDeviceProcAddr. + - The library will fetch pointers to all other functions it needs internally. +-# **If you fetch pointers to all Vulkan functions in a custom way**, e.g. using some loader like + [Volk](https://github.com/zeux/volk): + - Define `VMA_STATIC_VULKAN_FUNCTIONS` and `VMA_DYNAMIC_VULKAN_FUNCTIONS` to 0. + - Pass these pointers via structure #VmaVulkanFunctions. + +\code +VmaVulkanFunctions vulkanFunctions = {}; +vulkanFunctions.vkGetInstanceProcAddr = &vkGetInstanceProcAddr; +vulkanFunctions.vkGetDeviceProcAddr = &vkGetDeviceProcAddr; + +VmaAllocatorCreateInfo allocatorCreateInfo = {}; +allocatorCreateInfo.vulkanApiVersion = VK_API_VERSION_1_2; +allocatorCreateInfo.physicalDevice = physicalDevice; +allocatorCreateInfo.device = device; +allocatorCreateInfo.instance = instance; +allocatorCreateInfo.pVulkanFunctions = &vulkanFunctions; + +VmaAllocator allocator; +vmaCreateAllocator(&allocatorCreateInfo, &allocator); +\endcode + \section quick_start_resource_allocation Resource allocation @@ -17978,7 +17622,7 @@ 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. + already allocated and bound to it, plus #VmaAllocation objects that represents its underlying memory. \code VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; @@ -17986,7 +17630,7 @@ 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; +allocInfo.usage = VMA_MEMORY_USAGE_AUTO; VkBuffer buffer; VmaAllocation allocation; @@ -18010,8 +17654,8 @@ 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(), vmaFindMemoryTypeIndexForBufferInfo(), - vmaFindMemoryTypeIndexForImageInfo(). + can use function: vmaFindMemoryTypeIndexForBufferInfo(), + vmaFindMemoryTypeIndexForImageInfo(), 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. @@ -18022,9 +17666,10 @@ You can also combine multiple methods. vmaAllocateMemoryForBuffer(), vmaAllocateMemoryForImage(). For binding you should use functions: vmaBindBufferMemory(), vmaBindImageMemory() or their extended versions: vmaBindBufferMemory2(), vmaBindImageMemory2(). --# If you want to create a buffer or an image, allocate memory for it and bind +-# **This is the easiest and recommended way to use this library:** + 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 easiest and recommended way to use this library. + vmaCreateImage(). When using 3. or 4., the library internally queries Vulkan for memory types supported for that buffer or image (function `vkGetBufferMemoryRequirements()`) @@ -18042,11 +17687,12 @@ It is valid, although not very useful. 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. +Since version 3 of the library, it is recommended to use #VMA_MEMORY_USAGE_AUTO to let it select best memory type for your resource automatically. 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: +transfer only once or infrequently and then used for rendering every frame as a uniform buffer, you can +do it using following code. The buffer will most likely end up in a memory type with +`VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT` to be fast to access by the GPU device. \code VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; @@ -18054,13 +17700,56 @@ 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; +allocInfo.usage = VMA_MEMORY_USAGE_AUTO; VkBuffer buffer; VmaAllocation allocation; vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); \endcode +If you have a preference for putting the resource in GPU (device) memory or CPU (host) memory +on systems with discrete graphics card that have the memories separate, you can use +#VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE or #VMA_MEMORY_USAGE_AUTO_PREFER_HOST. + +When using `VMA_MEMORY_USAGE_AUTO*` while you want to map the allocated memory, +you also need to specify one of the host access flags: +#VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT. +This will help the library decide about preferred memory type to ensure it has `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` +so you can map it. + +For example, a staging buffer that will be filled via mapped pointer and then +used as a source of transfer to the buffer decribed previously can be created like this. +It will likely and up in a memory type that is `HOST_VISIBLE` and `HOST_COHERENT` +but not `HOST_CACHED` (meaning uncached, write-combined) and not `DEVICE_LOCAL` (meaning system RAM). + +\code +VkBufferCreateInfo stagingBufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +stagingBufferInfo.size = 65536; +stagingBufferInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + +VmaAllocationCreateInfo stagingAllocInfo = {}; +stagingAllocInfo.usage = VMA_MEMORY_USAGE_AUTO; +stagingAllocInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT; + +VkBuffer stagingBuffer; +VmaAllocation stagingAllocation; +vmaCreateBuffer(allocator, &stagingBufferInfo, &stagingAllocInfo, &stagingBuffer, &stagingAllocation, nullptr); +\endcode + +For more examples of creating different kinds of resources, see chapter \ref usage_patterns. + +Usage values `VMA_MEMORY_USAGE_AUTO*` are legal to use only when the library knows +about the resource being created by having `VkBufferCreateInfo` / `VkImageCreateInfo` passed, +so they work with functions like: vmaCreateBuffer(), vmaCreateImage(), vmaFindMemoryTypeIndexForBufferInfo() etc. +If you allocate raw memory using function vmaAllocateMemory(), you have to use other means of selecting +memory type, as decribed below. + +\note +Old usage values (`VMA_MEMORY_USAGE_GPU_ONLY`, `VMA_MEMORY_USAGE_CPU_ONLY`, +`VMA_MEMORY_USAGE_CPU_TO_GPU`, `VMA_MEMORY_USAGE_GPU_TO_CPU`, `VMA_MEMORY_USAGE_CPU_COPY`) +are still available and work same way as in previous versions of the library +for backward compatibility, but they are not recommended. + \section choosing_memory_type_required_preferred_flags Required and preferred flags You can specify more detailed requirements by filling members @@ -18074,7 +17763,7 @@ use following 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; +allocInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT | VMA_ALLOCATION_CREATE_MAPPED_BIT; VkBuffer buffer; VmaAllocation allocation; @@ -18084,8 +17773,8 @@ vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullpt 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. +Value passed in VmaAllocationCreateInfo::usage is internally converted to a set of required and preferred flags, +plus some extra "magic" (heuristics). \section choosing_memory_type_explicit_memory_types Explicit memory types @@ -18152,6 +17841,13 @@ Mapping the same `VkDeviceMemory` block multiple times is illegal - only one map This includes mapping disjoint regions. Mapping is not reference-counted internally by Vulkan. Because of this, Vulkan Memory Allocator provides following facilities: +\note If you want to be able to map an allocation, you need to specify one of the flags +#VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT +in VmaAllocationCreateInfo::flags. These flags are required for an allocation to be mappable +when using #VMA_MEMORY_USAGE_AUTO or other `VMA_MEMORY_USAGE_AUTO*` enum values. +For other usage values they are ignored and every such allocation made in `HOST_VISIBLE` memory type is mappable, +but they can still be used for consistency. + \section memory_mapping_mapping_functions Mapping functions The library provides following functions for mapping of a specific #VmaAllocation: vmaMapMemory(), vmaUnmapMemory(). @@ -18164,16 +17860,15 @@ Example: \code // Having these objects initialized: - struct ConstantBuffer { ... }; -ConstantBuffer constantBufferData; +ConstantBuffer constantBufferData = ... -VmaAllocator allocator; -VkBuffer constantBuffer; -VmaAllocation constantBufferAllocation; +VmaAllocator allocator = ... +VkBuffer constantBuffer = ... +VmaAllocation constantBufferAllocation = ... // You can map and fill your buffer using following code: @@ -18210,8 +17905,9 @@ 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; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; VkBuffer buf; VmaAllocation alloc; @@ -18222,18 +17918,12 @@ vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allo 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. -- Keeping many large memory blocks mapped may impact performance or stability of some debugging tools. +\note #VMA_ALLOCATION_CREATE_MAPPED_BIT by itself doesn't guarantee that the allocation will end up +in a mappable memory type. +For this, you need to also specify #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or +#VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT. +#VMA_ALLOCATION_CREATE_MAPPED_BIT only guarantees that if the memory is `HOST_VISIBLE`, the allocation will be mapped on creation. +For an example of how to make use of this fact, see section \ref usage_patterns_advanced_data_uploading. \section memory_mapping_cache_control Cache flush and invalidate @@ -18254,87 +17944,9 @@ In any memory type that is `HOST_VISIBLE` but not `HOST_COHERENT`, all allocatio 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) +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.pMappedData != nullptr) -{ - // Allocation ended up in mappable memory. - // It is 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 +`HOST_VISIBLE`, so on PC you may not need to bother. \page staying_within_budget Staying within budget @@ -18355,12 +17967,12 @@ operating system: \section staying_within_budget_querying_for_budget Querying for budget -To query for current memory usage and available budget, use function vmaGetBudget(). +To query for current memory usage and available budget, use function vmaGetHeapBudgets(). Returned structure #VmaBudget contains quantities expressed in bytes, per Vulkan memory heap. Please note that this function returns different information and works faster than -vmaCalculateStats(). vmaGetBudget() can be called every frame or even before every -allocation, while vmaCalculateStats() is intended to be used rarely, +vmaCalculateStatistics(). vmaGetHeapBudgets() can be called every frame or even before every +allocation, while vmaCalculateStatistics() is intended to be used rarely, only to obtain statistical information, e.g. for debugging purposes. It is recommended to use VK_EXT_memory_budget device extension to obtain information @@ -18390,22 +18002,27 @@ budget, by default the library still tries to create it, leaving it to the Vulka implementation whether the allocation succeeds or fails. You can change this behavior by using #VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT flag. With it, the allocation is not made if it would exceed the budget or if the budget is already exceeded. -Some other allocations become lost instead to make room for it, if the mechanism of -[lost allocations](@ref lost_allocations) is used. -If that is not possible, the allocation fails with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. +VMA then tries to make the allocation from the next eligible Vulkan memory type. +The all of them fail, the call then fails with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. Example usage pattern may be to pass the #VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT flag when creating resources that are not essential for the application (e.g. the texture of a specific object) and not to pass it when creating critically important resources (e.g. render targets). +On AMD graphics cards there is a custom vendor extension available: VK_AMD_memory_overallocation_behavior +that allows to control the behavior of the Vulkan implementation in out-of-memory cases - +whether it should fail with an error code or still allow the allocation. +Usage of this extension involves only passing extra structure on Vulkan device creation, +so it is out of scope of this library. + Finally, you can also use #VMA_ALLOCATION_CREATE_NEVER_ALLOCATE_BIT flag to make sure a new allocation is created only when it fits inside one of the existing memory blocks. If it would require to allocate a new block, if fails instead with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. This also ensures that the function call is very fast because it never goes to Vulkan to obtain a new block. -Please note that creating \ref custom_memory_pools with VmaPoolCreateInfo::minBlockCount -set to more than 0 will try to allocate memory blocks without checking whether they +\note Creating \ref custom_memory_pools with VmaPoolCreateInfo::minBlockCount +set to more than 0 will currently try to allocate memory blocks without checking whether they fit within budget. @@ -18479,7 +18096,7 @@ finalMemReq.memoryTypeBits = img1MemReq.memoryTypeBits & img2MemReq.memoryTypeBi // Validate if(finalMemReq.memoryTypeBits != 0) VmaAllocationCreateInfo allocCreateInfo = {}; -allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; VmaAllocation alloc; res = vmaAllocateMemory(allocator, &finalMemReq, &allocCreateInfo, &alloc, nullptr); @@ -18508,7 +18125,7 @@ See chapter 11.8. "Memory Aliasing" of Vulkan specification or `VK_IMAGE_CREATE_ - You can create more complex layout where different images and buffers are bound at different offsets inside one large allocation. For example, one can imagine a big texture used in some render passes, aliasing with a set of many small buffers -used between in some further passes. To bind a resource at non-zero offset of an allocation, +used between in some further passes. To bind a resource at non-zero offset in an allocation, use vmaBindBufferMemory2() / vmaBindImageMemory2(). - Before allocating memory for the resources you want to alias, check `memoryTypeBits` returned in memory requirements of each resource to make sure the bits overlap. @@ -18533,6 +18150,7 @@ It can be useful if you want to: - Reserve minimum or fixed amount of Vulkan memory always preallocated for that pool. - Use extra parameters for a set of your allocations that are available in #VmaPoolCreateInfo but not in #VmaAllocationCreateInfo - e.g., custom minimum alignment, custom `pNext` chain. +- Perform defragmentation on a specific subset of your allocations. To use custom memory pools: @@ -18544,14 +18162,28 @@ To use custom memory pools: Example: \code +// Find memoryTypeIndex for the pool. +VkBufferCreateInfo sampleBufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +sampleBufCreateInfo.size = 0x10000; // Doesn't matter. +sampleBufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo sampleAllocCreateInfo = {}; +sampleAllocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; + +uint32_t memTypeIndex; +VkResult res = vmaFindMemoryTypeIndexForBufferInfo(allocator, + &sampleBufCreateInfo, &sampleAllocCreateInfo, &memTypeIndex); +// Check res... + // Create a pool that can have at most 2 blocks, 128 MiB each. VmaPoolCreateInfo poolCreateInfo = {}; -poolCreateInfo.memoryTypeIndex = ... +poolCreateInfo.memoryTypeIndex = memTypeIndex; poolCreateInfo.blockSize = 128ull * 1024 * 1024; poolCreateInfo.maxBlockCount = 2; VmaPool pool; -vmaCreatePool(allocator, &poolCreateInfo, &pool); +res = vmaCreatePool(allocator, &poolCreateInfo, &pool); +// Check res... // Allocate a buffer out of it. VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; @@ -18563,8 +18195,8 @@ allocCreateInfo.pool = pool; VkBuffer buf; VmaAllocation alloc; -VmaAllocationInfo allocInfo; -vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); +res = vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, nullptr); +// Check res... \endcode You have to free all allocations made from this pool before destroying it. @@ -18574,6 +18206,19 @@ vmaDestroyBuffer(allocator, buf, alloc); vmaDestroyPool(allocator, pool); \endcode +New versions of this library support creating dedicated allocations in custom pools. +It is supported only when VmaPoolCreateInfo::blockSize = 0. +To use this feature, set VmaAllocationCreateInfo::pool to the pointer to your custom pool and +VmaAllocationCreateInfo::flags to #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. + +\note Excessive use of custom pools is a common mistake when using this library. +Custom pools may be useful for special purposes - when you want to +keep certain type of resources separate e.g. to reserve minimum amount of memory +for them or limit maximum amount of memory they can occupy. For most +resources this is not needed and so it is not recommended to create #VmaPool +objects and allocations out of them. Allocating from the default pool is sufficient. + + \section custom_memory_pools_MemTypeIndex Choosing memory type index When creating a pool, you must explicitly specify memory type index. @@ -18584,11 +18229,11 @@ 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. +exampleBufCreateInfo.size = 1024; // Doesn't matter +exampleBufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; VmaAllocationCreateInfo allocCreateInfo = {}; -allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; // Change if needed. +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; uint32_t memTypeIndex; vmaFindMemoryTypeIndexForBufferInfo(allocator, &exampleBufCreateInfo, &allocCreateInfo, &memTypeIndex); @@ -18628,6 +18273,7 @@ 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. +You don't need to specify explicitly which of these options you are going to use - it is detected automatically. \subsection linear_algorithm_free_at_once Free-at-once @@ -18663,7 +18309,7 @@ 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 make allocation from the upper stack, add flag #VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT to VmaAllocationCreateInfo::flags. ![Double stack](../gfx/Linear_allocator_7_double_stack.png) @@ -18685,52 +18331,11 @@ you can achieve behavior of a ring buffer / queue. ![Ring buffer](../gfx/Linear_allocator_5_ring_buffer.png) -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 with lost allocations](../gfx/Linear_allocator_6_ring_buffer_lost.png) - 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 +\note \ref defragmentation is not supported in custom pools created with #VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT. -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 is 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. - -![Buddy allocator](../gfx/Buddy_allocator.png) - -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 @@ -18740,354 +18345,185 @@ 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()`, vmaBindBufferMemory(), vmaBindImageMemory() - 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 buffers; -std::vector allocations; - - -const uint32_t allocCount = (uint32_t)allocations.size(); -std::vector 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); - vmaBindBufferMemory(allocator, allocations[i], buffers[i]); - } -} -\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. +To mitigate this problem, you can use defragmentation feature. +It doesn't happen automatically though and needs your cooperation, +because VMA is a low level library that only allocates memory. +It cannot recreate buffers and images in a new place as it doesn't remember the contents of `VkBufferCreateInfo` / `VkImageCreateInfo` structures. +It cannot copy their contents as it doesn't record any commands to a command buffer. Example: \code -// Given following variables already initialized: -VkDevice device; -VmaAllocator allocator; -VkCommandBuffer commandBuffer; -std::vector buffers; -std::vector allocations; - - -const uint32_t allocCount = (uint32_t)allocations.size(); -std::vector 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 is "GPU" this time. -defragInfo.maxGpuAllocationsToMove = UINT32_MAX; // Notice it is "GPU" this time. -defragInfo.commandBuffer = commandBuffer; +VmaDefragmentationInfo defragInfo = {}; +defragInfo.pool = myPool; +defragInfo.flags = VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT; VmaDefragmentationContext defragCtx; -vmaDefragmentationBegin(allocator, &defragInfo, nullptr, &defragCtx); +VkResult res = vmaBeginDefragmentation(allocator, &defragInfo, &defragCtx); +// Check res... -vkEndCommandBuffer(commandBuffer); - -// Submit commandBuffer. -// Wait for a fence that ensures commandBuffer execution finished. - -vmaDefragmentationEnd(allocator, defragCtx); - -for(uint32_t i = 0; i < allocCount; ++i) +for(;;) { - if(allocationsChanged[i]) + VmaDefragmentationPassMoveInfo pass; + res = vmaBeginDefragmentationPass(allocator, defragCtx, &pass); + if(res == VK_SUCCESS) + break; + else if(res != VK_INCOMPLETE) + // Handle error... + + for(uint32_t i = 0; i < pass.moveCount; ++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. + // Inspect pass.pMoves[i].srcAllocation, identify what buffer/image it represents. VmaAllocationInfo allocInfo; - vmaGetAllocationInfo(allocator, allocations[i], &allocInfo); - vmaBindBufferMemory(allocator, allocations[i], buffers[i]); + vmaGetAllocationInfo(allocator, pMoves[i].srcAllocation, &allocInfo); + MyEngineResourceData* resData = (MyEngineResourceData*)allocInfo.pUserData; + + // Recreate and bind this buffer/image at: pass.pMoves[i].dstMemory, pass.pMoves[i].dstOffset. + VkImageCreateInfo imgCreateInfo = ... + VkImage newImg; + res = vkCreateImage(device, &imgCreateInfo, nullptr, &newImg); + // Check res... + res = vmaBindImageMemory(allocator, pMoves[i].dstTmpAllocation, newImg); + // Check res... + + // Issue a vkCmdCopyBuffer/vkCmdCopyImage to copy its content to the new place. + vkCmdCopyImage(cmdBuf, resData->img, ..., newImg, ...); } -} -\endcode + + // Make sure the copy commands finished executing. + vkWaitForFences(...); -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 - -It is only legal to defragment allocations bound to: - -- buffers -- images created with `VK_IMAGE_CREATE_ALIAS_BIT`, `VK_IMAGE_TILING_LINEAR`, and - being currently in `VK_IMAGE_LAYOUT_GENERAL` or `VK_IMAGE_LAYOUT_PREINITIALIZED`. - -Defragmentation of images created with `VK_IMAGE_TILING_OPTIMAL` or in any other -layout may give undefined results. - -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 if -needed using an image memory barrier. - -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). - -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 maximize 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 is 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 is 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. - -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 is used in the -current frame? - -It is ensured because vmaTouchAllocation() / vmaGetAllocationInfo() not only returns allocation -status/parameters and checks whether it is not lost, but when it is 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. - -Q: What if my allocation may still be in use by the GPU when it is rendering a -previous frame while I already submit new frame on the CPU? - -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). - -Q: How do you inform the library when new frame starts? - -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) + // Destroy old buffers/images bound with pass.pMoves[i].srcAllocation. + for(uint32_t i = 0; i < pass.moveCount; ++i) { - // Check if its allocation is not lost + mark it as used in current frame. - if(vmaTouchAllocation(allocator, m_Alloc)) - { - // It is all OK - safe to use m_Buf. - return; - } + // ... + vkDestroyImage(device, resData->img, nullptr); } - // 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); + // Update appropriate descriptors to point to the new places... + + res = vmaEndDefragmentationPass(allocator, defragCtx, &pass); + if(res == VK_SUCCESS) + break; + else if(res != VK_INCOMPLETE) + // Handle error... } + +vmaEndDefragmentation(allocator, defragCtx, 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. +Although functions like vmaCreateBuffer(), vmaCreateImage(), vmaDestroyBuffer(), vmaDestroyImage() +create/destroy an allocation and a buffer/image at once, these are just a shortcut for +creating the resource, allocating memory, and binding them together. +Defragmentation works on memory allocations only. You must handle the rest manually. +Defragmentation is an iterative process that should repreat "passes" as long as related functions +return `VK_INCOMPLETE` not `VK_SUCCESS`. +In each pass: -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. +1. vmaBeginDefragmentationPass() function call: + - Calculates and returns the list of allocations to be moved in this pass. + Note this can be a time-consuming process. + - Reserves destination memory for them by creating temporary destination allocations + that you can query for their `VkDeviceMemory` + offset using vmaGetAllocationInfo(). +2. Inside the pass, **you should**: + - Inspect the returned list of allocations to be moved. + - Create new buffers/images and bind them at the returned destination temporary allocations. + - Copy data from source to destination resources if necessary. + - Destroy the source buffers/images, but NOT their allocations. +3. vmaEndDefragmentationPass() function call: + - Frees the source memory reserved for the allocations that are moved. + - Modifies source #VmaAllocation objects that are moved to point to the destination reserved memory. + - Frees `VkDeviceMemory` blocks that became empty. -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. +Unlike in previous iterations of the defragmentation API, there is no list of "movable" allocations passed as a parameter. +Defragmentation algorithm tries to move all suitable allocations. +You can, however, refuse to move some of them inside a defragmentation pass, by setting +`pass.pMoves[i].operation` to #VMA_DEFRAGMENTATION_MOVE_OPERATION_IGNORE. +This is not recommended and may result in suboptimal packing of the allocations after defragmentation. +If you cannot ensure any allocation can be moved, it is better to keep movable allocations separate in a custom pool. -Q: Can I touch allocation that cannot become lost? +Inside a pass, for each allocation that should be moved: -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. +- You should copy its data from the source to the destination place by calling e.g. `vkCmdCopyBuffer()`, `vkCmdCopyImage()`. + - You need to make sure these commands finished executing before destroying the source buffers/images and before calling vmaEndDefragmentationPass(). +- If a resource doesn't contain any meaningful data, e.g. it is a transient color attachment image to be cleared, + filled, and used temporarily in each rendering frame, you can just recreate this image + without copying its data. +- If the resource is in `HOST_VISIBLE` and `HOST_CACHED` memory, you can copy its data on the CPU + using `memcpy()`. +- If you cannot move the allocation, you can set `pass.pMoves[i].operation` to #VMA_DEFRAGMENTATION_MOVE_OPERATION_IGNORE. + This will cancel the move. + - vmaEndDefragmentationPass() will then free the destination memory + not the source memory of the allocation, leaving it unchanged. +- If you decide the allocation is unimportant and can be destroyed instead of moved (e.g. it wasn't used for long time), + you can set `pass.pMoves[i].operation` to #VMA_DEFRAGMENTATION_MOVE_OPERATION_DESTROY. + - vmaEndDefragmentationPass() will then free both source and destination memory, and will destroy the source #VmaAllocation object. + +You can defragment a specific custom pool by setting VmaDefragmentationInfo::pool +(like in the example above) or all the default pools by setting this member to null. + +Defragmentation is always performed in each pool separately. +Allocations are never moved between different Vulkan memory types. +The size of the destination memory reserved for a moved allocation is the same as the original one. +Alignment of an allocation as it was determined using `vkGetBufferMemoryRequirements()` etc. is also respected after defragmentation. +Buffers/images should be recreated with the same `VkBufferCreateInfo` / `VkImageCreateInfo` parameters as the original ones. + +You can perform the defragmentation incrementally to limit the number of allocations and bytes to be moved +in each pass, e.g. to call it in sync with render frames and not to experience too big hitches. +See members: VmaDefragmentationInfo::maxBytesPerPass, VmaDefragmentationInfo::maxAllocationsPerPass. + +It is also safe to perform the defragmentation asynchronously to render frames and other Vulkan and VMA +usage, possibly from multiple threads, with the exception that allocations +returned in VmaDefragmentationPassMoveInfo::pMoves shouldn't be destroyed until the defragmentation pass is ended. + +Mapping is preserved on allocations that are moved during defragmentation. +Whether through #VMA_ALLOCATION_CREATE_MAPPED_BIT or vmaMapMemory(), the allocations +are mapped at their new place. Of course, pointer to the mapped data changes, so it needs to be queried +using VmaAllocationInfo::pMappedData. + +\note Defragmentation is not supported in custom pools created with #VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT. \page statistics Statistics -This library contains functions that return information about its internal state, +This library contains several 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. +If you need to obtain basic statistics about memory usage per heap, together with current budget, +you can call function vmaGetHeapBudgets() and inspect structure #VmaBudget. +This is useful to keep track of memory usage and stay withing budget +(see also \ref staying_within_budget). +Example: -You can query for statistics of a custom pool using function vmaGetPoolStats(). -Information are returned using structure #VmaPoolStats. +\code +uint32_t heapIndex = ... -You can query for information about specific allocation using function vmaGetAllocationInfo(). +VmaBudget budgets[VK_MAX_MEMORY_HEAPS]; +vmaGetHeapBudgets(allocator, budgets); + +printf("My heap currently has %u allocations taking %llu B,\n", + budgets[heapIndex].statistics.allocationCount, + budgets[heapIndex].statistics.allocationBytes); +printf("allocated out of %u Vulkan device memory blocks taking %llu B,\n", + budgets[heapIndex].statistics.blockCount, + budgets[heapIndex].statistics.blockBytes); +printf("Vulkan reports total usage %llu B with budget %llu B.\n", + budgets[heapIndex].usage, + budgets[heapIndex].budget); +\endcode + +You can query for more detailed statistics per memory heap, type, and totals, +including minimum and maximum allocation size and unused range size, +by calling function vmaCalculateStatistics() and inspecting structure #VmaTotalStatistics. +This function is slower though, as it has to traverse all the internal data structures, +so it should be used only for debugging purposes. + +You can query for statistics of a custom pool using function vmaGetPoolStatistics() +or vmaCalculatePoolStatistics(). + +You can query for information about a specific allocation using function vmaGetAllocationInfo(). It fill structure #VmaAllocationInfo. \section statistics_json_dump JSON dump @@ -19104,7 +18540,7 @@ The format of this JSON string is not part of official documentation of the libr 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(). +The JSON string contains all the data that can be obtained using vmaCalculateStatistics(). 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. @@ -19119,20 +18555,21 @@ To do that, fill VmaAllocationCreateInfo::pUserData field when creating an allocation. It is 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. +It it useful to identify appropriate data structures in your engine given #VmaAllocation, +e.g. when doing \ref defragmentation. \code -VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; -// Fill bufferInfo... +VkBufferCreateInfo bufCreateInfo = ... MyBufferMetadata* pMetadata = CreateBufferMetadata(); VmaAllocationCreateInfo allocCreateInfo = {}; -allocCreateInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; allocCreateInfo.pUserData = pMetadata; VkBuffer buffer; VmaAllocation allocation; -vmaCreateBuffer(allocator, &bufferInfo, &allocCreateInfo, &buffer, &allocation, nullptr); +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buffer, &allocation, nullptr); \endcode The pointer may be later retrieved as VmaAllocationInfo::pUserData: @@ -19146,50 +18583,190 @@ MyBufferMetadata* pMetadata = (MyBufferMetadata*)allocInfo.pUserData; 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. +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. +An allocation can also carry a null-terminated string, giving a name to the allocation. +To set it, call vmaSetAllocationName(). 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); +vmaSetAllocationName(allocator, allocation, imageName.c_str()); \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. +The string can be later retrieved by inspecting VmaAllocationInfo::pName. +It is also printed in JSON report created by vmaBuildStatsString(). + +\note Setting string name to VMA allocation doesn't automatically set it to the Vulkan buffer or image created with it. +You must do it manually using an extension like VK_EXT_debug_utils, which is independent of this library. + + +\page virtual_allocator Virtual allocator + +As an extra feature, the core allocation algorithm of the library is exposed through a simple and convenient API of "virtual allocator". +It doesn't allocate any real GPU memory. It just keeps track of used and free regions of a "virtual block". +You can use it to allocate your own memory or other objects, even completely unrelated to Vulkan. +A common use case is sub-allocation of pieces of one large GPU buffer. + +\section virtual_allocator_creating_virtual_block Creating virtual block + +To use this functionality, there is no main "allocator" object. +You don't need to have #VmaAllocator object created. +All you need to do is to create a separate #VmaVirtualBlock object for each block of memory you want to be managed by the allocator: + +-# Fill in #VmaVirtualBlockCreateInfo structure. +-# Call vmaCreateVirtualBlock(). Get new #VmaVirtualBlock object. + +Example: \code -VmaAllocationInfo allocInfo; -vmaGetAllocationInfo(allocator, allocation, &allocInfo); -const char* imageName = (const char*)allocInfo.pUserData; -printf("Image name: %s\n", imageName); +VmaVirtualBlockCreateInfo blockCreateInfo = {}; +blockCreateInfo.size = 1048576; // 1 MB + +VmaVirtualBlock block; +VkResult res = vmaCreateVirtualBlock(&blockCreateInfo, &block); \endcode -That string is also printed in JSON report created by vmaBuildStatsString(). +\section virtual_allocator_making_virtual_allocations Making virtual allocations -\note Passing string name to VMA allocation doesn't automatically set it to the Vulkan buffer or image created with it. -You must do it manually using an extension like VK_EXT_debug_utils, which is independent of this library. +#VmaVirtualBlock object contains internal data structure that keeps track of free and occupied regions +using the same code as the main Vulkan memory allocator. +Similarly to #VmaAllocation for standard GPU allocations, there is #VmaVirtualAllocation type +that represents an opaque handle to an allocation withing the virtual block. + +In order to make such allocation: + +-# Fill in #VmaVirtualAllocationCreateInfo structure. +-# Call vmaVirtualAllocate(). Get new #VmaVirtualAllocation object that represents the allocation. + You can also receive `VkDeviceSize offset` that was assigned to the allocation. + +Example: + +\code +VmaVirtualAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.size = 4096; // 4 KB + +VmaVirtualAllocation alloc; +VkDeviceSize offset; +res = vmaVirtualAllocate(block, &allocCreateInfo, &alloc, &offset); +if(res == VK_SUCCESS) +{ + // Use the 4 KB of your memory starting at offset. +} +else +{ + // Allocation failed - no space for it could be found. Handle this error! +} +\endcode + +\section virtual_allocator_deallocation Deallocation + +When no longer needed, an allocation can be freed by calling vmaVirtualFree(). +You can only pass to this function an allocation that was previously returned by vmaVirtualAllocate() +called for the same #VmaVirtualBlock. + +When whole block is no longer needed, the block object can be released by calling vmaDestroyVirtualBlock(). +All allocations must be freed before the block is destroyed, which is checked internally by an assert. +However, if you don't want to call vmaVirtualFree() for each allocation, you can use vmaClearVirtualBlock() to free them all at once - +a feature not available in normal Vulkan memory allocator. Example: + +\code +vmaVirtualFree(block, alloc); +vmaDestroyVirtualBlock(block); +\endcode + +\section virtual_allocator_allocation_parameters Allocation parameters + +You can attach a custom pointer to each allocation by using vmaSetVirtualAllocationUserData(). +Its default value is null. +It can be used to store any data that needs to be associated with that allocation - e.g. an index, a handle, or a pointer to some +larger data structure containing more information. Example: + +\code +struct CustomAllocData +{ + std::string m_AllocName; +}; +CustomAllocData* allocData = new CustomAllocData(); +allocData->m_AllocName = "My allocation 1"; +vmaSetVirtualAllocationUserData(block, alloc, allocData); +\endcode + +The pointer can later be fetched, along with allocation offset and size, by passing the allocation handle to function +vmaGetVirtualAllocationInfo() and inspecting returned structure #VmaVirtualAllocationInfo. +If you allocated a new object to be used as the custom pointer, don't forget to delete that object before freeing the allocation! +Example: + +\code +VmaVirtualAllocationInfo allocInfo; +vmaGetVirtualAllocationInfo(block, alloc, &allocInfo); +delete (CustomAllocData*)allocInfo.pUserData; + +vmaVirtualFree(block, alloc); +\endcode + +\section virtual_allocator_alignment_and_units Alignment and units + +It feels natural to express sizes and offsets in bytes. +If an offset of an allocation needs to be aligned to a multiply of some number (e.g. 4 bytes), you can fill optional member +VmaVirtualAllocationCreateInfo::alignment to request it. Example: + +\code +VmaVirtualAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.size = 4096; // 4 KB +allocCreateInfo.alignment = 4; // Returned offset must be a multiply of 4 B + +VmaVirtualAllocation alloc; +res = vmaVirtualAllocate(block, &allocCreateInfo, &alloc, nullptr); +\endcode + +Alignments of different allocations made from one block may vary. +However, if all alignments and sizes are always multiply of some size e.g. 4 B or `sizeof(MyDataStruct)`, +you can express all sizes, alignments, and offsets in multiples of that size instead of individual bytes. +It might be more convenient, but you need to make sure to use this new unit consistently in all the places: + +- VmaVirtualBlockCreateInfo::size +- VmaVirtualAllocationCreateInfo::size and VmaVirtualAllocationCreateInfo::alignment +- Using offset returned by vmaVirtualAllocate() or in VmaVirtualAllocationInfo::offset + +\section virtual_allocator_statistics Statistics + +You can obtain statistics of a virtual block using vmaGetVirtualBlockStatistics() +(to get brief statistics that are fast to calculate) +or vmaCalculateVirtualBlockStatistics() (to get more detailed statistics, slower to calculate). +The functions fill structures #VmaStatistics, #VmaDetailedStatistics respectively - same as used by the normal Vulkan memory allocator. +Example: + +\code +VmaStatistics stats; +vmaGetVirtualBlockStatistics(block, &stats); +printf("My virtual block has %llu bytes used by %u virtual allocations\n", + stats.allocationBytes, stats.allocationCount); +\endcode + +You can also request a full list of allocations and free regions as a string in JSON format by calling +vmaBuildVirtualBlockStatsString(). +Returned string must be later freed using vmaFreeVirtualBlockStatsString(). +The format of this string differs from the one returned by the main Vulkan allocator, but it is similar. + +\section virtual_allocator_additional_considerations Additional considerations + +The "virtual allocator" functionality is implemented on a level of individual memory blocks. +Keeping track of a whole collection of blocks, allocating new ones when out of free space, +deleting empty ones, and deciding which one to try first for a new allocation must be implemented by the user. + +Alternative allocation algorithms are supported, just like in custom pools of the real GPU memory. +See enum #VmaVirtualBlockCreateFlagBits to learn how to specify them (e.g. #VMA_VIRTUAL_BLOCK_CREATE_LINEAR_ALGORITHM_BIT). +You can find their description in chapter \ref custom_memory_pools. +Allocation strategies are also supported. +See enum #VmaVirtualAllocationCreateFlagBits to learn how to specify them (e.g. #VMA_VIRTUAL_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT). + +Following features are supported only by the allocator of the real GPU memory and not by virtual allocations: +buffer-image granularity, `VMA_DEBUG_MARGIN`, `VMA_MIN_ALIGNMENT`. \page debugging_memory_usage Debugging incorrect memory usage @@ -19209,17 +18786,15 @@ To do it, define macro `VMA_DEBUG_INITIALIZE_ALLOCATIONS` to 1. #include "vk_mem_alloc.h" \endcode -It makes memory of all new allocations initialized to bit pattern `0xDCDCDCDC`. +It makes memory of 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`. +Memory initialization works only with memory types that are `HOST_VISIBLE` and with allocations that can be mapped. 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 @@ -19229,7 +18804,7 @@ By default, allocations are laid out in memory blocks next to each other if poss ![Allocations without margin](../gfx/Margins_1.png) 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. +number of bytes as a margin after every allocation. \code #define VMA_DEBUG_MARGIN 16 @@ -19243,9 +18818,6 @@ being overwritten outside of allocation boundaries. It is not 100% certain thoug 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 @@ -19253,12 +18825,13 @@ 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. +Margins do not apply to \ref virtual_allocator. + \section debugging_memory_usage_corruption_detection Corruption detection You can additionally define macro `VMA_DEBUG_DETECT_CORRUPTION` to 1 to enable validation @@ -19271,7 +18844,7 @@ of contents of the margins. \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. +(it must be multiply of 4) after every allocation is filled with a magic number. This idea is also know as "canary". Memory is automatically mapped and unmapped if necessary. @@ -19289,58 +18862,6 @@ 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 - -Recording functionality is disabled by default. -To enable it, define following macro before every include of this library: - -\code -#define VMA_RECORDING_ENABLED 1 -\endcode - -To record sequence of calls to a file: Fill in -VmaAllocatorCreateInfo::pRecordSettings member while creating #VmaAllocator -object. File is opened and written during whole lifetime of the allocator. - -To replay file: 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 - -Documentation of file format can be found in file: "docs/Recording file format.md". -It is 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. - - \page opengl_interop OpenGL Interop VMA provides some features that help with interoperability with OpenGL. @@ -19361,8 +18882,7 @@ also dedicated allocations or other allocations made from default pools, an alternative solution is to fill in VmaAllocatorCreateInfo::pTypeExternalMemoryHandleTypes. It should point to an array with `VkExternalMemoryHandleTypeFlagsKHR` to be automatically passed by the library through `VkExportMemoryAllocateInfoKHR` on each allocation made from a specific memory type. -This is currently the only method to use if you need exported dedicated allocations, as they cannot be created out of custom pools. -This will change in future versions of the library though. +Please note that new versions of the library also support dedicated allocations created in custom pools. You should not mix these two methods in a way that allows to apply both to the same memory type. Otherwise, `VkExportMemoryAllocateInfoKHR` structure would be attached twice to the `pNext` chain of `VkMemoryAllocateInfo`. @@ -19382,8 +18902,6 @@ from a function like `vkGetBufferMemoryRequirements`, which is called by VMA aut If you want to create a buffer with a specific minimum alignment out of default pools, use special function vmaCreateBufferWithAlignment(), which takes additional parameter `minAlignment`. -This is currently the only method to use if you need exported dedicated allocations, as they cannot be created out of custom pools. -This will change in future versions of the library though. Note the problem of alignment affects only resources placed inside bigger `VkDeviceMemory` blocks and not dedicated allocations, as these, by definition, always have alignment = 0 because the resource is bound to the beginning of its dedicated block. @@ -19392,34 +18910,14 @@ Contrary to Direct3D 12, Vulkan doesn't have a concept of alignment of the entir \page usage_patterns Recommended usage patterns +Vulkan gives great flexibility in memory allocation. +This chapter shows the most common 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_common_mistakes Common mistakes - -Use of CPU_TO_GPU instead of CPU_ONLY memory - -#VMA_MEMORY_USAGE_CPU_TO_GPU is recommended only for resources that will be -mapped and written by the CPU, as well as read directly by the GPU - like some -buffers or textures updated every frame (dynamic). If you create a staging copy -of a resource to be written by CPU and then used as a source of transfer to -another resource placed in the GPU memory, that staging resource should be -created with #VMA_MEMORY_USAGE_CPU_ONLY. Please read the descriptions of these -enums carefully for details. - -Unnecessary use of custom pools - -\ref custom_memory_pools may be useful for special purposes - when you want to -keep certain type of resources separate e.g. to reserve minimum amount of memory -for them, limit maximum amount of memory they can occupy, or make some of them -push out the other through the mechanism of \ref lost_allocations. For most -resources this is not needed and so it is not recommended to create #VmaPool -objects and allocations out of them. Allocating from the default pool is sufficient. - -\section usage_patterns_simple Simple patterns - -\subsection usage_patterns_simple_render_targets Render targets +\section usage_patterns_gpu_only GPU-only resource When: Any resources that you frequently write and read on GPU, @@ -19427,123 +18925,219 @@ e.g. images used as color attachments (aka "render targets"), depth-stencil atta images/buffers used as storage image/buffer (aka "Unordered Access View (UAV)"). What to do: -Create them in video memory that is fastest to access from GPU using -#VMA_MEMORY_USAGE_GPU_ONLY. +Let the library select the optimal memory type, which will likely have `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. -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. +\code +VkImageCreateInfo imgCreateInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO }; +imgCreateInfo.imageType = VK_IMAGE_TYPE_2D; +imgCreateInfo.extent.width = 3840; +imgCreateInfo.extent.height = 2160; +imgCreateInfo.extent.depth = 1; +imgCreateInfo.mipLevels = 1; +imgCreateInfo.arrayLayers = 1; +imgCreateInfo.format = VK_FORMAT_R8G8B8A8_UNORM; +imgCreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL; +imgCreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; +imgCreateInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; +imgCreateInfo.samples = VK_SAMPLE_COUNT_1_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT; +allocCreateInfo.priority = 1.0f; + +VkImage img; +VmaAllocation alloc; +vmaCreateImage(allocator, &imgCreateInfo, &allocCreateInfo, &img, &alloc, nullptr); +\endcode + +Also consider: +Consider 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 with different sizes +e.g. when display resolution changes. Prefer to create such resources first and all other GPU resources (like textures and vertex buffers) later. +When VK_EXT_memory_priority extension is enabled, it is also worth setting high priority to such allocation +to decrease chances to be evicted to system memory by the operating system. -\subsection usage_patterns_simple_immutable_resources Immutable resources +\section usage_patterns_staging_copy_upload Staging copy for upload When: -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. +A "staging" buffer than you want to map and fill from CPU code, then use as a source od transfer +to some GPU resource. What to do: -Create them in video memory that is fastest to access from GPU using -#VMA_MEMORY_USAGE_GPU_ONLY. +Use flag #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT. +Let the library select the optimal memory type, which will always have `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT`. -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. +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 65536; +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; -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`. +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; -\subsection usage_patterns_dynamic_resources Dynamic resources +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +... + +memcpy(allocInfo.pMappedData, myData, myDataSize); +\endcode + +Also consider: +You can map the allocation using vmaMapMemory() or you can create it as persistenly mapped +using #VMA_ALLOCATION_CREATE_MAPPED_BIT, as in the example above. + + +\section usage_patterns_readback Readback When: -Any resources that change frequently (aka "dynamic"), e.g. every frame or every draw call, -written on CPU, read on GPU. - -What to do: -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. -Also note that textures written directly from the host through a mapped pointer need to be in LINEAR not OPTIMAL layout. - -\subsection usage_patterns_readback Readback - -When: -Resources that contain data written by GPU that you want to read back on CPU, +Buffers for data written by or transferred from the GPU that you want to read back on the CPU, e.g. results of some computations. What to do: -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. +Use flag #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT. +Let the library select the optimal memory type, which will always have `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` +and `VK_MEMORY_PROPERTY_HOST_CACHED_BIT`. -\section usage_patterns_advanced Advanced patterns +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 65536; +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT; -\subsection usage_patterns_integrated_graphics Detecting integrated graphics +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; -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`. +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); -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 +const float* downloadedData = (const float*)allocInfo.pMappedData; +\endcode -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 transfer each time. --# Create just a single copy using #VMA_MEMORY_USAGE_CPU_TO_GPU, map it and fill it on CPU, - read it directly on GPU. --# Create just a single copy using #VMA_MEMORY_USAGE_CPU_ONLY, map it and fill it on CPU, - read it directly on GPU. +\section usage_patterns_advanced_data_uploading Advanced data uploading -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: +For resources that you frequently write on CPU via mapped pointer and +freqnently read on GPU e.g. as a uniform buffer (also called "dynamic"), multiple options are possible: -- On integrated graphics use (2) or (3) to avoid unnecessary 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). +-# Easiest solution is to have one copy of the resource in `HOST_VISIBLE` memory, + even if it means system RAM (not `DEVICE_LOCAL`) on systems with a discrete graphics card, + and make the device reach out to that resource directly. + - Reads performed by the device will then go through PCI Express bus. + The performace of this access may be limited, but it may be fine depending on the size + of this resource (whether it is small enough to quickly end up in GPU cache) and the sparsity + of access. +-# On systems with unified memory (e.g. AMD APU or Intel integrated graphics, mobile chips), + a memory type may be available that is both `HOST_VISIBLE` (available for mapping) and `DEVICE_LOCAL` + (fast to access from the GPU). Then, it is likely the best choice for such type of resource. +-# Systems with a discrete graphics card and separate video memory may or may not expose + a memory type that is both `HOST_VISIBLE` and `DEVICE_LOCAL`, also known as Base Address Register (BAR). + If they do, it represents a piece of VRAM (or entire VRAM, if ReBAR is enabled in the motherboard BIOS) + that is available to CPU for mapping. + - Writes performed by the host to that memory go through PCI Express bus. + The performance of these writes may be limited, but it may be fine, especially on PCIe 4.0, + as long as rules of using uncached and write-combined memory are followed - only sequential writes and no reads. +-# Finally, you may need or prefer to create a separate copy of the resource in `DEVICE_LOCAL` memory, + a separate "staging" copy in `HOST_VISIBLE` memory and perform an explicit transfer command between them. -Similarly, for resources that you frequently write on GPU and read on CPU, multiple -solutions are possible: +Thankfully, VMA offers an aid to create and use such resources in the the way optimal +for the current Vulkan device. To help the library make the best choice, +use flag #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT together with +#VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT. +It will then prefer a memory type that is both `DEVICE_LOCAL` and `HOST_VISIBLE` (integrated memory or BAR), +but if no such memory type is available or allocation from it fails +(PC graphics cards have only 256 MB of BAR by default, unless ReBAR is supported and enabled in BIOS), +it will fall back to `DEVICE_LOCAL` memory for fast GPU access. +It is then up to you to detect that the allocation ended up in a memory type that is not `HOST_VISIBLE`, +so you need to create another "staging" allocation and perform explicit transfers. --# 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. +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 65536; +bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | + VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); -You should take some measurements to decide which option is faster in case of your specific -resource. +VkMemoryPropertyFlags memPropFlags; +vmaGetAllocationMemoryProperties(allocator, alloc, &memPropFlags); -Note that textures accessed directly from the host through a mapped pointer need to be in LINEAR layout, -which may slow down their usage on the device. -Textures accessed only by the device and transfer operations can use OPTIMAL layout. +if(memPropFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) +{ + // Allocation ended up in a mappable memory and is already mapped - write to it directly. -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). + // [Executed in runtime]: + memcpy(allocInfo.pMappedData, myData, myDataSize); +} +else +{ + // Allocation ended up in a non-mappable memory - need to transfer. + VkBufferCreateInfo stagingBufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; + stagingBufCreateInfo.size = 65536; + stagingBufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + + VmaAllocationCreateInfo stagingAllocCreateInfo = {}; + stagingAllocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; + stagingAllocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; + + VkBuffer stagingBuf; + VmaAllocation stagingAlloc; + VmaAllocationInfo stagingAllocInfo; + vmaCreateBuffer(allocator, &stagingBufCreateInfo, &stagingAllocCreateInfo, + &stagingBuf, &stagingAlloc, stagingAllocInfo); + + // [Executed in runtime]: + memcpy(stagingAllocInfo.pMappedData, myData, myDataSize); + //vkCmdPipelineBarrier: VK_ACCESS_HOST_WRITE_BIT --> VK_ACCESS_TRANSFER_READ_BIT + VkBufferCopy bufCopy = { + 0, // srcOffset + 0, // dstOffset, + myDataSize); // size + vkCmdCopyBuffer(cmdBuf, stagingBuf, buf, 1, &bufCopy); +} +\endcode + +\section usage_patterns_other_use_cases Other use cases + +Here are some other, less obvious use cases and their recommended settings: + +- An image that is used only as transfer source and destination, but it should stay on the device, + as it is used to temporarily store a copy of some texture, e.g. from the current to the next frame, + for temporal antialiasing or other temporal effects. + - Use `VkImageCreateInfo::usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT` + - Use VmaAllocationCreateInfo::usage = #VMA_MEMORY_USAGE_AUTO +- An image that is used only as transfer source and destination, but it should be placed + in the system RAM despite it doesn't need to be mapped, because it serves as a "swap" copy to evict + least recently used textures from VRAM. + - Use `VkImageCreateInfo::usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT` + - Use VmaAllocationCreateInfo::usage = #VMA_MEMORY_USAGE_AUTO_PREFER_HOST, + as VMA needs a hint here to differentiate from the previous case. +- A buffer that you want to map and write from the CPU, directly read from the GPU + (e.g. as a uniform or vertex buffer), but you have a clear preference to place it in device or + host memory due to its large size. + - Use `VkBufferCreateInfo::usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT` + - Use VmaAllocationCreateInfo::usage = #VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE or #VMA_MEMORY_USAGE_AUTO_PREFER_HOST + - Use VmaAllocationCreateInfo::flags = #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT \page configuration Configuration @@ -19582,6 +19176,8 @@ by using a helper library like [volk](https://github.com/zeux/volk). Third, VMA tries to fetch remaining pointers that are still null by calling `vkGetInstanceProcAddr` and `vkGetDeviceProcAddr` on its own. +You need to only fill in VmaVulkanFunctions::vkGetInstanceProcAddr and VmaVulkanFunctions::vkGetDeviceProcAddr. +Other pointers will be fetched automatically. If you want to disable this feature, set configuration macro: `#define VMA_DYNAMIC_VULKAN_FUNCTIONS 0`. Finally, all the function pointers required by the library (considering selected @@ -19606,7 +19202,7 @@ VmaAllocatorCreateInfo::pDeviceMemoryCallbacks. \section heap_memory_limit Device heap memory limit When device memory of certain heap runs out of free space, new allocations may -fail (returning error code) or they may succeed, silently pushing some existing +fail (returning error code) or they may succeed, silently pushing some existing_ memory blocks from GPU VRAM to system RAM (which degrades performance). This behavior is implementation-dependent - it depends on GPU vendor and graphics driver. @@ -19627,10 +19223,14 @@ 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. +to do some internal optimizations. The extension is supported by this library. +It will be used automatically when enabled. -The extension is supported by this library. It will be used automatically when -enabled. To enable it: +It has been promoted to core Vulkan 1.1, so if you use eligible Vulkan version +and inform VMA about it by setting VmaAllocatorCreateInfo::vulkanApiVersion, +you are all set. + +Otherwise, if you want to use it as an extension: 1 . When creating Vulkan device, check if following 2 device extensions are supported (call `vkEnumerateDeviceExtensionProperties()`). @@ -19642,7 +19242,7 @@ If yes, enable them (fill `VkDeviceCreateInfo::ppEnabledExtensionNames`). 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 +your #VmaAllocator to inform the library that you enabled required extensions and you want the library to use them. \code @@ -19657,7 +19257,7 @@ 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. +_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 @@ -19671,6 +19271,86 @@ To learn more about this extension, see: +\page vk_ext_memory_priority VK_EXT_memory_priority + +VK_EXT_memory_priority is a device extension that allows to pass additional "priority" +value to Vulkan memory allocations that the implementation may use prefer certain +buffers and images that are critical for performance to stay in device-local memory +in cases when the memory is over-subscribed, while some others may be moved to the system memory. + +VMA offers convenient usage of this extension. +If you enable it, you can pass "priority" parameter when creating allocations or custom pools +and the library automatically passes the value to Vulkan using this extension. + +If you want to use this extension in connection with VMA, follow these steps: + +\section vk_ext_memory_priority_initialization Initialization + +1) Call `vkEnumerateDeviceExtensionProperties` for the physical device. +Check if the extension is supported - if returned array of `VkExtensionProperties` contains "VK_EXT_memory_priority". + +2) Call `vkGetPhysicalDeviceFeatures2` for the physical device instead of old `vkGetPhysicalDeviceFeatures`. +Attach additional structure `VkPhysicalDeviceMemoryPriorityFeaturesEXT` to `VkPhysicalDeviceFeatures2::pNext` to be returned. +Check if the device feature is really supported - check if `VkPhysicalDeviceMemoryPriorityFeaturesEXT::memoryPriority` is true. + +3) While creating device with `vkCreateDevice`, enable this extension - add "VK_EXT_memory_priority" +to the list passed as `VkDeviceCreateInfo::ppEnabledExtensionNames`. + +4) While creating the device, also don't set `VkDeviceCreateInfo::pEnabledFeatures`. +Fill in `VkPhysicalDeviceFeatures2` structure instead and pass it as `VkDeviceCreateInfo::pNext`. +Enable this device feature - attach additional structure `VkPhysicalDeviceMemoryPriorityFeaturesEXT` to +`VkPhysicalDeviceFeatures2::pNext` chain and set its member `memoryPriority` to `VK_TRUE`. + +5) While creating #VmaAllocator with vmaCreateAllocator() inform VMA that you +have enabled this extension and feature - add #VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT +to VmaAllocatorCreateInfo::flags. + +\section vk_ext_memory_priority_usage Usage + +When using this extension, you should initialize following member: + +- VmaAllocationCreateInfo::priority when creating a dedicated allocation with #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +- VmaPoolCreateInfo::priority when creating a custom pool. + +It should be a floating-point value between `0.0f` and `1.0f`, where recommended default is `0.5f`. +Memory allocated with higher value can be treated by the Vulkan implementation as higher priority +and so it can have lower chances of being pushed out to system memory, experiencing degraded performance. + +It might be a good idea to create performance-critical resources like color-attachment or depth-stencil images +as dedicated and set high priority to them. For example: + +\code +VkImageCreateInfo imgCreateInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO }; +imgCreateInfo.imageType = VK_IMAGE_TYPE_2D; +imgCreateInfo.extent.width = 3840; +imgCreateInfo.extent.height = 2160; +imgCreateInfo.extent.depth = 1; +imgCreateInfo.mipLevels = 1; +imgCreateInfo.arrayLayers = 1; +imgCreateInfo.format = VK_FORMAT_R8G8B8A8_UNORM; +imgCreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL; +imgCreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; +imgCreateInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; +imgCreateInfo.samples = VK_SAMPLE_COUNT_1_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT; +allocCreateInfo.priority = 1.0f; + +VkImage img; +VmaAllocation alloc; +vmaCreateImage(allocator, &imgCreateInfo, &allocCreateInfo, &img, &alloc, nullptr); +\endcode + +`priority` member is ignored in the following situations: + +- Allocations created in custom pools: They inherit the priority, along with all other allocation parameters + from the parametrs passed in #VmaPoolCreateInfo when the pool was created. +- Allocations created in default pools: They inherit the priority from the parameters + VMA used when creating default pools, which means `priority == 0.5f`. + + \page vk_amd_device_coherent_memory VK_AMD_device_coherent_memory VK_AMD_device_coherent_memory is a device extension that enables access to @@ -19714,7 +19394,7 @@ out of the special `DEVICE_COHERENT` and `DEVICE_UNCACHED` memory types on eligi devices. There are multiple ways to do it, for example: - You can request or prefer to allocate out of such memory types by adding - `VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD` to VmaAllocationCreateInfo::requiredFlags + `VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD` to VmaAllocationCreateInfo::requiredFlags or VmaAllocationCreateInfo::preferredFlags. Those flags can be freely mixed with other ways of \ref choosing_memory_type, like setting VmaAllocationCreateInfo::usage. - If you manually found memory type index to use for this purpose, force allocation @@ -19732,7 +19412,7 @@ accompanying this library. Device extension VK_KHR_buffer_device_address allow to fetch raw GPU pointer to a buffer and pass it for usage in a shader code. -It is promoted to core Vulkan 1.2. +It has been promoted to core Vulkan 1.2. If you want to use this feature in connection with VMA, follow these steps: @@ -19785,6 +19465,7 @@ accompanying this library. - 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. + This includes allocation and deallocation from default memory pool, as well as custom #VmaPool. - 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. @@ -19792,6 +19473,29 @@ accompanying this library. you must not call vmaGetAllocationInfo() and vmaMapMemory() from different threads at the same time if you pass the same #VmaAllocation object to these functions. +- #VmaVirtualBlock is not safe to be used from multiple threads simultaneously. + +\section general_considerations_versioning_and_compatibility Versioning and compatibility + +The library uses [**Semantic Versioning**](https://semver.org/), +which means version numbers follow convention: Major.Minor.Patch (e.g. 2.3.0), where: + +- Incremented Patch version means a release is backward- and forward-compatible, + introducing only some internal improvements, bug fixes, optimizations etc. + or changes that are out of scope of the official API described in this documentation. +- Incremented Minor version means a release is backward-compatible, + so existing code that uses the library should continue to work, while some new + symbols could have been added: new structures, functions, new values in existing + enums and bit flags, new structure members, but not new function parameters. +- Incrementing Major version means a release could break some backward compatibility. + +All changes between official releases are documented in file "CHANGELOG.md". + +\warning Backward compatiblity is considered on the level of C++ source code, not binary linkage. +Adding new members to existing structures is treated as backward compatible if initializing +the new members to binary zero results in the old behavior. +You should always fully initialize all library structures to zeros and not rely on their +exact binary size. \section general_considerations_validation_layer_warnings Validation layer warnings @@ -19807,9 +19511,7 @@ to just ignore them. `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 became lost. - - It may happen also when you use [defragmentation](@ref defragmentation). + - It may happen when you use [defragmentation](@ref defragmentation). \section general_considerations_allocation_algorithm Allocation algorithm @@ -19817,10 +19519,7 @@ 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 create such block with size / 2, size / 4, size / 8. -# 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 @@ -19831,28 +19530,29 @@ The library uses following algorithm for allocation, in order: Features deliberately excluded from the scope of this library: -- **Data transfer.** Uploading (streaming) and downloading data of buffers and images - between CPU and GPU memory and related synchronization is responsibility of the user. - Defining some "texture" object that would automatically stream its data from a - staging copy in CPU memory to GPU memory would rather be a feature of another, - higher-level library implemented on top of VMA. -- **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 is 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. - Success of an allocation is just checked with an assert. -- **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. - There are many preprocessor macros that make some variables unused, function parameters unreferenced, - or conditional expressions constant in some configurations. - The code of this library should not be bigger or more complicated just to silence these warnings. - It is recommended to disable such warnings instead. -- This is a C++ library with C interface. **Bindings or ports to any other programming languages** are welcome as external projects but - are not going to be included into this repository. +-# **Data transfer.** Uploading (streaming) and downloading data of buffers and images + between CPU and GPU memory and related synchronization is responsibility of the user. + Defining some "texture" object that would automatically stream its data from a + staging copy in CPU memory to GPU memory would rather be a feature of another, + higher-level library implemented on top of VMA. + VMA doesn't record any commands to a `VkCommandBuffer`. It just allocates memory. +-# **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 is 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. + Success of an allocation is just checked with an assert. +-# **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. + There are many preprocessor macros that make some variables unused, function parameters unreferenced, + or conditional expressions constant in some configurations. + The code of this library should not be bigger or more complicated just to silence these warnings. + It is recommended to disable such warnings instead. +-# This is a C++ library with C interface. **Bindings or ports to any other programming languages** are welcome as external projects but + are not going to be included into this repository. */ diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkanbuilders.h b/thirdparty/ZVulkan/include/zvulkan/vulkanbuilders.h index e607ece..104d9ce 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkanbuilders.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkanbuilders.h @@ -2,6 +2,117 @@ #include "vulkanobjects.h" #include +#include + +class VulkanCompatibleDevice; + +class VulkanInstanceBuilder +{ +public: + VulkanInstanceBuilder(); + + VulkanInstanceBuilder& ApiVersionsToTry(const std::vector& versions); + VulkanInstanceBuilder& RequireExtension(const std::string& extensionName); + VulkanInstanceBuilder& RequireSurfaceExtensions(bool enable = true); + VulkanInstanceBuilder& OptionalExtension(const std::string& extensionName); + VulkanInstanceBuilder& DebugLayer(bool enable = true); + + std::shared_ptr Create(); + +private: + std::vector apiVersionsToTry; + std::set requiredExtensions; + std::set optionalExtensions; + bool debugLayer = false; +}; + +#ifdef VK_USE_PLATFORM_WIN32_KHR + +class VulkanSurfaceBuilder +{ +public: + VulkanSurfaceBuilder(); + + VulkanSurfaceBuilder& Win32Window(HWND handle); + + std::shared_ptr Create(std::shared_ptr instance); + +private: + HWND hwnd = {}; +}; + +#endif + +class VulkanDeviceBuilder +{ +public: + VulkanDeviceBuilder(); + + VulkanDeviceBuilder& RequireExtension(const std::string& extensionName); + VulkanDeviceBuilder& OptionalExtension(const std::string& extensionName); + VulkanDeviceBuilder& OptionalRayQuery(); + VulkanDeviceBuilder& OptionalDescriptorIndexing(); + VulkanDeviceBuilder& Surface(std::shared_ptr surface); + VulkanDeviceBuilder& SelectDevice(int index); + + std::vector FindDevices(const std::shared_ptr& instance); + std::shared_ptr Create(std::shared_ptr instance); + +private: + std::set requiredDeviceExtensions; + std::set optionalDeviceExtensions; + std::shared_ptr surface; + int deviceIndex = 0; +}; + +class VulkanSwapChainBuilder +{ +public: + VulkanSwapChainBuilder(); + + std::shared_ptr Create(VulkanDevice* device); +}; + +class CommandPoolBuilder +{ +public: + CommandPoolBuilder(); + + CommandPoolBuilder& QueueFamily(int index); + CommandPoolBuilder& DebugName(const char* name) { debugName = name; return *this; } + + std::unique_ptr Create(VulkanDevice* device); + +private: + const char* debugName = nullptr; + int queueFamilyIndex = -1; +}; + +class SemaphoreBuilder +{ +public: + SemaphoreBuilder(); + + SemaphoreBuilder& DebugName(const char* name) { debugName = name; return *this; } + + std::unique_ptr Create(VulkanDevice* device); + +private: + const char* debugName = nullptr; +}; + +class FenceBuilder +{ +public: + FenceBuilder(); + + FenceBuilder& DebugName(const char* name) { debugName = name; return *this; } + + std::unique_ptr Create(VulkanDevice* device); + +private: + const char* debugName = nullptr; +}; class ImageBuilder { @@ -73,6 +184,7 @@ public: BufferBuilder& Size(size_t size); BufferBuilder& Usage(VkBufferUsageFlags bufferUsage, VmaMemoryUsage memoryUsage = VMA_MEMORY_USAGE_GPU_ONLY, VmaAllocationCreateFlags allocFlags = 0); BufferBuilder& MemoryType(VkMemoryPropertyFlags requiredFlags, VkMemoryPropertyFlags preferredFlags, uint32_t memoryTypeBits = 0); + BufferBuilder& MinAlignment(VkDeviceSize memoryAlignment); BufferBuilder& DebugName(const char* name) { debugName = name; return *this; } std::unique_ptr Create(VulkanDevice *device); @@ -81,6 +193,27 @@ private: VkBufferCreateInfo bufferInfo = {}; VmaAllocationCreateInfo allocInfo = {}; const char* debugName = nullptr; + VkDeviceSize minAlignment = 0; +}; + +enum class ShaderType +{ + Vertex, + TessControl, + TessEvaluation, + Geometry, + Fragment, + Compute +}; + +class ShaderIncludeResult +{ +public: + ShaderIncludeResult(std::string name, std::string text) : name(std::move(name)), text(std::move(text)) { } + ShaderIncludeResult(std::string error) : text(std::move(error)) { } + + std::string name; // fully resolved name of the included header file + std::string text; // the file contents - or include error message if name is empty }; class ShaderBuilder @@ -91,16 +224,23 @@ public: static void Init(); static void Deinit(); - ShaderBuilder& VertexShader(const std::string &code); - ShaderBuilder& FragmentShader(const std::string&code); + ShaderBuilder& Type(ShaderType type); + ShaderBuilder& AddSource(const std::string& name, const std::string& code); + + ShaderBuilder& OnIncludeSystem(std::function onIncludeSystem); + ShaderBuilder& OnIncludeLocal(std::function onIncludeLocal); + ShaderBuilder& DebugName(const char* name) { debugName = name; return *this; } std::unique_ptr Create(const char *shadername, VulkanDevice *device); private: - std::string code; + std::vector> sources; + std::function onIncludeSystem; + std::function onIncludeLocal; int stage = 0; const char* debugName = nullptr; + friend class ShaderBuilderIncluderImpl; }; class AccelerationStructureBuilder @@ -125,6 +265,7 @@ class ComputePipelineBuilder public: ComputePipelineBuilder(); + ComputePipelineBuilder& Cache(VulkanPipelineCache* cache); ComputePipelineBuilder& Layout(VulkanPipelineLayout *layout); ComputePipelineBuilder& ComputeShader(VulkanShader *shader); ComputePipelineBuilder& DebugName(const char* name) { debugName = name; return *this; } @@ -134,6 +275,7 @@ public: private: VkComputePipelineCreateInfo pipelineInfo = {}; VkPipelineShaderStageCreateInfo stageInfo = {}; + VulkanPipelineCache* cache = nullptr; const char* debugName = nullptr; }; @@ -213,6 +355,7 @@ class GraphicsPipelineBuilder public: GraphicsPipelineBuilder(); + GraphicsPipelineBuilder& Cache(VulkanPipelineCache* cache); GraphicsPipelineBuilder& Subpass(int subpass); GraphicsPipelineBuilder& Layout(VulkanPipelineLayout *layout); GraphicsPipelineBuilder& RenderPass(VulkanRenderPass *renderPass); @@ -266,6 +409,7 @@ private: std::vector vertexInputAttributes; std::vector dynamicStates; + VulkanPipelineCache* cache = nullptr; const char* debugName = nullptr; }; @@ -288,6 +432,24 @@ private: const char* debugName = nullptr; }; +class PipelineCacheBuilder +{ +public: + PipelineCacheBuilder(); + + PipelineCacheBuilder& InitialData(const void* data, size_t size); + PipelineCacheBuilder& Flags(VkPipelineCacheCreateFlags flags); + + PipelineCacheBuilder& DebugName(const char* name) { debugName = name; return *this; } + + std::unique_ptr Create(VulkanDevice* device); + +private: + VkPipelineCacheCreateInfo pipelineCacheInfo = {}; + std::vector initData; + const char* debugName = nullptr; +}; + class RenderPassBuilder { public: diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkancompatibledevice.h b/thirdparty/ZVulkan/include/zvulkan/vulkancompatibledevice.h index c8f2f30..9a2cd5e 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkancompatibledevice.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkancompatibledevice.h @@ -14,9 +14,6 @@ public: bool GraphicsTimeQueries = false; - std::vector EnabledDeviceExtensions; + std::set EnabledDeviceExtensions; VulkanDeviceFeatures EnabledFeatures; - - static std::vector FindDevices(const std::shared_ptr& instance, const std::shared_ptr& surface); - static VulkanCompatibleDevice SelectDevice(const std::shared_ptr& instance, const std::shared_ptr& surface, int vk_device); }; diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkandevice.h b/thirdparty/ZVulkan/include/zvulkan/vulkandevice.h index 918bc28..c7d0822 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkandevice.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkandevice.h @@ -21,7 +21,7 @@ public: VulkanDevice(std::shared_ptr instance, std::shared_ptr surface, const VulkanCompatibleDevice& selectedDevice); ~VulkanDevice(); - std::vector EnabledDeviceExtensions; + std::set EnabledDeviceExtensions; VulkanDeviceFeatures EnabledFeatures; VulkanPhysicalDevice PhysicalDevice; @@ -39,7 +39,7 @@ public: int PresentFamily = -1; bool GraphicsTimeQueries = false; - bool SupportsDeviceExtension(const char* ext) const; + bool SupportsExtension(const char* ext) const; void SetObjectName(const char* name, uint64_t handle, VkObjectType type); diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkaninstance.h b/thirdparty/ZVulkan/include/zvulkan/vulkaninstance.h index 490debe..8522a8a 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkaninstance.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkaninstance.h @@ -5,8 +5,6 @@ #elif defined(__APPLE__) #define VK_USE_PLATFORM_MACOS_MVK #define VK_USE_PLATFORM_METAL_EXT -#else -#define VK_USE_PLATFORM_XLIB_KHR #endif #include "volk/volk.h" @@ -21,6 +19,7 @@ #include #include #include +#include class VulkanDeviceFeatures { @@ -32,47 +31,41 @@ public: VkPhysicalDeviceDescriptorIndexingFeatures DescriptorIndexing = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT }; }; +class VulkanDeviceProperties +{ +public: + VkPhysicalDeviceProperties Properties = {}; + VkPhysicalDeviceMemoryProperties Memory = {}; + VkPhysicalDeviceAccelerationStructurePropertiesKHR AccelerationStructure = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ACCELERATION_STRUCTURE_PROPERTIES_KHR }; + VkPhysicalDeviceDescriptorIndexingProperties DescriptorIndexing = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT }; +}; + class VulkanPhysicalDevice { public: VkPhysicalDevice Device = VK_NULL_HANDLE; std::vector Extensions; std::vector QueueFamilies; - VkPhysicalDeviceProperties Properties = {}; - VkPhysicalDeviceMemoryProperties MemoryProperties = {}; + VulkanDeviceProperties Properties; VulkanDeviceFeatures Features; }; class VulkanInstance { public: - VulkanInstance(bool wantDebugLayer); + VulkanInstance(std::vector apiVersionsToTry, std::set requiredExtensions, std::set optionalExtensions, bool wantDebugLayer); ~VulkanInstance(); - std::vector RequiredExtensions = - { - VK_KHR_SURFACE_EXTENSION_NAME, -#if defined(VK_USE_PLATFORM_WIN32_KHR) - VK_KHR_WIN32_SURFACE_EXTENSION_NAME -#elif defined(VK_USE_PLATFORM_MACOS_MVK) - VK_MVK_MACOS_SURFACE_EXTENSION_NAME -#elif defined(VK_USE_PLATFORM_XLIB_KHR) - VK_KHR_XLIB_SURFACE_EXTENSION_NAME -#endif - }; - std::vector OptionalExtensions = - { - VK_EXT_SWAPCHAIN_COLOR_SPACE_EXTENSION_NAME, - VK_KHR_GET_SURFACE_CAPABILITIES_2_EXTENSION_NAME, - VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME - }; - std::vector ApiVersionsToTry = { VK_API_VERSION_1_2, VK_API_VERSION_1_1, VK_API_VERSION_1_0 }; + std::vector ApiVersionsToTry; + + std::set RequiredExtensions; + std::set OptionalExtensions; std::vector AvailableLayers; std::vector AvailableExtensions; - std::vector EnabledValidationLayers; - std::vector EnabledExtensions; + std::set EnabledValidationLayers; + std::set EnabledExtensions; std::vector PhysicalDevices; diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkanobjects.h b/thirdparty/ZVulkan/include/zvulkan/vulkanobjects.h index 0b447ec..c7842f6 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkanobjects.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkanobjects.h @@ -305,6 +305,24 @@ private: VulkanPipelineLayout &operator=(const VulkanPipelineLayout &) = delete; }; +class VulkanPipelineCache +{ +public: + VulkanPipelineCache(VulkanDevice* device, VkPipelineCache cache); + ~VulkanPipelineCache(); + + void SetDebugName(const char* name) { device->SetObjectName(name, (uint64_t)cache, VK_OBJECT_TYPE_PIPELINE_CACHE); } + + std::vector GetCacheData(); + + VulkanDevice* device; + VkPipelineCache cache; + +private: + VulkanPipelineCache(const VulkanPipelineCache&) = delete; + VulkanPipelineCache& operator=(const VulkanPipelineCache&) = delete; +}; + class VulkanRenderPass { public: @@ -1105,7 +1123,8 @@ inline VulkanImage::VulkanImage(VulkanDevice *device, VkImage image, VmaAllocati inline VulkanImage::~VulkanImage() { - vmaDestroyImage(device->allocator, image, allocation); + if (allocation) + vmaDestroyImage(device->allocator, image, allocation); } inline void *VulkanImage::Map(size_t offset, size_t size) @@ -1178,6 +1197,33 @@ inline VulkanPipelineLayout::~VulkanPipelineLayout() ///////////////////////////////////////////////////////////////////////////// +inline VulkanPipelineCache::VulkanPipelineCache(VulkanDevice* device, VkPipelineCache cache) : device(device), cache(cache) +{ +} + +inline VulkanPipelineCache::~VulkanPipelineCache() +{ + vkDestroyPipelineCache(device->device, cache, nullptr); +} + +inline std::vector VulkanPipelineCache::GetCacheData() +{ + size_t dataSize = 0; + VkResult result = vkGetPipelineCacheData(device->device, cache, &dataSize, nullptr); + CheckVulkanError(result, "Could not get cache data size"); + + std::vector buffer; + buffer.resize(dataSize); + result = vkGetPipelineCacheData(device->device, cache, &dataSize, buffer.data()); + if (result == VK_INCOMPLETE) + VulkanError("Could not get cache data (incomplete)"); + CheckVulkanError(result, "Could not get cache data"); + buffer.resize(dataSize); + return buffer; +} + +///////////////////////////////////////////////////////////////////////////// + inline VulkanRenderPass::VulkanRenderPass(VulkanDevice *device, VkRenderPass renderPass) : device(device), renderPass(renderPass) { } diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkansurface.h b/thirdparty/ZVulkan/include/zvulkan/vulkansurface.h index 5d3d41c..398bfd3 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkansurface.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkansurface.h @@ -2,45 +2,19 @@ #include "vulkaninstance.h" +class VulkanSurface +{ +public: + VulkanSurface(std::shared_ptr instance, VkSurfaceKHR surface); + ~VulkanSurface(); + + std::shared_ptr Instance; + VkSurfaceKHR Surface = VK_NULL_HANDLE; + #ifdef VK_USE_PLATFORM_WIN32_KHR -class VulkanSurface -{ -public: VulkanSurface(std::shared_ptr instance, HWND window); - ~VulkanSurface(); - - std::shared_ptr Instance; - VkSurfaceKHR Surface = VK_NULL_HANDLE; HWND Window = 0; -}; - -#elif defined(VK_USE_PLATFORM_XLIB_KHR) - -#include - -class VulkanSurface -{ -public: - VulkanSurface(std::shared_ptr instance, Display* disp, Window wind); - ~VulkanSurface(); - - std::shared_ptr Instance; - VkSurfaceKHR Surface = VK_NULL_HANDLE; - Display* disp = nullptr; - Window wind; -}; - -#else - -class VulkanSurface -{ -public: - VulkanSurface(std::shared_ptr instance); - ~VulkanSurface(); - - std::shared_ptr Instance; - VkSurfaceKHR Surface = VK_NULL_HANDLE; -}; #endif +}; diff --git a/thirdparty/ZVulkan/include/zvulkan/vulkanswapchain.h b/thirdparty/ZVulkan/include/zvulkan/vulkanswapchain.h index 8c5d720..6c5e97d 100644 --- a/thirdparty/ZVulkan/include/zvulkan/vulkanswapchain.h +++ b/thirdparty/ZVulkan/include/zvulkan/vulkanswapchain.h @@ -1,50 +1,61 @@ #pragma once #include "vulkandevice.h" +#include "vulkanobjects.h" class VulkanSemaphore; class VulkanFence; +class VulkanSurfaceCapabilities +{ +public: + VkSurfaceCapabilitiesKHR Capabilites = { }; +#ifdef WIN32 + VkSurfaceCapabilitiesFullScreenExclusiveEXT FullScreenExclusive = { VK_STRUCTURE_TYPE_SURFACE_CAPABILITIES_FULL_SCREEN_EXCLUSIVE_EXT }; +#else + struct { void* pNext = nullptr; VkBool32 fullScreenExclusiveSupported = 0; } FullScreenExclusive; +#endif + std::vector PresentModes; + std::vector Formats; +}; + class VulkanSwapChain { public: - VulkanSwapChain(VulkanDevice *device, bool vsync); + VulkanSwapChain(VulkanDevice* device); ~VulkanSwapChain(); - uint32_t acquireImage(int width, int height, bool exclusivefullscreen, VulkanSemaphore *semaphore = nullptr, VulkanFence *fence = nullptr); - void queuePresent(uint32_t imageIndex, VulkanSemaphore *semaphore = nullptr); + void Create(int width, int height, int imageCount, bool vsync, bool hdr, bool exclusivefullscreen); + bool Lost() const { return lost; } - bool vsync; - VkSwapchainKHR swapChain = VK_NULL_HANDLE; - VkSurfaceFormatKHR swapChainFormat; - VkPresentModeKHR swapChainPresentMode; + int Width() const { return actualExtent.width; } + int Height() const { return actualExtent.height; } + VkSurfaceFormatKHR Format() const { return format; } - std::vector swapChainImages; - std::vector swapChainImageViews; + int ImageCount() const { return (int)images.size(); } + VulkanImage* GetImage(int index) { return images[index].get(); } + VulkanImageView* GetImageView(int index) { return views[index].get(); } - VkExtent2D actualExtent; - - bool newSwapChain = true; + int AcquireImage(VulkanSemaphore* semaphore = nullptr, VulkanFence* fence = nullptr); + void QueuePresent(int imageIndex, VulkanSemaphore* semaphore = nullptr); private: - void selectFormat(); - void selectPresentMode(bool fullscreen); - bool createSwapChain(int width, int height, bool fullscreen, VkSwapchainKHR oldSwapChain = VK_NULL_HANDLE); - void createViews(); - void getImages(); - void releaseResources(); - void releaseViews(); - void recreate(int width, int height, bool fullscreen); + void SelectFormat(const VulkanSurfaceCapabilities& caps, bool hdr); - std::vector getSurfaceFormats(); - std::vector getPresentModes(bool exclusivefullscreen); + bool CreateSwapchain(int width, int height, int imageCount, bool vsync, bool hdr, bool exclusivefullscreen); - VulkanDevice *device = nullptr; + VulkanSurfaceCapabilities GetSurfaceCapabilities(bool exclusivefullscreen); - int lastSwapWidth = 0; - int lastSwapHeight = 0; - bool lastFullscreen = false; + VulkanDevice* device = nullptr; + bool lost = true; - VulkanSwapChain(const VulkanSwapChain &) = delete; - VulkanSwapChain &operator=(const VulkanSwapChain &) = delete; + VkExtent2D actualExtent = {}; + VkSwapchainKHR swapchain = VK_NULL_HANDLE; + VkSurfaceFormatKHR format = {}; + VkPresentModeKHR presentMode; + std::vector> images; + std::vector> views; + + VulkanSwapChain(const VulkanSwapChain&) = delete; + VulkanSwapChain& operator=(const VulkanSwapChain&) = delete; }; diff --git a/thirdparty/ZVulkan/src/vulkanbuilders.cpp b/thirdparty/ZVulkan/src/vulkanbuilders.cpp index a88f8dd..5a0db35 100644 --- a/thirdparty/ZVulkan/src/vulkanbuilders.cpp +++ b/thirdparty/ZVulkan/src/vulkanbuilders.cpp @@ -1,5 +1,8 @@ #include "vulkanbuilders.h" +#include "vulkansurface.h" +#include "vulkancompatibledevice.h" +#include "vulkanswapchain.h" #include "glslang/glslang/Public/ShaderLang.h" #include "glslang/spirv/GlslangToSpv.h" @@ -125,29 +128,143 @@ ShaderBuilder::ShaderBuilder() { } -ShaderBuilder& ShaderBuilder::VertexShader(const std::string& c) +ShaderBuilder& ShaderBuilder::Type(ShaderType type) { - code = c; - stage = EShLanguage::EShLangVertex; + switch (type) + { + case ShaderType::Vertex: stage = EShLanguage::EShLangVertex; break; + case ShaderType::TessControl: stage = EShLanguage::EShLangTessControl; break; + case ShaderType::TessEvaluation: stage = EShLanguage::EShLangTessEvaluation; break; + case ShaderType::Geometry: stage = EShLanguage::EShLangGeometry; break; + case ShaderType::Fragment: stage = EShLanguage::EShLangFragment; break; + case ShaderType::Compute: stage = EShLanguage::EShLangCompute; break; + } return *this; } -ShaderBuilder& ShaderBuilder::FragmentShader(const std::string& c) +ShaderBuilder& ShaderBuilder::AddSource(const std::string& name, const std::string& code) { - code = c; - stage = EShLanguage::EShLangFragment; + sources.push_back({ name, code }); return *this; } +ShaderBuilder& ShaderBuilder::OnIncludeSystem(std::function onIncludeSystem) +{ + this->onIncludeSystem = std::move(onIncludeSystem); + return *this; +} + +ShaderBuilder& ShaderBuilder::OnIncludeLocal(std::function onIncludeLocal) +{ + this->onIncludeLocal = std::move(onIncludeLocal); + return *this; +} + +class ShaderBuilderIncluderImpl : public glslang::TShader::Includer +{ +public: + ShaderBuilderIncluderImpl(ShaderBuilder* shaderBuilder) : shaderBuilder(shaderBuilder) + { + } + + IncludeResult* includeSystem(const char* headerName, const char* includerName, size_t inclusionDepth) override + { + if (!shaderBuilder->onIncludeSystem) + { + return nullptr; + } + + try + { + std::unique_ptr result; + try + { + result = std::make_unique(shaderBuilder->onIncludeSystem(headerName, includerName, inclusionDepth)); + } + catch (const std::exception& e) + { + result = std::make_unique(e.what()); + } + + if (!result || (result->name.empty() && result->text.empty())) + { + return nullptr; + } + + IncludeResult* outer = new IncludeResult(result->name, result->text.data(), result->text.size(), result.get()); + result.release(); + return outer; + } + catch (...) + { + return nullptr; + } + } + + IncludeResult* includeLocal(const char* headerName, const char* includerName, size_t inclusionDepth) override + { + if (!shaderBuilder->onIncludeLocal) + { + return nullptr; + } + + try + { + std::unique_ptr result; + try + { + result = std::make_unique(shaderBuilder->onIncludeLocal(headerName, includerName, inclusionDepth)); + } + catch (const std::exception& e) + { + result = std::make_unique(e.what()); + } + + if (!result || (result->name.empty() && result->text.empty())) + { + return nullptr; + } + + IncludeResult* outer = new IncludeResult(result->name, result->text.data(), result->text.size(), result.get()); + result.release(); + return outer; + } + catch (...) + { + return nullptr; + } + } + + void releaseInclude(IncludeResult* result) override + { + if (result) + { + delete (ShaderIncludeResult*)result->userData; + delete result; + } + } + +private: + ShaderBuilder* shaderBuilder = nullptr; +}; + std::unique_ptr ShaderBuilder::Create(const char *shadername, VulkanDevice *device) { EShLanguage stage = (EShLanguage)this->stage; - const char *sources[] = { code.c_str() }; + + std::vector namesC, sourcesC; + std::vector lengthsC; + for (const auto& s : sources) + { + namesC.push_back(s.first.c_str()); + sourcesC.push_back(s.second.c_str()); + lengthsC.push_back((int)s.second.size()); + } TBuiltInResource resources = DefaultTBuiltInResource; glslang::TShader shader(stage); - shader.setStrings(sources, 1); + shader.setStringsWithLengthsAndNames(sourcesC.data(), lengthsC.data(), namesC.data(), (int)sources.size()); shader.setEnvInput(glslang::EShSourceGlsl, stage, glslang::EShClientVulkan, 100); if (device->Instance->ApiVersion >= VK_API_VERSION_1_2) { @@ -159,7 +276,9 @@ std::unique_ptr ShaderBuilder::Create(const char *shadername, Vulk shader.setEnvClient(glslang::EShClientVulkan, glslang::EShTargetVulkan_1_0); shader.setEnvTarget(glslang::EShTargetSpv, glslang::EShTargetSpv_1_0); } - bool compileSuccess = shader.parse(&resources, 110, false, EShMsgVulkanRules); + + ShaderBuilderIncluderImpl includer(this); + bool compileSuccess = shader.parse(&resources, 110, false, EShMsgVulkanRules, includer); if (!compileSuccess) { throw std::runtime_error(std::string("Shader compile failed: ") + shader.getInfoLog()); @@ -206,6 +325,54 @@ std::unique_ptr ShaderBuilder::Create(const char *shadername, Vulk ///////////////////////////////////////////////////////////////////////////// +CommandPoolBuilder::CommandPoolBuilder() +{ +} + +CommandPoolBuilder& CommandPoolBuilder::QueueFamily(int index) +{ + queueFamilyIndex = index; + return *this; +} + +std::unique_ptr CommandPoolBuilder::Create(VulkanDevice* device) +{ + auto obj = std::make_unique(device, queueFamilyIndex != -1 ? queueFamilyIndex : device->GraphicsFamily); + if (debugName) + obj->SetDebugName(debugName); + return obj; +} + +///////////////////////////////////////////////////////////////////////////// + +SemaphoreBuilder::SemaphoreBuilder() +{ +} + +std::unique_ptr SemaphoreBuilder::Create(VulkanDevice* device) +{ + auto obj = std::make_unique(device); + if (debugName) + obj->SetDebugName(debugName); + return obj; +} + +///////////////////////////////////////////////////////////////////////////// + +FenceBuilder::FenceBuilder() +{ +} + +std::unique_ptr FenceBuilder::Create(VulkanDevice* device) +{ + auto obj = std::make_unique(device); + if (debugName) + obj->SetDebugName(debugName); + return obj; +} + +///////////////////////////////////////////////////////////////////////////// + ImageBuilder::ImageBuilder() { imageInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO; @@ -476,13 +643,27 @@ BufferBuilder& BufferBuilder::MemoryType(VkMemoryPropertyFlags requiredFlags, Vk return *this; } +BufferBuilder& BufferBuilder::MinAlignment(VkDeviceSize memoryAlignment) +{ + minAlignment = memoryAlignment; + return *this; +} + std::unique_ptr BufferBuilder::Create(VulkanDevice* device) { VkBuffer buffer; VmaAllocation allocation; - VkResult result = vmaCreateBuffer(device->allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); - CheckVulkanError(result, "Could not allocate memory for vulkan buffer"); + if (minAlignment == 0) + { + VkResult result = vmaCreateBuffer(device->allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); + CheckVulkanError(result, "Could not allocate memory for vulkan buffer"); + } + else + { + VkResult result = vmaCreateBufferWithAlignment(device->allocator, &bufferInfo, &allocInfo, minAlignment, &buffer, &allocation, nullptr); + CheckVulkanError(result, "Could not allocate memory for vulkan buffer"); + } auto obj = std::make_unique(device, buffer, allocation, bufferInfo.size); if (debugName) @@ -538,6 +719,12 @@ ComputePipelineBuilder::ComputePipelineBuilder() stageInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO; } +ComputePipelineBuilder& ComputePipelineBuilder::Cache(VulkanPipelineCache* cache) +{ + this->cache = cache; + return *this; +} + ComputePipelineBuilder& ComputePipelineBuilder::Layout(VulkanPipelineLayout* layout) { pipelineInfo.layout = layout->layout; @@ -557,7 +744,7 @@ ComputePipelineBuilder& ComputePipelineBuilder::ComputeShader(VulkanShader* shad std::unique_ptr ComputePipelineBuilder::Create(VulkanDevice* device) { VkPipeline pipeline; - vkCreateComputePipelines(device->device, VK_NULL_HANDLE, 1, &pipelineInfo, nullptr, &pipeline); + vkCreateComputePipelines(device->device, cache ? cache->cache : VK_NULL_HANDLE, 1, &pipelineInfo, nullptr, &pipeline); auto obj = std::make_unique(device, pipeline); if (debugName) obj->SetDebugName(debugName); @@ -758,6 +945,10 @@ GraphicsPipelineBuilder::GraphicsPipelineBuilder() inputAssembly.primitiveRestartEnable = VK_FALSE; viewportState.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO; + viewportState.viewportCount = 1; + viewportState.pViewports = &viewport; + viewportState.scissorCount = 1; + viewportState.pScissors = &scissor; depthStencil.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO; depthStencil.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL; @@ -816,6 +1007,12 @@ GraphicsPipelineBuilder& GraphicsPipelineBuilder::RasterizationSamples(VkSampleC return *this; } +GraphicsPipelineBuilder& GraphicsPipelineBuilder::Cache(VulkanPipelineCache* cache) +{ + this->cache = cache; + return *this; +} + GraphicsPipelineBuilder& GraphicsPipelineBuilder::Subpass(int subpass) { pipelineInfo.subpass = subpass; @@ -848,9 +1045,6 @@ GraphicsPipelineBuilder& GraphicsPipelineBuilder::Viewport(float x, float y, flo viewport.height = height; viewport.minDepth = minDepth; viewport.maxDepth = maxDepth; - - viewportState.viewportCount = 1; - viewportState.pViewports = &viewport; return *this; } @@ -860,9 +1054,6 @@ GraphicsPipelineBuilder& GraphicsPipelineBuilder::Scissor(int x, int y, int widt scissor.offset.y = y; scissor.extent.width = width; scissor.extent.height = height; - - viewportState.scissorCount = 1; - viewportState.pScissors = &scissor; return *this; } @@ -1038,7 +1229,7 @@ GraphicsPipelineBuilder& GraphicsPipelineBuilder::AddDynamicState(VkDynamicState std::unique_ptr GraphicsPipelineBuilder::Create(VulkanDevice* device) { VkPipeline pipeline = 0; - VkResult result = vkCreateGraphicsPipelines(device->device, VK_NULL_HANDLE, 1, &pipelineInfo, nullptr, &pipeline); + VkResult result = vkCreateGraphicsPipelines(device->device, cache ? cache->cache : VK_NULL_HANDLE, 1, &pipelineInfo, nullptr, &pipeline); CheckVulkanError(result, "Could not create graphics pipeline"); auto obj = std::make_unique(device, pipeline); if (debugName) @@ -1086,6 +1277,54 @@ std::unique_ptr PipelineLayoutBuilder::Create(VulkanDevice ///////////////////////////////////////////////////////////////////////////// +PipelineCacheBuilder::PipelineCacheBuilder() +{ + pipelineCacheInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO; +} + +PipelineCacheBuilder& PipelineCacheBuilder::InitialData(const void* data, size_t size) +{ + initData.resize(size); + memcpy(initData.data(), data, size); + return *this; +} + +PipelineCacheBuilder& PipelineCacheBuilder::Flags(VkPipelineCacheCreateFlags flags) +{ + pipelineCacheInfo.flags = flags; + return *this; +} + +std::unique_ptr PipelineCacheBuilder::Create(VulkanDevice* device) +{ + pipelineCacheInfo.pInitialData = nullptr; + pipelineCacheInfo.initialDataSize = 0; + + // Check if the saved cache data is compatible with our device: + if (initData.size() >= sizeof(VkPipelineCacheHeaderVersionOne)) + { + VkPipelineCacheHeaderVersionOne* header = (VkPipelineCacheHeaderVersionOne*)initData.data(); + if (header->headerVersion == VK_PIPELINE_CACHE_HEADER_VERSION_ONE && + header->vendorID == device->PhysicalDevice.Properties.Properties.vendorID && + header->deviceID == device->PhysicalDevice.Properties.Properties.deviceID && + memcmp(header->pipelineCacheUUID, device->PhysicalDevice.Properties.Properties.pipelineCacheUUID, VK_UUID_SIZE) == 0) + { + pipelineCacheInfo.pInitialData = initData.data(); + pipelineCacheInfo.initialDataSize = initData.size(); + } + } + + VkPipelineCache pipelineCache; + VkResult result = vkCreatePipelineCache(device->device, &pipelineCacheInfo, nullptr, &pipelineCache); + CheckVulkanError(result, "Could not create pipeline cache"); + auto obj = std::make_unique(device, pipelineCache); + if (debugName) + obj->SetDebugName(debugName); + return obj; +} + +///////////////////////////////////////////////////////////////////////////// + RenderPassBuilder::RenderPassBuilder() { renderPassInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; @@ -1433,3 +1672,267 @@ void WriteDescriptors::Execute(VulkanDevice* device) if (!writes.empty()) vkUpdateDescriptorSets(device->device, (uint32_t)writes.size(), writes.data(), 0, nullptr); } + +///////////////////////////////////////////////////////////////////////////// + +VulkanInstanceBuilder::VulkanInstanceBuilder() +{ + apiVersionsToTry = { VK_API_VERSION_1_2, VK_API_VERSION_1_1, VK_API_VERSION_1_0 }; + + OptionalExtension(VK_KHR_GET_SURFACE_CAPABILITIES_2_EXTENSION_NAME); + OptionalExtension(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME); +} + +VulkanInstanceBuilder& VulkanInstanceBuilder::ApiVersionsToTry(const std::vector& versions) +{ + apiVersionsToTry = versions; + return *this; +} + +VulkanInstanceBuilder& VulkanInstanceBuilder::RequireExtension(const std::string& extensionName) +{ + requiredExtensions.insert(extensionName); + return *this; +} + +VulkanInstanceBuilder& VulkanInstanceBuilder::RequireSurfaceExtensions(bool enable) +{ + if (enable) + { + RequireExtension(VK_KHR_SURFACE_EXTENSION_NAME); + +#if defined(VK_USE_PLATFORM_WIN32_KHR) + RequireExtension(VK_KHR_WIN32_SURFACE_EXTENSION_NAME); +#elif defined(VK_USE_PLATFORM_MACOS_MVK) + RequireExtension(VK_MVK_MACOS_SURFACE_EXTENSION_NAME); +#elif defined(VK_USE_PLATFORM_XLIB_KHR) + RequireExtension(VK_KHR_XLIB_SURFACE_EXTENSION_NAME); +#endif + + OptionalExtension(VK_EXT_SWAPCHAIN_COLOR_SPACE_EXTENSION_NAME); // For HDR support + } + return *this; +} + +VulkanInstanceBuilder& VulkanInstanceBuilder::OptionalExtension(const std::string& extensionName) +{ + optionalExtensions.insert(extensionName); + return *this; +} + +VulkanInstanceBuilder& VulkanInstanceBuilder::DebugLayer(bool enable) +{ + debugLayer = enable; + return *this; +} + +std::shared_ptr VulkanInstanceBuilder::Create() +{ + return std::make_shared(apiVersionsToTry, requiredExtensions, optionalExtensions, debugLayer); +} + +///////////////////////////////////////////////////////////////////////////// + +#ifdef VK_USE_PLATFORM_WIN32_KHR + +VulkanSurfaceBuilder::VulkanSurfaceBuilder() +{ +} + +VulkanSurfaceBuilder& VulkanSurfaceBuilder::Win32Window(HWND hwnd) +{ + this->hwnd = hwnd; + return *this; +} + +std::shared_ptr VulkanSurfaceBuilder::Create(std::shared_ptr instance) +{ + return std::make_shared(std::move(instance), hwnd); +} + +#endif + +///////////////////////////////////////////////////////////////////////////// + +VulkanDeviceBuilder::VulkanDeviceBuilder() +{ + OptionalExtension(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME); + OptionalExtension(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME); +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::RequireExtension(const std::string& extensionName) +{ + requiredDeviceExtensions.insert(extensionName); + return *this; +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::OptionalExtension(const std::string& extensionName) +{ + optionalDeviceExtensions.insert(extensionName); + return *this; +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::OptionalRayQuery() +{ + OptionalExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME); + OptionalExtension(VK_KHR_ACCELERATION_STRUCTURE_EXTENSION_NAME); + OptionalExtension(VK_KHR_DEFERRED_HOST_OPERATIONS_EXTENSION_NAME); + OptionalExtension(VK_KHR_RAY_QUERY_EXTENSION_NAME); + return *this; +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::OptionalDescriptorIndexing() +{ + OptionalExtension(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME); + return *this; +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::Surface(std::shared_ptr surface) +{ + if (surface) + { + RequireExtension(VK_KHR_SWAPCHAIN_EXTENSION_NAME); +#if defined(VK_USE_PLATFORM_WIN32_KHR) + OptionalExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME); +#endif + } + this->surface = std::move(surface); + return *this; +} + +VulkanDeviceBuilder& VulkanDeviceBuilder::SelectDevice(int index) +{ + deviceIndex = index; + return *this; +} + +std::vector VulkanDeviceBuilder::FindDevices(const std::shared_ptr& instance) +{ + std::vector supportedDevices; + + for (size_t idx = 0; idx < instance->PhysicalDevices.size(); idx++) + { + const auto& info = instance->PhysicalDevices[idx]; + + // Check if all required extensions are there + std::set requiredExtensionSearch = requiredDeviceExtensions; + for (const auto& ext : info.Extensions) + requiredExtensionSearch.erase(ext.extensionName); + if (!requiredExtensionSearch.empty()) + continue; + + // Check if all required features are there + if (info.Features.Features.samplerAnisotropy != VK_TRUE || + info.Features.Features.fragmentStoresAndAtomics != VK_TRUE) + continue; + + VulkanCompatibleDevice dev; + dev.Device = &instance->PhysicalDevices[idx]; + dev.EnabledDeviceExtensions = requiredDeviceExtensions; + + // Enable optional extensions we are interested in, if they are available on this device + for (const auto& ext : dev.Device->Extensions) + { + if (optionalDeviceExtensions.find(ext.extensionName) != optionalDeviceExtensions.end()) + { + dev.EnabledDeviceExtensions.insert(ext.extensionName); + } + } + + // Enable optional features we are interested in, if they are available on this device + auto& enabledFeatures = dev.EnabledFeatures; + auto& deviceFeatures = dev.Device->Features; + enabledFeatures.Features.samplerAnisotropy = deviceFeatures.Features.samplerAnisotropy; + enabledFeatures.Features.fragmentStoresAndAtomics = deviceFeatures.Features.fragmentStoresAndAtomics; + enabledFeatures.Features.depthClamp = deviceFeatures.Features.depthClamp; + enabledFeatures.Features.shaderClipDistance = deviceFeatures.Features.shaderClipDistance; + enabledFeatures.BufferDeviceAddress.bufferDeviceAddress = deviceFeatures.BufferDeviceAddress.bufferDeviceAddress; + enabledFeatures.AccelerationStructure.accelerationStructure = deviceFeatures.AccelerationStructure.accelerationStructure; + enabledFeatures.RayQuery.rayQuery = deviceFeatures.RayQuery.rayQuery; + enabledFeatures.DescriptorIndexing.runtimeDescriptorArray = deviceFeatures.DescriptorIndexing.runtimeDescriptorArray; + enabledFeatures.DescriptorIndexing.descriptorBindingPartiallyBound = deviceFeatures.DescriptorIndexing.descriptorBindingPartiallyBound; + enabledFeatures.DescriptorIndexing.descriptorBindingSampledImageUpdateAfterBind = deviceFeatures.DescriptorIndexing.descriptorBindingSampledImageUpdateAfterBind; + enabledFeatures.DescriptorIndexing.descriptorBindingVariableDescriptorCount = deviceFeatures.DescriptorIndexing.descriptorBindingVariableDescriptorCount; + + // Figure out which queue can present + if (surface) + { + for (int i = 0; i < (int)info.QueueFamilies.size(); i++) + { + VkBool32 presentSupport = false; + VkResult result = vkGetPhysicalDeviceSurfaceSupportKHR(info.Device, i, surface->Surface, &presentSupport); + if (result == VK_SUCCESS && info.QueueFamilies[i].queueCount > 0 && presentSupport) + { + dev.PresentFamily = i; + break; + } + } + } + + // The vulkan spec states that graphics and compute queues can always do transfer. + // Furthermore the spec states that graphics queues always can do compute. + // Last, the spec makes it OPTIONAL whether the VK_QUEUE_TRANSFER_BIT is set for such queues, but they MUST support transfer. + // + // In short: pick the first graphics queue family for everything. + for (int i = 0; i < (int)info.QueueFamilies.size(); i++) + { + const auto& queueFamily = info.QueueFamilies[i]; + if (queueFamily.queueCount > 0 && (queueFamily.queueFlags & VK_QUEUE_GRAPHICS_BIT)) + { + dev.GraphicsFamily = i; + dev.GraphicsTimeQueries = queueFamily.timestampValidBits != 0; + break; + } + } + + // Only use device if we found the required graphics and present queues + if (dev.GraphicsFamily != -1 && (!surface || dev.PresentFamily != -1)) + { + supportedDevices.push_back(dev); + } + } + + // The device order returned by Vulkan can be anything. Prefer discrete > integrated > virtual gpu > cpu > other + auto sortFunc = [&](const auto& a, const auto b) + { + // Sort by GPU type first. This will ensure the "best" device is most likely to map to vk_device 0 + static const int typeSort[] = { 4, 1, 0, 2, 3 }; + int sortA = a.Device->Properties.Properties.deviceType < 5 ? typeSort[a.Device->Properties.Properties.deviceType] : (int)a.Device->Properties.Properties.deviceType; + int sortB = b.Device->Properties.Properties.deviceType < 5 ? typeSort[b.Device->Properties.Properties.deviceType] : (int)b.Device->Properties.Properties.deviceType; + if (sortA != sortB) + return sortA < sortB; + + // Then sort by the device's unique ID so that vk_device uses a consistent order + int sortUUID = memcmp(a.Device->Properties.Properties.pipelineCacheUUID, b.Device->Properties.Properties.pipelineCacheUUID, VK_UUID_SIZE); + return sortUUID < 0; + }; + std::stable_sort(supportedDevices.begin(), supportedDevices.end(), sortFunc); + + return supportedDevices; +} + +std::shared_ptr VulkanDeviceBuilder::Create(std::shared_ptr instance) +{ + if (instance->PhysicalDevices.empty()) + VulkanError("No Vulkan devices found. The graphics card may have no vulkan support or the driver may be too old."); + + std::vector supportedDevices = FindDevices(instance); + if (supportedDevices.empty()) + VulkanError("No Vulkan device found supports the minimum requirements of this application"); + + size_t selected = deviceIndex; + if (selected >= supportedDevices.size()) + selected = 0; + return std::make_shared(instance, surface, supportedDevices[selected]); +} + +///////////////////////////////////////////////////////////////////////////// + +VulkanSwapChainBuilder::VulkanSwapChainBuilder() +{ +} + +std::shared_ptr VulkanSwapChainBuilder::Create(VulkanDevice* device) +{ + return std::make_shared(device); +} diff --git a/thirdparty/ZVulkan/src/vulkandevice.cpp b/thirdparty/ZVulkan/src/vulkandevice.cpp index 81085d1..0adfff5 100644 --- a/thirdparty/ZVulkan/src/vulkandevice.cpp +++ b/thirdparty/ZVulkan/src/vulkandevice.cpp @@ -33,18 +33,20 @@ VulkanDevice::~VulkanDevice() ReleaseResources(); } -bool VulkanDevice::SupportsDeviceExtension(const char* ext) const +bool VulkanDevice::SupportsExtension(const char* ext) const { - return std::find(EnabledDeviceExtensions.begin(), EnabledDeviceExtensions.end(), ext) != EnabledDeviceExtensions.end(); + return + EnabledDeviceExtensions.find(ext) != EnabledDeviceExtensions.end() || + Instance->EnabledExtensions.find(ext) != Instance->EnabledExtensions.end(); } void VulkanDevice::CreateAllocator() { VmaAllocatorCreateInfo allocinfo = {}; allocinfo.vulkanApiVersion = Instance->ApiVersion; - if (SupportsDeviceExtension(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME) && SupportsDeviceExtension(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME) && SupportsExtension(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME)) allocinfo.flags |= VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT; - if (SupportsDeviceExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) allocinfo.flags |= VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT; allocinfo.physicalDevice = PhysicalDevice.Device; allocinfo.device = device; @@ -91,7 +93,7 @@ void VulkanDevice::CreateDevice() deviceFeatures2.features = EnabledFeatures.Features; void** next = const_cast(&deviceCreateInfo.pNext); - if (SupportsDeviceExtension(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME)) { *next = &deviceFeatures2; next = &deviceFeatures2.pNext; @@ -101,22 +103,22 @@ void VulkanDevice::CreateDevice() deviceCreateInfo.pEnabledFeatures = &deviceFeatures2.features; } - if (SupportsDeviceExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) { *next = &EnabledFeatures.BufferDeviceAddress; next = &EnabledFeatures.BufferDeviceAddress.pNext; } - if (SupportsDeviceExtension(VK_KHR_ACCELERATION_STRUCTURE_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_ACCELERATION_STRUCTURE_EXTENSION_NAME)) { *next = &EnabledFeatures.AccelerationStructure; next = &EnabledFeatures.AccelerationStructure.pNext; } - if (SupportsDeviceExtension(VK_KHR_RAY_QUERY_EXTENSION_NAME)) + if (SupportsExtension(VK_KHR_RAY_QUERY_EXTENSION_NAME)) { *next = &EnabledFeatures.RayQuery; next = &EnabledFeatures.RayQuery.pNext; } - if (SupportsDeviceExtension(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME)) + if (SupportsExtension(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME)) { *next = &EnabledFeatures.DescriptorIndexing; next = &EnabledFeatures.DescriptorIndexing.pNext; diff --git a/thirdparty/ZVulkan/src/vulkaninstance.cpp b/thirdparty/ZVulkan/src/vulkaninstance.cpp index 647bcab..939dc73 100644 --- a/thirdparty/ZVulkan/src/vulkaninstance.cpp +++ b/thirdparty/ZVulkan/src/vulkaninstance.cpp @@ -6,7 +6,8 @@ #include #include -VulkanInstance::VulkanInstance(bool wantDebugLayer) : WantDebugLayer(wantDebugLayer) +VulkanInstance::VulkanInstance(std::vector apiVersionsToTry, std::set requiredExtensions, std::set optionalExtensions, bool wantDebugLayer) + : ApiVersionsToTry(std::move(apiVersionsToTry)), RequiredExtensions(std::move(requiredExtensions)), OptionalExtensions(std::move(optionalExtensions)), WantDebugLayer(wantDebugLayer) { try { @@ -64,8 +65,8 @@ void VulkanInstance::CreateInstance() { if (layer.layerName == debugLayer) { - EnabledValidationLayers.push_back(layer.layerName); - EnabledExtensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME); + EnabledValidationLayers.insert(layer.layerName); + EnabledExtensions.insert(VK_EXT_DEBUG_UTILS_EXTENSION_NAME); debugLayerFound = true; break; } @@ -75,15 +76,20 @@ void VulkanInstance::CreateInstance() // Enable optional instance extensions we are interested in for (const auto& ext : AvailableExtensions) { - for (const auto& opt : OptionalExtensions) + if (OptionalExtensions.find(ext.extensionName) != OptionalExtensions.end()) { - if (strcmp(ext.extensionName, opt) == 0) - { - EnabledExtensions.push_back(opt); - } + EnabledExtensions.insert(ext.extensionName); } } + std::vector enabledValidationLayersCStr; + for (const std::string& layer : EnabledValidationLayers) + enabledValidationLayersCStr.push_back(layer.c_str()); + + std::vector enabledExtensionsCStr; + for (const std::string& ext : EnabledExtensions) + enabledExtensionsCStr.push_back(ext.c_str()); + // Try get the highest vulkan version we can get VkResult result = VK_ERROR_INITIALIZATION_FAILED; for (uint32_t apiVersion : ApiVersionsToTry) @@ -100,9 +106,9 @@ void VulkanInstance::CreateInstance() createInfo.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO; createInfo.pApplicationInfo = &appInfo; createInfo.enabledExtensionCount = (uint32_t)EnabledExtensions.size(); - createInfo.enabledLayerCount = (uint32_t)EnabledValidationLayers.size(); - createInfo.ppEnabledLayerNames = EnabledValidationLayers.data(); - createInfo.ppEnabledExtensionNames = EnabledExtensions.data(); + createInfo.enabledLayerCount = (uint32_t)enabledValidationLayersCStr.size(); + createInfo.ppEnabledLayerNames = enabledValidationLayersCStr.data(); + createInfo.ppEnabledExtensionNames = enabledExtensionsCStr.data(); result = vkCreateInstance(&createInfo, nullptr, &Instance); if (result >= VK_SUCCESS) @@ -157,9 +163,6 @@ std::vector VulkanInstance::GetPhysicalDevices(VkInstance auto& dev = devinfo[i]; dev.Device = devices[i]; - vkGetPhysicalDeviceMemoryProperties(dev.Device, &dev.MemoryProperties); - vkGetPhysicalDeviceProperties(dev.Device, &dev.Properties); - uint32_t queueFamilyCount = 0; vkGetPhysicalDeviceQueueFamilyProperties(dev.Device, &queueFamilyCount, nullptr); dev.QueueFamilies.resize(queueFamilyCount); @@ -180,11 +183,32 @@ std::vector VulkanInstance::GetPhysicalDevices(VkInstance return false; }; + vkGetPhysicalDeviceMemoryProperties(dev.Device, &dev.Properties.Memory); + if (apiVersion != VK_API_VERSION_1_0) { + VkPhysicalDeviceProperties2 deviceProperties2 = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2 }; + + void** next = const_cast(&deviceProperties2.pNext); + if (checkForExtension(VK_KHR_ACCELERATION_STRUCTURE_EXTENSION_NAME)) + { + *next = &dev.Properties.AccelerationStructure; + next = &dev.Properties.AccelerationStructure.pNext; + } + if (checkForExtension(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME)) + { + *next = &dev.Properties.DescriptorIndexing; + next = &dev.Properties.DescriptorIndexing.pNext; + } + + vkGetPhysicalDeviceProperties2(dev.Device, &deviceProperties2); + dev.Properties.Properties = deviceProperties2.properties; + dev.Properties.AccelerationStructure.pNext = nullptr; + dev.Properties.DescriptorIndexing.pNext = nullptr; + VkPhysicalDeviceFeatures2 deviceFeatures2 = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2 }; - void** next = const_cast(&deviceFeatures2.pNext); + next = const_cast(&deviceFeatures2.pNext); if (checkForExtension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) { *next = &dev.Features.BufferDeviceAddress; @@ -215,6 +239,7 @@ std::vector VulkanInstance::GetPhysicalDevices(VkInstance } else { + vkGetPhysicalDeviceProperties(dev.Device, &dev.Properties.Properties); vkGetPhysicalDeviceFeatures(dev.Device, &dev.Features.Features); } } diff --git a/thirdparty/ZVulkan/src/vulkansurface.cpp b/thirdparty/ZVulkan/src/vulkansurface.cpp index 848cbcf..ae6ede1 100644 --- a/thirdparty/ZVulkan/src/vulkansurface.cpp +++ b/thirdparty/ZVulkan/src/vulkansurface.cpp @@ -2,6 +2,15 @@ #include "vulkansurface.h" #include "vulkaninstance.h" +VulkanSurface::VulkanSurface(std::shared_ptr instance, VkSurfaceKHR surface) : Instance(std::move(instance)), Surface(surface) +{ +} + +VulkanSurface::~VulkanSurface() +{ + vkDestroySurfaceKHR(Instance->Instance, Surface, nullptr); +} + #ifdef VK_USE_PLATFORM_WIN32_KHR VulkanSurface::VulkanSurface(std::shared_ptr instance, HWND window) : Instance(std::move(instance)), Window(window) @@ -15,39 +24,4 @@ VulkanSurface::VulkanSurface(std::shared_ptr instance, HWND wind VulkanError("Could not create vulkan surface"); } -VulkanSurface::~VulkanSurface() -{ - vkDestroySurfaceKHR(Instance->Instance, Surface, nullptr); -} - -#elif defined(VK_USE_PLATFORM_XLIB_KHR) - -VulkanSurface::VulkanSurface(std::shared_ptr instance, Display* disp, Window wind) : Instance(std::move(instance)), disp(disp), wind(wind) -{ - VkXlibSurfaceCreateInfoKHR createInfo = { VK_STRUCTURE_TYPE_XLIB_SURFACE_CREATE_INFO_KHR }; - createInfo.dpy = disp; - createInfo.window = wind; - - VkResult result = vkCreateXlibSurfaceKHR(Instance->Instance, &createInfo, nullptr, &Surface); - if (result != VK_SUCCESS) - VulkanError("Could not create vulkan surface"); -} - -VulkanSurface::~VulkanSurface() -{ - vkDestroySurfaceKHR(Instance->Instance, Surface, nullptr); -} - -#else - -VulkanSurface::VulkanSurface(std::shared_ptr instance) : Instance(std::move(instance)) -{ - VulkanError("VulkanSurface not implemented on this platform"); -} - -VulkanSurface::~VulkanSurface() -{ - vkDestroySurfaceKHR(Instance->Instance, Surface, nullptr); -} - #endif diff --git a/thirdparty/ZVulkan/src/vulkanswapchain.cpp b/thirdparty/ZVulkan/src/vulkanswapchain.cpp index 982e44f..495c684 100644 --- a/thirdparty/ZVulkan/src/vulkanswapchain.cpp +++ b/thirdparty/ZVulkan/src/vulkanswapchain.cpp @@ -2,205 +2,163 @@ #include "vulkanswapchain.h" #include "vulkanobjects.h" #include "vulkansurface.h" +#include "vulkanbuilders.h" -VulkanSwapChain::VulkanSwapChain(VulkanDevice *device, bool vsync) : vsync(vsync), device(device) +VulkanSwapChain::VulkanSwapChain(VulkanDevice* device) : device(device) { } VulkanSwapChain::~VulkanSwapChain() { - releaseResources(); + views.clear(); + images.clear(); + if (swapchain) + vkDestroySwapchainKHR(device->device, swapchain, nullptr); } -uint32_t VulkanSwapChain::acquireImage(int width, int height, bool fullscreen, VulkanSemaphore *semaphore, VulkanFence *fence) +void VulkanSwapChain::Create(int width, int height, int imageCount, bool vsync, bool hdr, bool exclusivefullscreen) { - if (width <= 0 || height <= 0) - return 0xffffffff; + views.clear(); + images.clear(); - if (lastSwapWidth != width || lastSwapHeight != height || !swapChain || lastFullscreen != fullscreen) + CreateSwapchain(width, height, imageCount, vsync, hdr, exclusivefullscreen); + + if (exclusivefullscreen && lost) { - recreate(width, height, fullscreen); - lastSwapWidth = width; - lastSwapHeight = height; - lastFullscreen = fullscreen; + // We could not acquire exclusive fullscreen. Fall back to normal fullsceen instead. + CreateSwapchain(width, height, imageCount, vsync, hdr, false); } - uint32_t imageIndex; - while (true) + if (swapchain) { - if (!swapChain) - { - imageIndex = 0xffffffff; - break; - } - - VkResult result = vkAcquireNextImageKHR(device->device, swapChain, 1'000'000'000, semaphore ? semaphore->semaphore : VK_NULL_HANDLE, fence ? fence->fence : VK_NULL_HANDLE, &imageIndex); - if (result == VK_SUCCESS) - { - break; - } - else if (result == VK_SUBOPTIMAL_KHR) - { - lastSwapWidth = 0; - lastSwapHeight = 0; - lastFullscreen = false; - break; - } - else if (result == VK_ERROR_OUT_OF_DATE_KHR) - { - recreate(width, height, fullscreen); - } - else if (result == VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT) - { -#ifdef WIN32 - if (GetForegroundWindow() == device->Surface->Window) - { - vkAcquireFullScreenExclusiveModeEXT(device->device, swapChain); - } -#endif - imageIndex = 0xffffffff; - break; - } - else if (result == VK_NOT_READY || result == VK_TIMEOUT) - { - imageIndex = 0xffffffff; - break; - } - else - { - throw std::runtime_error("Failed to acquire next image!"); - } - } - return imageIndex; -} - -void VulkanSwapChain::queuePresent(uint32_t imageIndex, VulkanSemaphore *semaphore) -{ - VkPresentInfoKHR presentInfo = {}; - presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR; - presentInfo.waitSemaphoreCount = semaphore ? 1 : 0; - presentInfo.pWaitSemaphores = semaphore ? &semaphore->semaphore : VK_NULL_HANDLE; - presentInfo.swapchainCount = 1; - presentInfo.pSwapchains = &swapChain; - presentInfo.pImageIndices = &imageIndex; - presentInfo.pResults = nullptr; - VkResult result = vkQueuePresentKHR(device->PresentQueue, &presentInfo); - if (result == VK_SUBOPTIMAL_KHR || result == VK_ERROR_OUT_OF_DATE_KHR) - { - lastSwapWidth = 0; - lastSwapHeight = 0; - lastFullscreen = false; - } - else if (result == VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT) - { -#ifdef WIN32 - if (GetForegroundWindow() == device->Surface->Window) - { - vkAcquireFullScreenExclusiveModeEXT(device->device, swapChain); - } -#endif - } - else if (result == VK_ERROR_OUT_OF_HOST_MEMORY || result == VK_ERROR_OUT_OF_DEVICE_MEMORY) - { - // The spec says we can recover from this. - // However, if we are out of memory it is better to crash now than in some other weird place further away from the source of the problem. - - throw std::runtime_error("vkQueuePresentKHR failed: out of memory"); - } - else if (result == VK_ERROR_DEVICE_LOST) - { - throw std::runtime_error("vkQueuePresentKHR failed: device lost"); - } - else if (result == VK_ERROR_SURFACE_LOST_KHR) - { - throw std::runtime_error("vkQueuePresentKHR failed: surface lost"); - } - else if (result != VK_SUCCESS) - { - throw std::runtime_error("vkQueuePresentKHR failed"); - } -} - -void VulkanSwapChain::recreate(int width, int height, bool fullscreen) -{ - newSwapChain = true; - - releaseViews(); - swapChainImages.clear(); - - selectFormat(); - selectPresentMode(fullscreen); - - VkSwapchainKHR oldSwapChain = swapChain; - createSwapChain(width, height, fullscreen, oldSwapChain); - if (oldSwapChain) - vkDestroySwapchainKHR(device->device, oldSwapChain, nullptr); - - if (swapChain) - { - getImages(); - createViews(); - } -} - -bool VulkanSwapChain::createSwapChain(int width, int height, bool fullscreen, VkSwapchainKHR oldSwapChain) -{ - VkSurfaceCapabilitiesKHR surfaceCapabilities; -#ifdef WIN32 - bool exclusivefullscreen = false; - if (fullscreen && device->SupportsDeviceExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) - { - VkPhysicalDeviceSurfaceInfo2KHR info = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SURFACE_INFO_2_KHR }; - VkSurfaceFullScreenExclusiveInfoEXT exclusiveInfo = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_INFO_EXT }; - VkSurfaceFullScreenExclusiveWin32InfoEXT exclusiveWin32Info = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_WIN32_INFO_EXT }; - info.surface = device->Surface->Surface; - info.pNext = &exclusiveInfo; - exclusiveInfo.fullScreenExclusive = VK_FULL_SCREEN_EXCLUSIVE_APPLICATION_CONTROLLED_EXT; - exclusiveInfo.pNext = &exclusiveWin32Info; - exclusiveWin32Info.hmonitor = MonitorFromWindow(device->Surface->Window, MONITOR_DEFAULTTONEAREST); - - VkSurfaceCapabilities2KHR capabilites = { VK_STRUCTURE_TYPE_SURFACE_CAPABILITIES_2_KHR }; - VkSurfaceCapabilitiesFullScreenExclusiveEXT exclusiveCapabilities = { VK_STRUCTURE_TYPE_SURFACE_CAPABILITIES_FULL_SCREEN_EXCLUSIVE_EXT }; - capabilites.pNext = &exclusiveCapabilities; - - VkResult result = vkGetPhysicalDeviceSurfaceCapabilities2KHR(device->PhysicalDevice.Device, &info, &capabilites); + uint32_t imageCount; + VkResult result = vkGetSwapchainImagesKHR(device->device, swapchain, &imageCount, nullptr); if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfaceCapabilities2EXT failed"); + throw std::runtime_error("vkGetSwapchainImagesKHR failed"); - surfaceCapabilities = capabilites.surfaceCapabilities; - exclusivefullscreen = exclusiveCapabilities.fullScreenExclusiveSupported == VK_TRUE; + std::vector swapchainImages; + swapchainImages.resize(imageCount); + result = vkGetSwapchainImagesKHR(device->device, swapchain, &imageCount, swapchainImages.data()); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetSwapchainImagesKHR failed (2)"); - exclusivefullscreen = false; // Force it off for now since it doesn't work when vsync is false for some reason + for (VkImage vkimage : swapchainImages) + { + auto image = std::make_unique(device, vkimage, nullptr, actualExtent.width, actualExtent.height, 1, 1); + auto view = ImageViewBuilder() + .Type(VK_IMAGE_VIEW_TYPE_2D) + .Image(image.get(), format.format) + .DebugName("SwapchainImageView") + .Create(device); + images.push_back(std::move(image)); + views.push_back(std::move(view)); + } + } +} + +void VulkanSwapChain::SelectFormat(const VulkanSurfaceCapabilities& caps, bool hdr) +{ + if (caps.Formats.size() == 1 && caps.Formats.front().format == VK_FORMAT_UNDEFINED) + { + format.format = VK_FORMAT_B8G8R8A8_UNORM; + format.colorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR; + return; + } + + if (hdr) + { + for (const auto& f : caps.Formats) + { + if (f.format == VK_FORMAT_R16G16B16A16_SFLOAT && f.colorSpace == VK_COLOR_SPACE_EXTENDED_SRGB_LINEAR_EXT) + { + format = f; + return; + } + } + } + + for (const auto& f : caps.Formats) + { + if (f.format == VK_FORMAT_B8G8R8A8_UNORM && f.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR) + { + format = f; + return; + } + } + + format = caps.Formats.front(); +} + +bool VulkanSwapChain::CreateSwapchain(int width, int height, int imageCount, bool vsync, bool hdr, bool exclusivefullscreen) +{ + lost = false; + + VulkanSurfaceCapabilities caps = GetSurfaceCapabilities(exclusivefullscreen); + + if (exclusivefullscreen && (caps.PresentModes.empty() || !caps.FullScreenExclusive.fullScreenExclusiveSupported)) + { + // Try again without exclusive full screen. + exclusivefullscreen = false; + caps = GetSurfaceCapabilities(exclusivefullscreen); + } + + if (caps.PresentModes.empty()) + throw std::runtime_error("No surface present modes supported"); + + bool supportsFifoRelaxed = std::find(caps.PresentModes.begin(), caps.PresentModes.end(), VK_PRESENT_MODE_FIFO_RELAXED_KHR) != caps.PresentModes.end(); + bool supportsMailbox = std::find(caps.PresentModes.begin(), caps.PresentModes.end(), VK_PRESENT_MODE_MAILBOX_KHR) != caps.PresentModes.end(); + bool supportsImmediate = std::find(caps.PresentModes.begin(), caps.PresentModes.end(), VK_PRESENT_MODE_IMMEDIATE_KHR) != caps.PresentModes.end(); + + presentMode = VK_PRESENT_MODE_FIFO_KHR; + if (vsync) + { + if (supportsFifoRelaxed) + presentMode = VK_PRESENT_MODE_FIFO_RELAXED_KHR; } else { - VkResult result = vkGetPhysicalDeviceSurfaceCapabilitiesKHR(device->PhysicalDevice.Device, device->Surface->Surface, &surfaceCapabilities); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfaceCapabilitiesKHR failed"); - fullscreen = false; + if (exclusivefullscreen) // Exclusive full screen doesn't seem to support mailbox for some reason, even if it is advertised + { + if (supportsImmediate) + presentMode = VK_PRESENT_MODE_IMMEDIATE_KHR; + else if (supportsMailbox) + presentMode = VK_PRESENT_MODE_MAILBOX_KHR; + } + else + { + if (supportsMailbox) + presentMode = VK_PRESENT_MODE_MAILBOX_KHR; + else if (supportsImmediate) + presentMode = VK_PRESENT_MODE_IMMEDIATE_KHR; + } } -#endif + + SelectFormat(caps, hdr); actualExtent = { static_cast(width), static_cast(height) }; - actualExtent.width = std::max(surfaceCapabilities.minImageExtent.width, std::min(surfaceCapabilities.maxImageExtent.width, actualExtent.width)); - actualExtent.height = std::max(surfaceCapabilities.minImageExtent.height, std::min(surfaceCapabilities.maxImageExtent.height, actualExtent.height)); + actualExtent.width = std::max(caps.Capabilites.minImageExtent.width, std::min(caps.Capabilites.maxImageExtent.width, actualExtent.width)); + actualExtent.height = std::max(caps.Capabilites.minImageExtent.height, std::min(caps.Capabilites.maxImageExtent.height, actualExtent.height)); if (actualExtent.width == 0 || actualExtent.height == 0) { - swapChain = VK_NULL_HANDLE; + swapchain = VK_NULL_HANDLE; + lost = true; return false; } - uint32_t imageCount = surfaceCapabilities.minImageCount + 1; - if (surfaceCapabilities.maxImageCount > 0 && imageCount > surfaceCapabilities.maxImageCount) - imageCount = surfaceCapabilities.maxImageCount; - - imageCount = std::min(imageCount, (uint32_t)2); // Only use two buffers (triple buffering sucks! good for benchmarks, bad for mouse input) + if (caps.Capabilites.maxImageCount != 0) + imageCount = std::min(caps.Capabilites.maxImageCount, (uint32_t)imageCount); + imageCount = std::max(caps.Capabilites.minImageCount, (uint32_t)imageCount); VkSwapchainCreateInfoKHR swapChainCreateInfo = { VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR }; +#ifdef WIN32 + VkSurfaceFullScreenExclusiveInfoEXT exclusiveInfo = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_INFO_EXT }; + VkSurfaceFullScreenExclusiveWin32InfoEXT exclusiveWin32Info = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_WIN32_INFO_EXT }; +#endif + swapChainCreateInfo.surface = device->Surface->Surface; swapChainCreateInfo.minImageCount = imageCount; - swapChainCreateInfo.imageFormat = swapChainFormat.format; - swapChainCreateInfo.imageColorSpace = swapChainFormat.colorSpace; + swapChainCreateInfo.imageFormat = format.format; + swapChainCreateInfo.imageColorSpace = format.colorSpace; swapChainCreateInfo.imageExtent = actualExtent; swapChainCreateInfo.imageArrayLayers = 1; swapChainCreateInfo.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; @@ -219,188 +177,189 @@ bool VulkanSwapChain::createSwapChain(int width, int height, bool fullscreen, Vk swapChainCreateInfo.pQueueFamilyIndices = nullptr; } - swapChainCreateInfo.preTransform = surfaceCapabilities.currentTransform; + swapChainCreateInfo.preTransform = caps.Capabilites.currentTransform; swapChainCreateInfo.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR; // If alpha channel is passed on to the DWM or not - swapChainCreateInfo.presentMode = swapChainPresentMode; - swapChainCreateInfo.clipped = VK_FALSE;// VK_TRUE; - swapChainCreateInfo.oldSwapchain = oldSwapChain; + swapChainCreateInfo.presentMode = presentMode; + swapChainCreateInfo.clipped = VK_TRUE; // Applications SHOULD set this value to VK_TRUE if they do not expect to read back the content of presentable images before presenting them or after reacquiring them, and if their fragment shaders do not have any side effects that require them to run for all pixels in the presentable image + swapChainCreateInfo.oldSwapchain = swapchain; #ifdef WIN32 - VkSurfaceFullScreenExclusiveInfoEXT exclusiveInfo = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_INFO_EXT }; - VkSurfaceFullScreenExclusiveWin32InfoEXT exclusiveWin32Info = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_WIN32_INFO_EXT }; - if (exclusivefullscreen && device->SupportsDeviceExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + if (exclusivefullscreen) { swapChainCreateInfo.pNext = &exclusiveInfo; - exclusiveInfo.fullScreenExclusive = VK_FULL_SCREEN_EXCLUSIVE_APPLICATION_CONTROLLED_EXT; exclusiveInfo.pNext = &exclusiveWin32Info; + exclusiveInfo.fullScreenExclusive = VK_FULL_SCREEN_EXCLUSIVE_APPLICATION_CONTROLLED_EXT; exclusiveWin32Info.hmonitor = MonitorFromWindow(device->Surface->Window, MONITOR_DEFAULTTONEAREST); } #endif - VkResult result = vkCreateSwapchainKHR(device->device, &swapChainCreateInfo, nullptr, &swapChain); + VkResult result = vkCreateSwapchainKHR(device->device, &swapChainCreateInfo, nullptr, &swapchain); + + if (swapChainCreateInfo.oldSwapchain) + vkDestroySwapchainKHR(device->device, swapChainCreateInfo.oldSwapchain, nullptr); + if (result != VK_SUCCESS) { - swapChain = VK_NULL_HANDLE; + swapchain = VK_NULL_HANDLE; + lost = true; return false; } #ifdef WIN32 - if (exclusivefullscreen && device->SupportsDeviceExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + if (exclusivefullscreen) { - vkAcquireFullScreenExclusiveModeEXT(device->device, swapChain); + result = vkAcquireFullScreenExclusiveModeEXT(device->device, swapchain); + if (result != VK_SUCCESS) + { + lost = true; + } } #endif return true; } -void VulkanSwapChain::createViews() +int VulkanSwapChain::AcquireImage(VulkanSemaphore* semaphore, VulkanFence* fence) { - swapChainImageViews.reserve(swapChainImages.size()); - for (size_t i = 0; i < swapChainImages.size(); i++) + if (lost) + return -1; + + uint32_t imageIndex; + VkResult result = vkAcquireNextImageKHR(device->device, swapchain, 1'000'000'000, semaphore ? semaphore->semaphore : VK_NULL_HANDLE, fence ? fence->fence : VK_NULL_HANDLE, &imageIndex); + if (result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR) { - device->SetObjectName("SwapChainImage", (uint64_t)swapChainImages[i], VK_OBJECT_TYPE_IMAGE); - - VkImageViewCreateInfo createInfo = {}; - createInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; - createInfo.image = swapChainImages[i]; - createInfo.viewType = VK_IMAGE_VIEW_TYPE_2D; - createInfo.format = swapChainFormat.format; - createInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; - createInfo.subresourceRange.baseMipLevel = 0; - createInfo.subresourceRange.levelCount = 1; - createInfo.subresourceRange.baseArrayLayer = 0; - createInfo.subresourceRange.layerCount = 1; - - VkImageView view; - VkResult result = vkCreateImageView(device->device, &createInfo, nullptr, &view); - if (result != VK_SUCCESS) - throw std::runtime_error("Could not create image view for swapchain image"); - - device->SetObjectName("SwapChainImageView", (uint64_t)view, VK_OBJECT_TYPE_IMAGE_VIEW); - - swapChainImageViews.push_back(view); + return imageIndex; } -} - -void VulkanSwapChain::selectFormat() -{ - std::vector surfaceFormats = getSurfaceFormats(); - if (surfaceFormats.empty()) - throw std::runtime_error("No surface formats supported"); - - if (surfaceFormats.size() == 1 && surfaceFormats.front().format == VK_FORMAT_UNDEFINED) + else if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT) { - swapChainFormat.format = VK_FORMAT_B8G8R8A8_UNORM; - swapChainFormat.colorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR; - return; + lost = true; + return -1; } - - for (const auto &format : surfaceFormats) + else if (result == VK_NOT_READY || result == VK_TIMEOUT) { - if (format.format == VK_FORMAT_B8G8R8A8_UNORM && format.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR) - { - swapChainFormat = format; - return; - } - } - - swapChainFormat = surfaceFormats.front(); -} - -void VulkanSwapChain::selectPresentMode(bool fullscreen) -{ - std::vector presentModes = getPresentModes(fullscreen); - - if (presentModes.empty()) - throw std::runtime_error("No surface present modes supported"); - - swapChainPresentMode = VK_PRESENT_MODE_FIFO_KHR; - if (vsync) - { - bool supportsFifoRelaxed = std::find(presentModes.begin(), presentModes.end(), VK_PRESENT_MODE_FIFO_RELAXED_KHR) != presentModes.end(); - if (supportsFifoRelaxed) - swapChainPresentMode = VK_PRESENT_MODE_FIFO_RELAXED_KHR; + return -1; } else { - bool supportsMailbox = std::find(presentModes.begin(), presentModes.end(), VK_PRESENT_MODE_MAILBOX_KHR) != presentModes.end(); - bool supportsImmediate = std::find(presentModes.begin(), presentModes.end(), VK_PRESENT_MODE_IMMEDIATE_KHR) != presentModes.end(); - if (supportsMailbox) - swapChainPresentMode = VK_PRESENT_MODE_MAILBOX_KHR; - else if (supportsImmediate) - swapChainPresentMode = VK_PRESENT_MODE_IMMEDIATE_KHR; + throw std::runtime_error("Failed to acquire next image!"); } } -void VulkanSwapChain::getImages() +void VulkanSwapChain::QueuePresent(int imageIndex, VulkanSemaphore* semaphore) { - uint32_t imageCount; - VkResult result = vkGetSwapchainImagesKHR(device->device, swapChain, &imageCount, nullptr); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetSwapchainImagesKHR failed"); - - swapChainImages.resize(imageCount); - result = vkGetSwapchainImagesKHR(device->device, swapChain, &imageCount, swapChainImages.data()); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetSwapchainImagesKHR failed (2)"); -} - -void VulkanSwapChain::releaseViews() -{ - for (auto &view : swapChainImageViews) + uint32_t index = imageIndex; + VkPresentInfoKHR presentInfo = {}; + presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR; + presentInfo.waitSemaphoreCount = semaphore ? 1 : 0; + presentInfo.pWaitSemaphores = semaphore ? &semaphore->semaphore : VK_NULL_HANDLE; + presentInfo.swapchainCount = 1; + presentInfo.pSwapchains = &swapchain; + presentInfo.pImageIndices = &index; + presentInfo.pResults = nullptr; + VkResult result = vkQueuePresentKHR(device->PresentQueue, &presentInfo); + if (result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR) { - vkDestroyImageView(device->device, view, nullptr); + return; + } + else if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_ERROR_SURFACE_LOST_KHR || result == VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT) + { + lost = true; + } + else if (result == VK_ERROR_OUT_OF_HOST_MEMORY || result == VK_ERROR_OUT_OF_DEVICE_MEMORY) + { + // The spec says we can recover from this. + // However, if we are out of memory it is better to crash now than in some other weird place further away from the source of the problem. + + throw std::runtime_error("vkQueuePresentKHR failed: out of memory"); + } + else if (result == VK_ERROR_DEVICE_LOST) + { + throw std::runtime_error("vkQueuePresentKHR failed: device lost"); + } + else + { + throw std::runtime_error("vkQueuePresentKHR failed"); } - swapChainImageViews.clear(); } -void VulkanSwapChain::releaseResources() +VulkanSurfaceCapabilities VulkanSwapChain::GetSurfaceCapabilities(bool exclusivefullscreen) { - releaseViews(); - if (swapChain) - vkDestroySwapchainKHR(device->device, swapChain, nullptr); -} + // They sure made it easy to query something that isn't even time critical. Good job guys! -std::vector VulkanSwapChain::getSurfaceFormats() -{ - uint32_t surfaceFormatCount = 0; - VkResult result = vkGetPhysicalDeviceSurfaceFormatsKHR(device->PhysicalDevice.Device, device->Surface->Surface, &surfaceFormatCount, nullptr); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormatsKHR failed"); - else if (surfaceFormatCount == 0) - return {}; + VulkanSurfaceCapabilities caps; - std::vector surfaceFormats(surfaceFormatCount); - result = vkGetPhysicalDeviceSurfaceFormatsKHR(device->PhysicalDevice.Device, device->Surface->Surface, &surfaceFormatCount, surfaceFormats.data()); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormatsKHR failed"); - return surfaceFormats; -} - -std::vector VulkanSwapChain::getPresentModes(bool exclusivefullscreen) -{ + VkPhysicalDeviceSurfaceInfo2KHR surfaceInfo = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SURFACE_INFO_2_KHR }; #ifdef WIN32 - if (exclusivefullscreen && device->SupportsDeviceExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + VkSurfaceFullScreenExclusiveInfoEXT exclusiveInfo = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_INFO_EXT }; + VkSurfaceFullScreenExclusiveWin32InfoEXT exclusiveWin32Info = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_WIN32_INFO_EXT }; +#endif + +#ifdef WIN32 + if (exclusivefullscreen) { - VkPhysicalDeviceSurfaceInfo2KHR surfaceInfo = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SURFACE_INFO_2_KHR }; - VkSurfaceFullScreenExclusiveInfoEXT exclusiveInfo = { VK_STRUCTURE_TYPE_SURFACE_FULL_SCREEN_EXCLUSIVE_INFO_EXT }; - surfaceInfo.surface = device->Surface->Surface; - surfaceInfo.pNext = &exclusiveInfo; exclusiveInfo.fullScreenExclusive = VK_FULL_SCREEN_EXCLUSIVE_APPLICATION_CONTROLLED_EXT; + exclusiveWin32Info.hmonitor = MonitorFromWindow(device->Surface->Window, MONITOR_DEFAULTTONEAREST); + } +#endif + + surfaceInfo.surface = device->Surface->Surface; + + if (device->SupportsExtension(VK_KHR_GET_SURFACE_CAPABILITIES_2_EXTENSION_NAME)) + { + const void** next = &surfaceInfo.pNext; + +#ifdef WIN32 + if (exclusivefullscreen && device->SupportsExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + { + *next = &exclusiveInfo; + next = const_cast(&exclusiveInfo.pNext); + + *next = &exclusiveWin32Info; + next = &exclusiveWin32Info.pNext; + } +#endif + + VkSurfaceCapabilities2KHR caps2 = { VK_STRUCTURE_TYPE_SURFACE_CAPABILITIES_2_KHR }; + next = const_cast(&caps2.pNext); + +#ifdef WIN32 + if (exclusivefullscreen && device->SupportsExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + { + *next = &caps.FullScreenExclusive; + next = const_cast(&caps.FullScreenExclusive.pNext); + } +#endif + + VkResult result = vkGetPhysicalDeviceSurfaceCapabilities2KHR(device->PhysicalDevice.Device, &surfaceInfo, &caps2); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceCapabilities2KHR failed"); + + caps.Capabilites = caps2.surfaceCapabilities; + } + else + { + VkResult result = vkGetPhysicalDeviceSurfaceCapabilitiesKHR(device->PhysicalDevice.Device, device->Surface->Surface, &caps.Capabilites); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceCapabilitiesKHR failed"); + } + +#ifdef WIN32 + if (exclusivefullscreen && device->SupportsExtension(VK_EXT_FULL_SCREEN_EXCLUSIVE_EXTENSION_NAME)) + { + const void** next = &surfaceInfo.pNext; uint32_t presentModeCount = 0; VkResult result = vkGetPhysicalDeviceSurfacePresentModes2EXT(device->PhysicalDevice.Device, &surfaceInfo, &presentModeCount, nullptr); if (result != VK_SUCCESS) throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModes2EXT failed"); - else if (presentModeCount == 0) - return {}; - std::vector presentModes(presentModeCount); - result = vkGetPhysicalDeviceSurfacePresentModes2EXT(device->PhysicalDevice.Device, &surfaceInfo, &presentModeCount, presentModes.data()); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModes2EXT failed"); - return presentModes; + if (presentModeCount > 0) + { + caps.PresentModes.resize(presentModeCount); + result = vkGetPhysicalDeviceSurfacePresentModes2EXT(device->PhysicalDevice.Device, &surfaceInfo, &presentModeCount, caps.PresentModes.data()); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModes2EXT failed"); + } } else #endif @@ -409,13 +368,51 @@ std::vector VulkanSwapChain::getPresentModes(bool exclusiveful VkResult result = vkGetPhysicalDeviceSurfacePresentModesKHR(device->PhysicalDevice.Device, device->Surface->Surface, &presentModeCount, nullptr); if (result != VK_SUCCESS) throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModesKHR failed"); - else if (presentModeCount == 0) - return {}; - std::vector presentModes(presentModeCount); - result = vkGetPhysicalDeviceSurfacePresentModesKHR(device->PhysicalDevice.Device, device->Surface->Surface, &presentModeCount, presentModes.data()); - if (result != VK_SUCCESS) - throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModesKHR failed"); - return presentModes; + if (presentModeCount > 0) + { + caps.PresentModes.resize(presentModeCount); + result = vkGetPhysicalDeviceSurfacePresentModesKHR(device->PhysicalDevice.Device, device->Surface->Surface, &presentModeCount, caps.PresentModes.data()); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfacePresentModesKHR failed"); + } } + + if (device->SupportsExtension(VK_KHR_GET_SURFACE_CAPABILITIES_2_EXTENSION_NAME)) + { + uint32_t surfaceFormatCount = 0; + VkResult result = vkGetPhysicalDeviceSurfaceFormats2KHR(device->PhysicalDevice.Device, &surfaceInfo, &surfaceFormatCount, nullptr); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormats2KHR failed"); + + if (surfaceFormatCount > 0) + { + std::vector formats(surfaceFormatCount, { VK_STRUCTURE_TYPE_SURFACE_FORMAT_2_KHR }); + result = vkGetPhysicalDeviceSurfaceFormats2KHR(device->PhysicalDevice.Device, &surfaceInfo, &surfaceFormatCount, formats.data()); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormats2KHR failed"); + + for (VkSurfaceFormat2KHR& fmt : formats) + caps.Formats.push_back(fmt.surfaceFormat); + } + } + else + { + uint32_t surfaceFormatCount = 0; + VkResult result = vkGetPhysicalDeviceSurfaceFormatsKHR(device->PhysicalDevice.Device, device->Surface->Surface, &surfaceFormatCount, nullptr); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormatsKHR failed"); + + if (surfaceFormatCount > 0) + { + caps.Formats.resize(surfaceFormatCount); + result = vkGetPhysicalDeviceSurfaceFormatsKHR(device->PhysicalDevice.Device, device->Surface->Surface, &surfaceFormatCount, caps.Formats.data()); + if (result != VK_SUCCESS) + throw std::runtime_error("vkGetPhysicalDeviceSurfaceFormatsKHR failed"); + } + } + + caps.FullScreenExclusive.pNext = nullptr; + + return caps; }