commit 93b04b1b2b8e4255b7760a3567550bf8f4772c51 Author: matthew Date: Tue Jul 8 04:13:38 2025 +1000 initial commit diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..378eac2 --- /dev/null +++ b/.gitignore @@ -0,0 +1 @@ +build diff --git a/asset-packer/dub.json b/asset-packer/dub.json new file mode 100644 index 0000000..2b77fc9 --- /dev/null +++ b/asset-packer/dub.json @@ -0,0 +1,15 @@ +{ + "name": "asset-packer", + "targetName": "Packer", + "configurations": [ + { + "name": "default", + "targetType": "executable", + "targetPath": "../build", + "platforms": ["linux", "windows"], + "importPaths": ["./", "../src"], + "sourceFiles-linux": ["../src/util.d", "../build/libxxhash.a"], + "sourcePaths": ["./"], + } + ] +} diff --git a/asset-packer/includes.c b/asset-packer/includes.c new file mode 100644 index 0000000..3e95db3 --- /dev/null +++ b/asset-packer/includes.c @@ -0,0 +1 @@ +#include "../external/xxhash/xxhash.h" diff --git a/asset-packer/main.d b/asset-packer/main.d new file mode 100644 index 0000000..b96563b --- /dev/null +++ b/asset-packer/main.d @@ -0,0 +1,7 @@ +import std.stdio; +import util; + +void main() +{ + Log("Hello, World!"); +} diff --git a/build-vma.sh b/build-vma.sh new file mode 100755 index 0000000..1fa8e03 --- /dev/null +++ b/build-vma.sh @@ -0,0 +1,36 @@ +#!/bin/bash +set -eu + +if [ -x "$(command -v g++)" ]; then cpp_compiler="g++"; c_compiler="gcc"; +elif [ -x "$(command -v clang++)" ]; then cpp_compiler="clang++"; c_compiler="clang"; +else echo "Unable to find c++ cpp_compiler"; exit -1; fi; + +if [ -x "$(command -v mold)" ]; then linker_cmd="-fuse-ld=mold"; +elif [ -x "$(command -v lld)" ]; then linker_cmd="-fuse-ld=lld"; +elif [ -x "$(command -v ld)" ]; then linker_cmd="-fuse-ld=ld"; +else echo "Unable to find c/c++ linker"; exit -1; fi; + +# VMA +src="external/vma/vma.cpp" +flags="-std=c++20 -D_USE_MATH_DEFINES -Iexternal/vma -c -Wno-everything -static" +out="-o" +obj="build/vma.o" +lib="build/libvma.a" + +if ! [ -f build/libvma.a ]; then + $cpp_compiler $flags $src $out $obj + ar rcs $lib $obj + rm $obj +fi + +# XXHASH +src="external/xxhash/xxhash.c" +flags="-std=c99 -Wno-everything -Iexternal/xxhash -c -static" +obj="build/xxhash.o" +lib="build/libxxhash.a" + +if ! [ -f build/libxxhash.a ]; then + $c_compiler $flags $src $out $obj + ar rcs $lib $obj + rm $obj +fi diff --git a/dub.json b/dub.json new file mode 100644 index 0000000..21b9d4d --- /dev/null +++ b/dub.json @@ -0,0 +1,31 @@ +{ + "name": "main", + "targetType": "none", + "subPackages": [ + { + "name": "gears", + "targetType": "executable", + "targetName": "Gears", + "targetPath": "build", + "platforms": ["linux", "windows"], + "sourceFiles-linux": ["build/libvma.a", "build/libxxhash.a"], + "sourceFiles-windows": [], + "importPaths": ["src"], + "sourcePaths": ["src"], + "libs-linux": ["xcb", "X11", "X11-xcb", "vulkan", "stdc++"], + "libs-windows": [], + "preGenerateCommands-linux": ["./build-vma.sh"], + "preGenerateCommands-windows": [], + }, + { + "name": "packer", + "targetType": "executable", + "targetPath": "build", + "targetName": "Packer", + "platforms": ["linux", "windows"], + "importPaths": ["src", "asset-packer"], + "sourceFiles-linux": ["src/util.d", "build/libxxhash.a"], + "sourcePaths": ["asset-packer"], + } + ] +} diff --git a/external/vma/vk_mem_alloc.h b/external/vma/vk_mem_alloc.h new file mode 100644 index 0000000..81b2e35 --- /dev/null +++ b/external/vma/vk_mem_alloc.h @@ -0,0 +1,19112 @@ +// +// Copyright (c) 2017-2025 Advanced Micro Devices, Inc. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. +// + +#ifndef AMD_VULKAN_MEMORY_ALLOCATOR_H +#define AMD_VULKAN_MEMORY_ALLOCATOR_H + +/** \mainpage Vulkan Memory Allocator + +Version 3.2.1 + +Copyright (c) 2017-2025 Advanced Micro Devices, Inc. All rights reserved. \n +License: MIT \n +See also: [product page on GPUOpen](https://gpuopen.com/gaming-product/vulkan-memory-allocator/), +[repository on GitHub](https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator) + + +API documentation divided into groups: [Topics](topics.html) + +General documentation chapters: + +- User guide + - \subpage quick_start + - [Project setup](@ref quick_start_project_setup) + - [Initialization](@ref quick_start_initialization) + - [Resource allocation](@ref quick_start_resource_allocation) + - \subpage choosing_memory_type + - [Usage](@ref choosing_memory_type_usage) + - [Required and preferred flags](@ref choosing_memory_type_required_preferred_flags) + - [Explicit memory types](@ref choosing_memory_type_explicit_memory_types) + - [Custom memory pools](@ref choosing_memory_type_custom_memory_pools) + - [Dedicated allocations](@ref choosing_memory_type_dedicated_allocations) + - \subpage memory_mapping + - [Copy functions](@ref memory_mapping_copy_functions) + - [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) + - \subpage staying_within_budget + - [Querying for budget](@ref staying_within_budget_querying_for_budget) + - [Controlling memory usage](@ref staying_within_budget_controlling_memory_usage) + - \subpage resource_aliasing + - \subpage custom_memory_pools + - [Choosing memory type index](@ref custom_memory_pools_MemTypeIndex) + - [When not to use custom pools](@ref custom_memory_pools_when_not_use) + - [Linear allocation algorithm](@ref linear_algorithm) + - [Free-at-once](@ref linear_algorithm_free_at_once) + - [Stack](@ref linear_algorithm_stack) + - [Double stack](@ref linear_algorithm_double_stack) + - [Ring buffer](@ref linear_algorithm_ring_buffer) + - \subpage defragmentation + - \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) + - [Leak detection features](@ref debugging_memory_usage_leak_detection) + - \subpage other_api_interop +- \subpage usage_patterns + - [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) +- Extension support + - \subpage vk_khr_dedicated_allocation + - \subpage enabling_buffer_device_address + - \subpage vk_ext_memory_priority + - \subpage vk_amd_device_coherent_memory + - \subpage vk_khr_external_memory_win32 +- \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) + +\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 + +#if !defined(VULKAN_H_) +#include +#endif + +#if !defined(VMA_VULKAN_VERSION) + #if defined(VK_VERSION_1_4) + #define VMA_VULKAN_VERSION 1004000 + #elif 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 + extern PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr; + extern PFN_vkGetDeviceProcAddr vkGetDeviceProcAddr; + extern PFN_vkGetPhysicalDeviceProperties vkGetPhysicalDeviceProperties; + extern PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties; + extern PFN_vkAllocateMemory vkAllocateMemory; + extern PFN_vkFreeMemory vkFreeMemory; + extern PFN_vkMapMemory vkMapMemory; + extern PFN_vkUnmapMemory vkUnmapMemory; + extern PFN_vkFlushMappedMemoryRanges vkFlushMappedMemoryRanges; + extern PFN_vkInvalidateMappedMemoryRanges vkInvalidateMappedMemoryRanges; + extern PFN_vkBindBufferMemory vkBindBufferMemory; + extern PFN_vkBindImageMemory vkBindImageMemory; + extern PFN_vkGetBufferMemoryRequirements vkGetBufferMemoryRequirements; + extern PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements; + extern PFN_vkCreateBuffer vkCreateBuffer; + extern PFN_vkDestroyBuffer vkDestroyBuffer; + extern PFN_vkCreateImage vkCreateImage; + extern PFN_vkDestroyImage vkDestroyImage; + extern PFN_vkCmdCopyBuffer vkCmdCopyBuffer; + #if VMA_VULKAN_VERSION >= 1001000 + extern PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2; + extern PFN_vkGetImageMemoryRequirements2 vkGetImageMemoryRequirements2; + extern PFN_vkBindBufferMemory2 vkBindBufferMemory2; + extern PFN_vkBindImageMemory2 vkBindImageMemory2; + extern PFN_vkGetPhysicalDeviceMemoryProperties2 vkGetPhysicalDeviceMemoryProperties2; + #endif // #if VMA_VULKAN_VERSION >= 1001000 +#endif // #if defined(__ANDROID__) && VMA_STATIC_VULKAN_FUNCTIONS && VK_NO_PROTOTYPES + +#if !defined(VMA_DEDICATED_ALLOCATION) + #if VK_KHR_get_memory_requirements2 && VK_KHR_dedicated_allocation + #define VMA_DEDICATED_ALLOCATION 1 + #else + #define VMA_DEDICATED_ALLOCATION 0 + #endif +#endif + +#if !defined(VMA_BIND_MEMORY2) + #if VK_KHR_bind_memory2 + #define VMA_BIND_MEMORY2 1 + #else + #define VMA_BIND_MEMORY2 0 + #endif +#endif + +#if !defined(VMA_MEMORY_BUDGET) + #if VK_EXT_memory_budget && (VK_KHR_get_physical_device_properties2 || VMA_VULKAN_VERSION >= 1001000) + #define VMA_MEMORY_BUDGET 1 + #else + #define VMA_MEMORY_BUDGET 0 + #endif +#endif + +// Defined to 1 when VK_KHR_buffer_device_address device extension or equivalent core Vulkan 1.2 feature is defined in its headers. +#if !defined(VMA_BUFFER_DEVICE_ADDRESS) + #if VK_KHR_buffer_device_address || VMA_VULKAN_VERSION >= 1002000 + #define VMA_BUFFER_DEVICE_ADDRESS 1 + #else + #define VMA_BUFFER_DEVICE_ADDRESS 0 + #endif +#endif + +// Defined to 1 when VK_EXT_memory_priority device extension is defined in Vulkan headers. +#if !defined(VMA_MEMORY_PRIORITY) + #if VK_EXT_memory_priority + #define VMA_MEMORY_PRIORITY 1 + #else + #define VMA_MEMORY_PRIORITY 0 + #endif +#endif + +// Defined to 1 when VK_KHR_maintenance4 device extension is defined in Vulkan headers. +#if !defined(VMA_KHR_MAINTENANCE4) + #if VK_KHR_maintenance4 + #define VMA_KHR_MAINTENANCE4 1 + #else + #define VMA_KHR_MAINTENANCE4 0 + #endif +#endif + +// Defined to 1 when VK_KHR_maintenance5 device extension is defined in Vulkan headers. +#if !defined(VMA_KHR_MAINTENANCE5) + #if VK_KHR_maintenance5 + #define VMA_KHR_MAINTENANCE5 1 + #else + #define VMA_KHR_MAINTENANCE5 0 + #endif +#endif + + +// Defined to 1 when VK_KHR_external_memory device extension is defined in Vulkan headers. +#if !defined(VMA_EXTERNAL_MEMORY) + #if VK_KHR_external_memory + #define VMA_EXTERNAL_MEMORY 1 + #else + #define VMA_EXTERNAL_MEMORY 0 + #endif +#endif + +// Defined to 1 when VK_KHR_external_memory_win32 device extension is defined in Vulkan headers. +#if !defined(VMA_EXTERNAL_MEMORY_WIN32) + #if VK_KHR_external_memory_win32 + #define VMA_EXTERNAL_MEMORY_WIN32 1 + #else + #define VMA_EXTERNAL_MEMORY_WIN32 0 + #endif +#endif + +// Define these macros to decorate all public functions with additional code, +// before and after returned type, appropriately. This may be useful for +// exporting the functions when compiling VMA as a separate library. Example: +// #define VMA_CALL_PRE __declspec(dllexport) +// #define VMA_CALL_POST __cdecl +#ifndef VMA_CALL_PRE + #define VMA_CALL_PRE +#endif +#ifndef VMA_CALL_POST + #define VMA_CALL_POST +#endif + +// Define this macro to decorate pNext pointers with an attribute specifying the Vulkan +// structure that will be extended via the pNext chain. +#ifndef VMA_EXTENDS_VK_STRUCT + #define VMA_EXTENDS_VK_STRUCT(vkStruct) +#endif + +// Define this macro to decorate pointers with an attribute specifying the +// length of the array they point to if they are not null. +// +// The length may be one of +// - The name of another parameter in the argument list where the pointer is declared +// - The name of another member in the struct where the pointer is declared +// - The name of a member of a struct type, meaning the value of that member in +// the context of the call. For example +// VMA_LEN_IF_NOT_NULL("VkPhysicalDeviceMemoryProperties::memoryHeapCount"), +// this means the number of memory heaps available in the device associated +// with the VmaAllocator being dealt with. +#ifndef VMA_LEN_IF_NOT_NULL + #define VMA_LEN_IF_NOT_NULL(len) +#endif + +// The VMA_NULLABLE macro is defined to be _Nullable when compiling with Clang. +// see: https://clang.llvm.org/docs/AttributeReference.html#nullable +#ifndef VMA_NULLABLE + #ifdef __clang__ + #define VMA_NULLABLE _Nullable + #else + #define VMA_NULLABLE + #endif +#endif + +// The VMA_NOT_NULL macro is defined to be _Nonnull when compiling with Clang. +// see: https://clang.llvm.org/docs/AttributeReference.html#nonnull +#ifndef VMA_NOT_NULL + #ifdef __clang__ + #define VMA_NOT_NULL _Nonnull + #else + #define VMA_NOT_NULL + #endif +#endif + +// If non-dispatchable handles are represented as pointers then we can give +// then nullability annotations +#ifndef VMA_NOT_NULL_NON_DISPATCHABLE + #if defined(__LP64__) || defined(_WIN64) || (defined(__x86_64__) && !defined(__ILP32__) ) || defined(_M_X64) || defined(__ia64) || defined (_M_IA64) || defined(__aarch64__) || defined(__powerpc64__) + #define VMA_NOT_NULL_NON_DISPATCHABLE VMA_NOT_NULL + #else + #define VMA_NOT_NULL_NON_DISPATCHABLE + #endif +#endif + +#ifndef VMA_NULLABLE_NON_DISPATCHABLE + #if defined(__LP64__) || defined(_WIN64) || (defined(__x86_64__) && !defined(__ILP32__) ) || defined(_M_X64) || defined(__ia64) || defined (_M_IA64) || defined(__aarch64__) || defined(__powerpc64__) + #define VMA_NULLABLE_NON_DISPATCHABLE VMA_NULLABLE + #else + #define VMA_NULLABLE_NON_DISPATCHABLE + #endif +#endif + +#ifndef VMA_STATS_STRING_ENABLED + #define VMA_STATS_STRING_ENABLED 1 +#endif + +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// +// +// INTERFACE +// +//////////////////////////////////////////////////////////////////////////////// +//////////////////////////////////////////////////////////////////////////////// + +// 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 +@{ +*/ + +/// Flags for created #VmaAllocator. +typedef enum VmaAllocatorCreateFlagBits +{ + /** \brief Allocator and all objects created from it will not be synchronized internally, so you must guarantee they are used from only one thread at a time or synchronized externally by you. + + Using this flag may increase performance because internal mutexes are not used. + */ + VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT = 0x00000001, + /** \brief Enables usage of VK_KHR_dedicated_allocation extension. + + The flag works only if VmaAllocatorCreateInfo::vulkanApiVersion `== VK_API_VERSION_1_0`. + When it is `VK_API_VERSION_1_1`, the flag is ignored because the extension has been promoted to Vulkan 1.1. + + Using this extension will automatically allocate dedicated blocks of memory for + some buffers and images instead of suballocating place for them out of bigger + memory blocks (as if you explicitly used #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT + flag) when it is recommended by the driver. It may improve performance on some + GPUs. + + You may set this flag only if you found out that following device extensions are + supported, you enabled them while creating Vulkan device passed as + VmaAllocatorCreateInfo::device, and you want them to be used internally by this + library: + + - VK_KHR_get_memory_requirements2 (device extension) + - VK_KHR_dedicated_allocation (device extension) + + When this flag is set, you can experience following warnings reported by Vulkan + validation layer. You can ignore them. + + > vkBindBufferMemory(): Binding memory to buffer 0x2d but vkGetBufferMemoryRequirements() has not been called on that buffer. + */ + VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT = 0x00000002, + /** + Enables usage of VK_KHR_bind_memory2 extension. + + The flag works only if VmaAllocatorCreateInfo::vulkanApiVersion `== VK_API_VERSION_1_0`. + When it is `VK_API_VERSION_1_1`, the flag is ignored because the extension has been promoted to Vulkan 1.1. + + You may set this flag only if you found out that this device extension is supported, + you enabled it while creating Vulkan device passed as VmaAllocatorCreateInfo::device, + and you want it to be used internally by this library. + + The extension provides functions `vkBindBufferMemory2KHR` and `vkBindImageMemory2KHR`, + which allow to pass a chain of `pNext` structures while binding. + This flag is required if you use `pNext` parameter in vmaBindBufferMemory2() or vmaBindImageMemory2(). + */ + VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT = 0x00000004, + /** + Enables usage of VK_EXT_memory_budget extension. + + You may set this flag only if you found out that this device extension is supported, + you enabled it while creating Vulkan device passed as VmaAllocatorCreateInfo::device, + and you want it to be used internally by this library, along with another instance extension + VK_KHR_get_physical_device_properties2, which is required by it (or Vulkan 1.1, where this extension is promoted). + + The extension provides query for current memory usage and budget, which will probably + be more accurate than an estimation used by the library otherwise. + */ + VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT = 0x00000008, + /** + Enables usage of VK_AMD_device_coherent_memory extension. + + You may set this flag only if you: + + - found out that this device extension is supported and enabled it while creating Vulkan device passed as VmaAllocatorCreateInfo::device, + - checked that `VkPhysicalDeviceCoherentMemoryFeaturesAMD::deviceCoherentMemory` is true and set it while creating the Vulkan device, + - want it to be used internally by this library. + + The extension and accompanying device feature provide access to memory types with + `VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD` and `VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD` flags. + They are useful mostly for writing breadcrumb markers - a common method for debugging GPU crash/hang/TDR. + + When the extension is not enabled, such memory types are still enumerated, but their usage is illegal. + To protect from this error, if you don't create the allocator with this flag, it will refuse to allocate any memory or create a custom pool in such memory type, + returning `VK_ERROR_FEATURE_NOT_PRESENT`. + */ + VMA_ALLOCATOR_CREATE_AMD_DEVICE_COHERENT_MEMORY_BIT = 0x00000010, + /** + Enables usage of "buffer device address" feature, which allows you to use function + `vkGetBufferDeviceAddress*` to get raw GPU pointer to a buffer and pass it for usage inside a shader. + + You may set this flag only if you: + + 1. (For Vulkan version < 1.2) Found as available and enabled device extension + VK_KHR_buffer_device_address. + This extension is promoted to core Vulkan 1.2. + 2. Found as available and enabled device feature `VkPhysicalDeviceBufferDeviceAddressFeatures::bufferDeviceAddress`. + + When this flag is set, you can create buffers with `VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT` using VMA. + The library automatically adds `VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT` to + allocated memory blocks wherever it might be needed. + + For more information, see documentation chapter \ref enabling_buffer_device_address. + */ + VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT = 0x00000020, + /** + Enables usage of VK_EXT_memory_priority extension in the library. + + You may set this flag only if you found available and enabled this device extension, + along with `VkPhysicalDeviceMemoryPriorityFeaturesEXT::memoryPriority == VK_TRUE`, + while creating Vulkan device passed as VmaAllocatorCreateInfo::device. + + When this flag is used, VmaAllocationCreateInfo::priority and VmaPoolCreateInfo::priority + are used to set priorities of allocated Vulkan memory. Without it, these variables are ignored. + + A priority must be a floating-point value between 0 and 1, indicating the priority of the allocation relative to other memory allocations. + Larger values are higher priority. The granularity of the priorities is implementation-dependent. + It is automatically passed to every call to `vkAllocateMemory` done by the library using structure `VkMemoryPriorityAllocateInfoEXT`. + The value to be used for default priority is 0.5. + For more details, see the documentation of the VK_EXT_memory_priority extension. + */ + VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT = 0x00000040, + /** + Enables usage of VK_KHR_maintenance4 extension in the library. + + You may set this flag only if you found available and enabled this device extension, + while creating Vulkan device passed as VmaAllocatorCreateInfo::device. + */ + VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE4_BIT = 0x00000080, + /** + Enables usage of VK_KHR_maintenance5 extension in the library. + + You should set this flag if you found available and enabled this device extension, + while creating Vulkan device passed as VmaAllocatorCreateInfo::device. + */ + VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT = 0x00000100, + + /** + Enables usage of VK_KHR_external_memory_win32 extension in the library. + + You should set this flag if you found available and enabled this device extension, + while creating Vulkan device passed as VmaAllocatorCreateInfo::device. + For more information, see \ref vk_khr_external_memory_win32. + */ + VMA_ALLOCATOR_CREATE_KHR_EXTERNAL_MEMORY_WIN32_BIT = 0x00000200, + + 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. + + If you use this flag while creating a buffer or an image, `VkMemoryDedicatedAllocateInfo` + structure is applied if possible. + */ + 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 preferred. + */ + 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 recommended 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 +{ + /// 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; + PFN_vkFreeMemory VMA_NULLABLE vkFreeMemory; + PFN_vkMapMemory VMA_NULLABLE vkMapMemory; + PFN_vkUnmapMemory VMA_NULLABLE vkUnmapMemory; + PFN_vkFlushMappedMemoryRanges VMA_NULLABLE vkFlushMappedMemoryRanges; + PFN_vkInvalidateMappedMemoryRanges VMA_NULLABLE vkInvalidateMappedMemoryRanges; + PFN_vkBindBufferMemory VMA_NULLABLE vkBindBufferMemory; + PFN_vkBindImageMemory VMA_NULLABLE vkBindImageMemory; + PFN_vkGetBufferMemoryRequirements VMA_NULLABLE vkGetBufferMemoryRequirements; + PFN_vkGetImageMemoryRequirements VMA_NULLABLE vkGetImageMemoryRequirements; + PFN_vkCreateBuffer VMA_NULLABLE vkCreateBuffer; + PFN_vkDestroyBuffer VMA_NULLABLE vkDestroyBuffer; + PFN_vkCreateImage VMA_NULLABLE vkCreateImage; + 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 + /// Fetch from "vkGetPhysicalDeviceMemoryProperties2" on Vulkan >= 1.1, but you can also fetch it from "vkGetPhysicalDeviceMemoryProperties2KHR" if you enabled extension VK_KHR_get_physical_device_properties2. + PFN_vkGetPhysicalDeviceMemoryProperties2KHR VMA_NULLABLE vkGetPhysicalDeviceMemoryProperties2KHR; +#endif +#if VMA_KHR_MAINTENANCE4 || 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_vkGetDeviceBufferMemoryRequirementsKHR 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_vkGetDeviceImageMemoryRequirementsKHR VMA_NULLABLE vkGetDeviceImageMemoryRequirements; +#endif +#if VMA_EXTERNAL_MEMORY_WIN32 + PFN_vkGetMemoryWin32HandleKHR VMA_NULLABLE vkGetMemoryWin32HandleKHR; +#else + void* VMA_NULLABLE vkGetMemoryWin32HandleKHR; +#endif +} VmaVulkanFunctions; + +/// Description of a Allocator to be created. +typedef struct VmaAllocatorCreateInfo +{ + /// Flags for created allocator. Use #VmaAllocatorCreateFlagBits enum. + VmaAllocatorCreateFlags flags; + /// Vulkan physical device. + /** It must be valid throughout whole lifetime of created allocator. */ + VkPhysicalDevice VMA_NOT_NULL physicalDevice; + /// Vulkan device. + /** It must be valid throughout whole lifetime of created allocator. */ + VkDevice VMA_NOT_NULL device; + /// Preferred size of a single `VkDeviceMemory` block to be allocated from large heaps > 1 GiB. Optional. + /** Set to 0 to use default, which is currently 256 MiB. */ + VkDeviceSize preferredLargeHeapBlockSize; + /// Custom CPU memory allocation callbacks. Optional. + /** Optional, can be null. When specified, will also be used for all CPU-side memory allocations. */ + const VkAllocationCallbacks* VMA_NULLABLE pAllocationCallbacks; + /// Informative callbacks for `vkAllocateMemory`, `vkFreeMemory`. Optional. + /** Optional, can be null. */ + const VmaDeviceMemoryCallbacks* VMA_NULLABLE pDeviceMemoryCallbacks; + /** \brief Either null or a pointer to an array of limits on maximum number of bytes that can be allocated out of particular Vulkan memory heap. + + If not NULL, it must be a pointer to an array of + `VkPhysicalDeviceMemoryProperties::memoryHeapCount` elements, defining limit on + maximum number of bytes that can be allocated out of particular Vulkan memory + heap. + + Any of the elements may be equal to `VK_WHOLE_SIZE`, which means no limit on that + heap. This is also the default in case of `pHeapSizeLimit` = NULL. + + If there is a limit defined for a heap: + + - If user tries to allocate more memory from that heap using this allocator, + the allocation fails with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. + - If the limit is smaller than heap size reported in `VkMemoryHeap::size`, the + value of this limit will be reported instead when using vmaGetMemoryProperties(). + + Warning! Using this feature may not be equivalent to installing a GPU with + smaller amount of memory, because graphics driver doesn't necessary fail new + allocations with `VK_ERROR_OUT_OF_DEVICE_MEMORY` result when memory capacity is + exceeded. It may return success and just silently migrate some device memory + blocks to system RAM. This driver behavior can also be controlled using + VK_AMD_memory_overallocation_behavior extension. + */ + const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL("VkPhysicalDeviceMemoryProperties::memoryHeapCount") pHeapSizeLimit; + + /** \brief Pointers to Vulkan functions. Can be null. + + For details see [Pointers to Vulkan functions](@ref config_Vulkan_functions). + */ + const VmaVulkanFunctions* VMA_NULLABLE pVulkanFunctions; + /** \brief Handle to Vulkan instance object. + + Starting from version 3.0.0 this member is no longer optional, it must be set! + */ + VkInstance VMA_NOT_NULL instance; + /** \brief Optional. Vulkan version that the application uses. + + 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. + Only versions 1.0...1.4 are supported by the current implementation. + Leaving it initialized to zero is equivalent to `VK_API_VERSION_1_0`. + It must match the Vulkan version used by the application and supported on the selected physical device, + so it must be no higher than `VkApplicationInfo::apiVersion` passed to `vkCreateInstance` + and no higher than `VkPhysicalDeviceProperties::apiVersion` found on the physical device used. + */ + uint32_t vulkanApiVersion; +#if VMA_EXTERNAL_MEMORY + /** \brief Either null or a pointer to an array of external memory handle types for each Vulkan memory type. + + If not NULL, it must be a pointer to an array of `VkPhysicalDeviceMemoryProperties::memoryTypeCount` + elements, defining external memory handle types of particular Vulkan memory type, + to be passed using `VkExportMemoryAllocateInfoKHR`. + + Any of the elements may be equal to 0, which means not to use `VkExportMemoryAllocateInfoKHR` on this memory type. + This is also the default in case of `pTypeExternalMemoryHandleTypes` = NULL. + */ + const VkExternalMemoryHandleTypeFlagsKHR* VMA_NULLABLE VMA_LEN_IF_NOT_NULL("VkPhysicalDeviceMemoryProperties::memoryTypeCount") pTypeExternalMemoryHandleTypes; +#endif // #if VMA_EXTERNAL_MEMORY +} VmaAllocatorCreateInfo; + +/// Information about existing #VmaAllocator object. +typedef struct VmaAllocatorInfo +{ + /** \brief Handle to Vulkan instance object. + + This is the same value as has been passed through VmaAllocatorCreateInfo::instance. + */ + VkInstance VMA_NOT_NULL instance; + /** \brief Handle to Vulkan physical device object. + + This is the same value as has been passed through VmaAllocatorCreateInfo::physicalDevice. + */ + VkPhysicalDevice VMA_NOT_NULL physicalDevice; + /** \brief Handle to Vulkan device object. + + This is the same value as has been passed through VmaAllocatorCreateInfo::device. + */ + VkDevice VMA_NOT_NULL device; +} VmaAllocatorInfo; + +/** @} */ + +/** +\addtogroup group_stats +@{ +*/ + +/** \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(). +*/ +typedef struct VmaStatistics +{ + /** \brief Number of `VkDeviceMemory` objects - Vulkan memory blocks allocated. + */ + uint32_t blockCount; + /** \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; + /// 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; + +/** \brief General statistics from current state of the Allocator - +total memory usage across all memory heaps and types. + +These are slower to calculate. Use for debugging purposes. +See function vmaCalculateStatistics(). +*/ +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 for a specific memory heap. + +These are fast to calculate. +See function vmaGetHeapBudgets(). +*/ +typedef struct VmaBudget +{ + /** \brief Statistics fetched from the library. + */ + VmaStatistics statistics; + /** \brief Estimated current memory usage of the program, in bytes. + + Fetched from system using VK_EXT_memory_budget extension if enabled. + + 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. + + It might be different (most probably smaller) than `VkMemoryHeap::size[heapIndex]` due to factors + 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; + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + +/** \brief Parameters of new #VmaAllocation. + +To be used with functions like vmaCreateBuffer(), vmaCreateImage(), and many others. +*/ +typedef struct VmaAllocationCreateInfo +{ + /// Use #VmaAllocationCreateFlagBits enum. + VmaAllocationCreateFlags flags; + /** \brief Intended usage of memory. + + You can leave #VMA_MEMORY_USAGE_UNKNOWN if you specify memory requirements in other way. \n + If `pool` is not null, this member is ignored. + */ + VmaMemoryUsage usage; + /** \brief Flags that must be set in a Memory Type chosen for an allocation. + + Leave 0 if you specify memory requirements in other way. \n + If `pool` is not null, this member is ignored.*/ + VkMemoryPropertyFlags requiredFlags; + /** \brief Flags that preferably should be set in a memory type chosen for an allocation. + + Set to 0 if no additional flags are preferred. \n + If `pool` is not null, this member is ignored. */ + VkMemoryPropertyFlags preferredFlags; + /** \brief Bitmask containing one bit set for every memory type acceptable for this allocation. + + Value 0 is equivalent to `UINT32_MAX` - it means any memory type is accepted if + it meets other requirements specified by this structure, with no further + restrictions on memory type index. \n + If `pool` is not null, this member is ignored. + */ + uint32_t memoryTypeBits; + /** \brief Pool that this allocation should be created in. + + Leave `VK_NULL_HANDLE` to allocate from default pool. If not null, members: + `usage`, `requiredFlags`, `preferredFlags`, `memoryTypeBits` are ignored. + */ + VmaPool VMA_NULLABLE pool; + /** \brief Custom general-purpose pointer that will be stored in #VmaAllocation, can be read as VmaAllocationInfo::pUserData and changed using vmaSetAllocationUserData(). + + If #VMA_ALLOCATION_CREATE_USER_DATA_COPY_STRING_BIT is used, it must be either + null or pointer to a null-terminated string. The string will be then copied to + internal buffer, so it doesn't need to be valid after allocation call. + */ + void* VMA_NULLABLE pUserData; + /** \brief A floating-point value between 0 and 1, indicating the priority of the allocation 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 + and this allocation ends up as dedicated or is explicitly forced as dedicated using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. + Otherwise, it has the priority of a memory block where it is placed and this variable is ignored. + */ + 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 VMA_EXTENDS_VK_STRUCT(VkMemoryAllocateInfo) pMemoryAllocateNext; +} VmaPoolCreateInfo; + +/** @} */ + +/** +\addtogroup group_alloc +@{ +*/ + +/** +Parameters of #VmaAllocation objects, that can be retrieved using function vmaGetAllocationInfo(). + +There is also an extended version of this structure that carries additional parameters: #VmaAllocationInfo2. +*/ +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; + +/// Extended parameters of a #VmaAllocation object that can be retrieved using function vmaGetAllocationInfo2(). +typedef struct VmaAllocationInfo2 +{ + /** \brief Basic parameters of the allocation. + + If you need only these, you can use function vmaGetAllocationInfo() and structure #VmaAllocationInfo instead. + */ + VmaAllocationInfo allocationInfo; + /** \brief Size of the `VkDeviceMemory` block that the allocation belongs to. + + In case of an allocation with dedicated memory, it will be equal to `allocationInfo.size`. + */ + VkDeviceSize blockSize; + /** \brief `VK_TRUE` if the allocation has dedicated memory, `VK_FALSE` if it was placed as part of a larger memory block. + + When `VK_TRUE`, it also means `VkMemoryDedicatedAllocateInfo` was used when creating the allocation + (if VK_KHR_dedicated_allocation extension or Vulkan version >= 1.1 is enabled). + */ + VkBool32 dedicatedMemory; +} VmaAllocationInfo2; + +/** Callback function called during vmaBeginDefragmentation() to check custom criterion about ending current defragmentation pass. + +Should return true if the defragmentation needs to stop current pass. +*/ +typedef VkBool32 (VKAPI_PTR* PFN_vmaCheckDefragmentationBreakFunction)(void* VMA_NULLABLE pUserData); + +/** \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; + /** \brief Optional custom callback for stopping vmaBeginDefragmentation(). + + Have to return true for breaking current defragmentation pass. + */ + PFN_vmaCheckDefragmentationBreakFunction VMA_NULLABLE pfnBreakCallback; + /// \brief Optional data to pass to custom callback for stopping pass of defragmentation. + void* VMA_NULLABLE pBreakCallbackUserData; +} 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. + +This algorithm tries to find a memory type that: + +- Is allowed by memoryTypeBits. +- Contains all the flags from pAllocationCreateInfo->requiredFlags. +- Matches intended usage. +- Has as many flags from pAllocationCreateInfo->preferredFlags as possible. + +\return Returns VK_ERROR_FEATURE_NOT_PRESENT if not found. Receiving such result +from this function or any other allocating function probably means that your +device doesn't support any memory type with requested features for the specific +type of resource you want to use it for. Please check parameters of your +resource, like image layout (OPTIMAL versus LINEAR) or mip level count. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndex( + VmaAllocator VMA_NOT_NULL allocator, + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, + uint32_t* VMA_NOT_NULL pMemoryTypeIndex); + +/** +\brief Helps to find memoryTypeIndex, given VkBufferCreateInfo and VmaAllocationCreateInfo. + +It can be useful e.g. to determine value to be used as VmaPoolCreateInfo::memoryTypeIndex. +It internally creates a temporary, dummy buffer that never has memory bound. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( + VmaAllocator VMA_NOT_NULL allocator, + const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, + const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, + uint32_t* VMA_NOT_NULL pMemoryTypeIndex); + +/** +\brief Helps to find memoryTypeIndex, given VkImageCreateInfo and VmaAllocationCreateInfo. + +It can be useful e.g. to determine value to be used as VmaPoolCreateInfo::memoryTypeIndex. +It internally creates a temporary, dummy image that never has memory bound. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForImageInfo( + VmaAllocator VMA_NOT_NULL allocator, + const VkImageCreateInfo* VMA_NOT_NULL pImageCreateInfo, + const VmaAllocationCreateInfo* VMA_NOT_NULL pAllocationCreateInfo, + uint32_t* VMA_NOT_NULL pMemoryTypeIndex); + +/** \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. +*/ +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); + +/** \brief Destroys #VmaPool object and frees Vulkan device memory. +*/ +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. + +Note that when using the pool from multiple threads, returned information may immediately +become outdated. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStatistics( + VmaAllocator VMA_NOT_NULL allocator, + VmaPool VMA_NOT_NULL pool, + VmaStatistics* VMA_NOT_NULL pPoolStats); + +/** \brief Retrieves detailed statistics of existing #VmaPool object. + +\param allocator Allocator object. +\param pool Pool object. +\param[out] pPoolStats Statistics of specified pool. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaCalculatePoolStatistics( + VmaAllocator VMA_NOT_NULL allocator, + VmaPool VMA_NOT_NULL pool, + 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. + +Corruption detection is enabled only when `VMA_DEBUG_DETECT_CORRUPTION` macro is defined to nonzero, +`VMA_DEBUG_MARGIN` is defined to nonzero and the pool is created in memory type that is +`HOST_VISIBLE` and `HOST_COHERENT`. For more information, see [Corruption detection](@ref debugging_memory_usage_corruption_detection). + +Possible return values: + +- `VK_ERROR_FEATURE_NOT_PRESENT` - corruption detection is not enabled for specified pool. +- `VK_SUCCESS` - corruption detection has been performed and succeeded. +- `VK_ERROR_UNKNOWN` - corruption detection has been performed and found memory corruptions around one of the allocations. + `VMA_ASSERT` is also fired in that case. +- Other value: Error returned by Vulkan, e.g. memory mapping failure. +*/ +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. + +After the call `ppName` is either null or points to an internally-owned null-terminated string +containing name of the pool that was previously set. The pointer becomes invalid when the pool is +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); + +/** \brief Sets name of a custom pool. + +`pName` can be either null or pointer to a null-terminated string with new name for the pool. +Function makes internal copy of the string, so it can be changed or freed immediately after this call. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaSetPoolName( + VmaAllocator VMA_NOT_NULL allocator, + VmaPool VMA_NOT_NULL pool, + const char* VMA_NULLABLE pName); + +/** \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(). + +You should free the memory using vmaFreeMemory() or vmaFreeMemoryPages(). + +It is recommended to use vmaAllocateMemoryForBuffer(), vmaAllocateMemoryForImage(), +vmaCreateBuffer(), vmaCreateImage() instead whenever possible. +*/ +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, + 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 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(). + +Word "pages" is just a suggestion to use this function to allocate pieces of memory needed for sparse binding. +It is just a general purpose allocation function able to make multiple allocations at once. +It may be internally optimized to be more efficient than calling vmaAllocateMemory() `allocationCount` times. + +All allocations are made using same parameters. All of them are created out of the same memory pool and type. +If any allocation fails, all allocations already made within this function call are also freed, so that when +returned result is not `VK_SUCCESS`, `pAllocation` array is always entirely filled with `VK_NULL_HANDLE`. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryPages( + VmaAllocator VMA_NOT_NULL allocator, + 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, + VmaAllocationInfo* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) pAllocationInfo); + +/** \brief Allocates memory suitable for given `VkBuffer`. + +\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, + VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); + +/** \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, + VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); + +/** \brief Frees memory previously allocated using vmaAllocateMemory(), vmaAllocateMemoryForBuffer(), or vmaAllocateMemoryForImage(). + +Passing `VK_NULL_HANDLE` as `allocation` is valid. Such function call is just skipped. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaFreeMemory( + VmaAllocator VMA_NOT_NULL allocator, + const VmaAllocation VMA_NULLABLE allocation); + +/** \brief Frees memory and destroys multiple allocations. + +Word "pages" is just a suggestion to use this function to free pieces of memory used for sparse binding. +It is just a general purpose function to free memory and destroy allocations made using e.g. vmaAllocateMemory(), +vmaAllocateMemoryPages() and other functions. +It may be internally optimized to be more efficient than calling vmaFreeMemory() `allocationCount` times. + +Allocations in `pAllocations` array can come from any memory pools and types. +Passing `VK_NULL_HANDLE` as elements of `pAllocations` array is valid. Such entries are just skipped. +*/ +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); + +/** \brief Returns current information about specified allocation. + +Current parameters of given allocation are returned in `pAllocationInfo`. + +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). + +There is also a new function vmaGetAllocationInfo2() that offers extended information +about the allocation, returned using new structure #VmaAllocationInfo2. +*/ +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 extended information about specified allocation. + +Current parameters of given allocation are returned in `pAllocationInfo`. +Extended parameters in structure #VmaAllocationInfo2 include memory block size +and a flag telling whether the allocation has dedicated memory. +It can be useful e.g. for interop with OpenGL. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationInfo2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VmaAllocationInfo2* VMA_NOT_NULL pAllocationInfo); + +/** \brief Sets pUserData in given allocation to new value. + +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( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + void* VMA_NULLABLE pUserData); + +/** \brief Sets pName in given allocation to new value. + +`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 vmaSetAllocationName( + VmaAllocator VMA_NOT_NULL allocator, + 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); + + +#if VMA_EXTERNAL_MEMORY_WIN32 +/** +\brief Given an allocation, returns Win32 handle that may be imported by other processes or APIs. + +\param hTargetProcess Must be a valid handle to target process or null. If it's null, the function returns + handle for the current process. +\param[out] pHandle Output parameter that returns the handle. + +The function fills `pHandle` with handle that can be used in target process. +The handle is fetched using function `vkGetMemoryWin32HandleKHR`. +When no longer needed, you must close it using: + +\code +CloseHandle(handle); +\endcode + +You can close it any time, before or after destroying the allocation object. +It is reference-counted internally by Windows. + +Note the handle is returned for the entire `VkDeviceMemory` block that the allocation belongs to. +If the allocation is sub-allocated from a larger block, you may need to consider the offset of the allocation +(VmaAllocationInfo::offset). + +If the function fails with `VK_ERROR_FEATURE_NOT_PRESENT` error code, please double-check +that VmaVulkanFunctions::vkGetMemoryWin32HandleKHR function pointer is set, e.g. either by using `VMA_DYNAMIC_VULKAN_FUNCTIONS` +or by manually passing it through VmaAllocatorCreateInfo::pVulkanFunctions. + +For more information, see chapter \ref vk_khr_external_memory_win32. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaGetMemoryWin32Handle(VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, HANDLE hTargetProcess, HANDLE* VMA_NOT_NULL pHandle); +#endif // VMA_EXTERNAL_MEMORY_WIN32 + +/** \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. + +\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` +multiple times simultaneously, it is safe to call this function on allocations +assigned to the same memory block. Actual Vulkan memory will be mapped on first +mapping and unmapped on last unmapping. + +If the function succeeded, you must call vmaUnmapMemory() to unmap the +allocation when mapping is no longer needed or before freeing the allocation, at +the latest. + +It also safe to call this function multiple times on the same allocation. You +must call vmaUnmapMemory() same number of times as you called vmaMapMemory(). + +It is also safe to call this function on allocation created with +#VMA_ALLOCATION_CREATE_MAPPED_BIT flag. Its memory stays mapped all the time. +You must still call vmaUnmapMemory() same number of times as you called +vmaMapMemory(). You must not call vmaUnmapMemory() additional time to free the +"0-th" mapping made automatically due to #VMA_ALLOCATION_CREATE_MAPPED_BIT flag. + +This function fails when used on allocation made in memory type that is not +`HOST_VISIBLE`. + +This function 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. +*/ +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); + +/** \brief Unmaps memory represented by given allocation, mapped previously using vmaMapMemory(). + +For details, see description of vmaMapMemory(). + +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. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaUnmapMemory( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation); + +/** \brief Flushes memory of given allocation. + +Calls `vkFlushMappedMemoryRanges()` for memory associated with given range of given allocation. +It needs to be called after writing to a mapped memory for memory types that are not `HOST_COHERENT`. +Unmap operation doesn't do that automatically. + +- `offset` must be relative to the beginning of allocation. +- `size` can be `VK_WHOLE_SIZE`. It means all memory from `offset` the the end of given allocation. +- `offset` and `size` don't have to be aligned. + They are internally rounded down/up to multiply of `nonCoherentAtomSize`. +- If `size` is 0, this call is ignored. +- If memory type that the `allocation` belongs to is not `HOST_VISIBLE` or it is `HOST_COHERENT`, + this call is ignored. + +Warning! `offset` and `size` are relative to the contents of given `allocation`. +If you mean whole allocation, you can pass 0 and `VK_WHOLE_SIZE`, respectively. +Do not pass allocation's offset as `offset`!!! + +This function returns the `VkResult` from `vkFlushMappedMemoryRanges` if it is +called, otherwise `VK_SUCCESS`. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocation( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize offset, + VkDeviceSize size); + +/** \brief Invalidates memory of given allocation. + +Calls `vkInvalidateMappedMemoryRanges()` for memory associated with given range of given allocation. +It needs to be called before reading from a mapped memory for memory types that are not `HOST_COHERENT`. +Map operation doesn't do that automatically. + +- `offset` must be relative to the beginning of allocation. +- `size` can be `VK_WHOLE_SIZE`. It means all memory from `offset` the the end of given allocation. +- `offset` and `size` don't have to be aligned. + They are internally rounded down/up to multiply of `nonCoherentAtomSize`. +- If `size` is 0, this call is ignored. +- If memory type that the `allocation` belongs to is not `HOST_VISIBLE` or it is `HOST_COHERENT`, + this call is ignored. + +Warning! `offset` and `size` are relative to the contents of given `allocation`. +If you mean whole allocation, you can pass 0 and `VK_WHOLE_SIZE`, respectively. +Do not pass allocation's offset as `offset`!!! + +This function returns the `VkResult` from `vkInvalidateMappedMemoryRanges` if +it is called, otherwise `VK_SUCCESS`. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocation( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize offset, + VkDeviceSize size); + +/** \brief Flushes memory of given set of allocations. + +Calls `vkFlushMappedMemoryRanges()` for memory associated with given ranges of given allocations. +For more information, see documentation of vmaFlushAllocation(). + +\param allocator +\param allocationCount +\param allocations +\param offsets If not null, it must point to an array of offsets of regions to flush, relative to the beginning of respective allocations. Null means all offsets are zero. +\param sizes If not null, it must point to an array of sizes of regions to flush in respective allocations. Null means `VK_WHOLE_SIZE` for all allocations. + +This function returns the `VkResult` from `vkFlushMappedMemoryRanges` if it is +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 VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) offsets, + const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) sizes); + +/** \brief Invalidates memory of given set of allocations. + +Calls `vkInvalidateMappedMemoryRanges()` for memory associated with given ranges of given allocations. +For more information, see documentation of vmaInvalidateAllocation(). + +\param allocator +\param allocationCount +\param allocations +\param offsets If not null, it must point to an array of offsets of regions to flush, relative to the beginning of respective allocations. Null means all offsets are zero. +\param sizes If not null, it must point to an array of sizes of regions to flush in respective allocations. Null means `VK_WHOLE_SIZE` for all allocations. + +This function returns the `VkResult` from `vkInvalidateMappedMemoryRanges` if it is +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 VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) offsets, + const VkDeviceSize* VMA_NULLABLE VMA_LEN_IF_NOT_NULL(allocationCount) sizes); + +/** \brief Maps the allocation temporarily if needed, copies data from specified host pointer to it, and flushes the memory from the host caches if needed. + +\param allocator +\param pSrcHostPointer Pointer to the host data that become source of the copy. +\param dstAllocation Handle to the allocation that becomes destination of the copy. +\param dstAllocationLocalOffset Offset within `dstAllocation` where to write copied data, in bytes. +\param size Number of bytes to copy. + +This is a convenience function that allows to copy data from a host pointer to an allocation easily. +Same behavior can be achieved by calling vmaMapMemory(), `memcpy()`, vmaUnmapMemory(), vmaFlushAllocation(). + +This function can be called only for allocations created in a memory type that has `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` flag. +It can be ensured e.g. by using #VMA_MEMORY_USAGE_AUTO and #VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or +#VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT. +Otherwise, the function will fail and generate a Validation Layers error. + +`dstAllocationLocalOffset` is relative to the contents of given `dstAllocation`. +If you mean whole allocation, you should pass 0. +Do not pass allocation's offset within device memory block this parameter! +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCopyMemoryToAllocation( + VmaAllocator VMA_NOT_NULL allocator, + const void* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(size) pSrcHostPointer, + VmaAllocation VMA_NOT_NULL dstAllocation, + VkDeviceSize dstAllocationLocalOffset, + VkDeviceSize size); + +/** \brief Invalidates memory in the host caches if needed, maps the allocation temporarily if needed, and copies data from it to a specified host pointer. + +\param allocator +\param srcAllocation Handle to the allocation that becomes source of the copy. +\param srcAllocationLocalOffset Offset within `srcAllocation` where to read copied data, in bytes. +\param pDstHostPointer Pointer to the host memory that become destination of the copy. +\param size Number of bytes to copy. + +This is a convenience function that allows to copy data from an allocation to a host pointer easily. +Same behavior can be achieved by calling vmaInvalidateAllocation(), vmaMapMemory(), `memcpy()`, vmaUnmapMemory(). + +This function should be called only for allocations created in a memory type that has `VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` +and `VK_MEMORY_PROPERTY_HOST_CACHED_BIT` flag. +It can be ensured e.g. by using #VMA_MEMORY_USAGE_AUTO and #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT. +Otherwise, the function may fail and generate a Validation Layers error. +It may also work very slowly when reading from an uncached memory. + +`srcAllocationLocalOffset` is relative to the contents of given `srcAllocation`. +If you mean whole allocation, you should pass 0. +Do not pass allocation's offset within device memory block as this parameter! +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCopyAllocationToMemory( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL srcAllocation, + VkDeviceSize srcAllocationLocalOffset, + void* VMA_NOT_NULL VMA_LEN_IF_NOT_NULL(size) pDstHostPointer, + VkDeviceSize size); + +/** \brief Checks magic number in margins around all allocations in given memory types (in both default and custom pools) in search for corruptions. + +\param 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 +`HOST_VISIBLE` and `HOST_COHERENT`. For more information, see [Corruption detection](@ref debugging_memory_usage_corruption_detection). + +Possible return values: + +- `VK_ERROR_FEATURE_NOT_PRESENT` - corruption detection is not enabled for any of specified memory types. +- `VK_SUCCESS` - corruption detection has been performed and succeeded. +- `VK_ERROR_UNKNOWN` - corruption detection has been performed and found memory corruptions around one of the allocations. + `VMA_ASSERT` is also fired in that case. +- Other value: Error returned by Vulkan, e.g. memory mapping failure. +*/ +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] 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. + +For more information about defragmentation, see documentation chapter: +[Defragmentation](@ref defragmentation). +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentation( + VmaAllocator VMA_NOT_NULL allocator, + const VmaDefragmentationInfo* VMA_NOT_NULL pInfo, + VmaDefragmentationContext VMA_NULLABLE* VMA_NOT_NULL pContext); + +/** \brief Ends defragmentation process. + +\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 information 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_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 information 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_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo); + +/** \brief Binds buffer to allocation. + +Binds specified buffer to region of memory represented by specified allocation. +Gets `VkDeviceMemory` handle and offset from the allocation. +If you want to create a buffer, allocate memory for it and bind them together separately, +you should use this function for binding instead of standard `vkBindBufferMemory()`, +because it ensures proper synchronization so that when a `VkDeviceMemory` object is used by multiple +allocations, calls to `vkBind*Memory()` or `vkMapMemory()` won't happen from multiple threads simultaneously +(which is illegal in Vulkan). + +It is recommended to use function vmaCreateBuffer() instead of this one. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkBuffer VMA_NOT_NULL_NON_DISPATCHABLE buffer); + +/** \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. + +This function is similar to vmaBindBufferMemory(), but it provides additional parameters. + +If `pNext` is not null, #VmaAllocator object must have been created with #VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT flag +or with VmaAllocatorCreateInfo::vulkanApiVersion `>= VK_API_VERSION_1_1`. Otherwise the call fails. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + VkBuffer VMA_NOT_NULL_NON_DISPATCHABLE buffer, + const void* VMA_NULLABLE VMA_EXTENDS_VK_STRUCT(VkBindBufferMemoryInfoKHR) pNext); + +/** \brief Binds image to allocation. + +Binds specified image to region of memory represented by specified allocation. +Gets `VkDeviceMemory` handle and offset from the allocation. +If you want to create an image, allocate memory for it and bind them together separately, +you should use this function for binding instead of standard `vkBindImageMemory()`, +because it ensures proper synchronization so that when a `VkDeviceMemory` object is used by multiple +allocations, calls to `vkBind*Memory()` or `vkMapMemory()` won't happen from multiple threads simultaneously +(which is illegal in Vulkan). + +It is recommended to use function vmaCreateImage() instead of this one. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkImage VMA_NOT_NULL_NON_DISPATCHABLE image); + +/** \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. + +This function is similar to vmaBindImageMemory(), but it provides additional parameters. + +If `pNext` is not null, #VmaAllocator object must have been created with #VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT flag +or with VmaAllocatorCreateInfo::vulkanApiVersion `>= VK_API_VERSION_1_1`. Otherwise the call fails. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + VkImage VMA_NOT_NULL_NON_DISPATCHABLE image, + const void* VMA_NULLABLE VMA_EXTENDS_VK_STRUCT(VkBindImageMemoryInfoKHR) pNext); + +/** \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: + +-# Creates buffer. +-# Allocates appropriate memory for it. +-# Binds the buffer with the memory. + +If any of these operations fail, buffer and allocation are not created, +returned value is negative error code, `*pBuffer` and `*pAllocation` are null. + +If the function succeeded, you must destroy both buffer and allocation when you +no longer need them using either convenience function vmaDestroyBuffer() or +separately, using `vkDestroyBuffer()` and vmaFreeMemory(). + +If #VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT flag was used, +VK_KHR_dedicated_allocation extension is used internally to query driver whether +it requires or prefers the new buffer to have dedicated allocation. If yes, +and if dedicated allocation is possible +(#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. + +\note This function creates a new `VkBuffer`. Sub-allocation of parts of one large buffer, +although recommended as a good practice, is out of scope of this library and could be implemented +by the user as a higher-level logic on top of VMA. +*/ +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, + VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); + +/** \brief Creates a buffer with additional minimum alignment. + +Similar to vmaCreateBuffer() but provides additional parameter `minAlignment` which allows to specify custom, +minimum alignment to be used when placing the buffer inside a larger memory block, which may be needed e.g. +for interop with OpenGL. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBufferWithAlignment( + VmaAllocator VMA_NOT_NULL allocator, + 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, + 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(). + +\note There is a new version of this function augmented with parameter `allocationLocalOffset` - see vmaCreateAliasingBuffer2(). +*/ +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 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 allocationLocalOffset Additional offset to be added while binding, relative to the beginning of the allocation. Normally it should be 0. +\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(). + +\note This is a new version of the function augmented with parameter `allocationLocalOffset`. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingBuffer2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + 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: + +\code +vkDestroyBuffer(device, buffer, allocationCallbacks); +vmaFreeMemory(allocator, allocation); +\endcode + +It is safe to pass null as buffer and/or allocation. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyBuffer( + VmaAllocator VMA_NOT_NULL allocator, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE buffer, + VmaAllocation VMA_NULLABLE allocation); + +/// Function similar to vmaCreateBuffer(). +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, + VmaAllocationInfo* VMA_NULLABLE pAllocationInfo); + +/// Function similar to vmaCreateAliasingBuffer() but for images. +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); + +/// Function similar to vmaCreateAliasingBuffer2() but for images. +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingImage2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + 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: + +\code +vkDestroyImage(device, image, allocationCallbacks); +vmaFreeMemory(allocator, allocation); +\endcode + +It is safe to pass null as image and/or allocation. +*/ +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyImage( + VmaAllocator VMA_NOT_NULL allocator, + 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 +#endif + +#ifdef VMA_IMPLEMENTATION +#undef VMA_IMPLEMENTATION + +#include +#include +#include +#include +#include +#include + +#if !defined(VMA_CPP20) + #if __cplusplus >= 202002L || _MSVC_LANG >= 202002L // C++20 + #define VMA_CPP20 1 + #else + #define VMA_CPP20 0 + #endif +#endif + +#ifdef _MSC_VER + #include // For functions like __popcnt, _BitScanForward etc. +#endif +#if VMA_CPP20 + #include +#endif + +#if VMA_STATS_STRING_ENABLED + #include // For snprintf +#endif + +/******************************************************************************* +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 +internally, like: + + vulkanFunctions.vkAllocateMemory = &vkAllocateMemory; +*/ +#if !defined(VMA_STATIC_VULKAN_FUNCTIONS) && !defined(VK_NO_PROTOTYPES) + #define VMA_STATIC_VULKAN_FUNCTIONS 1 +#endif + +/* +Define this macro to 1 to make the library fetch pointers to Vulkan functions +internally, like: + + 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 +#endif + +#ifndef VMA_USE_STL_SHARED_MUTEX + #if __cplusplus >= 201703L || _MSVC_LANG >= 201703L // C++17 + #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. + #elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 190023918 && __cplusplus == 199711L && _MSVC_LANG >= 201703L + #define VMA_USE_STL_SHARED_MUTEX 1 + #else + #define VMA_USE_STL_SHARED_MUTEX 0 + #endif +#endif + +/* +Define this macro to include custom header files without having to edit this file directly, e.g.: + + // Inside of "my_vma_configuration_user_includes.h": + + #include "my_custom_assert.h" // for MY_CUSTOM_ASSERT + #include "my_custom_min.h" // for my_custom_min + #include + #include + + // 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. +*/ +#if !defined(VMA_CONFIGURATION_USER_INCLUDES_H) + #include // for assert + #include // for min, max, swap + #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. + #define VMA_NULL nullptr +#endif + +#ifndef VMA_FALLTHROUGH + #if __cplusplus >= 201703L || _MSVC_LANG >= 201703L // C++17 + #define VMA_FALLTHROUGH [[fallthrough]] + #else + #define VMA_FALLTHROUGH + #endif +#endif + +// Normal assert to check for programmer's errors, especially in Debug configuration. +#ifndef VMA_ASSERT + #ifdef NDEBUG + #define VMA_ASSERT(expr) + #else + #define VMA_ASSERT(expr) assert(expr) + #endif +#endif + +// Assert that will be called very often, like inside data structures e.g. operator[]. +// Making it non-empty can make program slow. +#ifndef VMA_HEAVY_ASSERT + #ifdef NDEBUG + #define VMA_HEAVY_ASSERT(expr) + #else + #define VMA_HEAVY_ASSERT(expr) //VMA_ASSERT(expr) + #endif +#endif + +// Assert used for reporting memory leaks - unfreed allocations. +#ifndef VMA_ASSERT_LEAK + #define VMA_ASSERT_LEAK(expr) VMA_ASSERT(expr) +#endif + +// If your compiler is not compatible with C++17 and definition of +// aligned_alloc() function is missing, uncommenting following line may help: + +//#include + +#if defined(__ANDROID_API__) && (__ANDROID_API__ < 16) +#include +static void* vma_aligned_alloc(size_t alignment, size_t size) +{ + // alignment must be >= sizeof(void*) + if(alignment < sizeof(void*)) + { + alignment = sizeof(void*); + } + + return memalign(alignment, size); +} +#elif defined(__APPLE__) || defined(__ANDROID__) || (defined(__linux__) && defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC)) +#include + +#if defined(__APPLE__) +#include +#endif + +static void* vma_aligned_alloc(size_t alignment, size_t size) +{ + // Unfortunately, aligned_alloc causes VMA to crash due to it returning null pointers. (At least under 11.4) + // Therefore, for now disable this specific exception until a proper solution is found. + //#if defined(__APPLE__) && (defined(MAC_OS_X_VERSION_10_16) || defined(__IPHONE_14_0)) + //#if MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_16 || __IPHONE_OS_VERSION_MAX_ALLOWED >= __IPHONE_14_0 + // // For C++14, usr/include/malloc/_malloc.h declares aligned_alloc()) only + // // with the MacOSX11.0 SDK in Xcode 12 (which is what adds + // // MAC_OS_X_VERSION_10_16), even though the function is marked + // // available for 10.15. That is why the preprocessor checks for 10.16 but + // // the __builtin_available checks for 10.15. + // // People who use C++17 could call aligned_alloc with the 10.15 SDK already. + // if (__builtin_available(macOS 10.15, iOS 13, *)) + // return aligned_alloc(alignment, size); + //#endif + //#endif + + // alignment must be >= sizeof(void*) + if(alignment < sizeof(void*)) + { + alignment = sizeof(void*); + } + + void *pointer; + if(posix_memalign(&pointer, alignment, size) == 0) + return pointer; + return VMA_NULL; +} +#elif defined(_WIN32) +static void* vma_aligned_alloc(size_t alignment, size_t size) +{ + return _aligned_malloc(size, alignment); +} +#elif __cplusplus >= 201703L || _MSVC_LANG >= 201703L // C++17 +static void* vma_aligned_alloc(size_t alignment, size_t size) +{ + return aligned_alloc(alignment, size); +} +#else +static void* vma_aligned_alloc(size_t alignment, size_t size) +{ + VMA_ASSERT(0 && "Could not implement aligned_alloc automatically. Please enable C++17 or later in your compiler or provide custom implementation of macro VMA_SYSTEM_ALIGNED_MALLOC (and VMA_SYSTEM_ALIGNED_FREE if needed) using the API of your system."); + return VMA_NULL; +} +#endif + +#if defined(_WIN32) +static void vma_aligned_free(void* ptr) +{ + _aligned_free(ptr); +} +#else +static void vma_aligned_free(void* VMA_NULLABLE ptr) +{ + free(ptr); +} +#endif + +#ifndef VMA_ALIGN_OF + #define VMA_ALIGN_OF(type) (alignof(type)) +#endif + +#ifndef VMA_SYSTEM_ALIGNED_MALLOC + #define VMA_SYSTEM_ALIGNED_MALLOC(size, alignment) vma_aligned_alloc((alignment), (size)) +#endif + +#ifndef VMA_SYSTEM_ALIGNED_FREE + // VMA_SYSTEM_FREE is the old name, but might have been defined by the user + #if defined(VMA_SYSTEM_FREE) + #define VMA_SYSTEM_ALIGNED_FREE(ptr) VMA_SYSTEM_FREE(ptr) + #else + #define VMA_SYSTEM_ALIGNED_FREE(ptr) vma_aligned_free(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 + +#ifndef VMA_MAX + #define VMA_MAX(v1, v2) ((std::max)((v1), (v2))) +#endif + +#ifndef VMA_SORT + #define VMA_SORT(beg, end, cmp) std::sort(beg, end, cmp) +#endif + +#ifndef VMA_DEBUG_LOG_FORMAT + #define VMA_DEBUG_LOG_FORMAT(format, ...) + /* + #define VMA_DEBUG_LOG_FORMAT(format, ...) do { \ + printf((format), __VA_ARGS__); \ + printf("\n"); \ + } while(false) + */ +#endif + +#ifndef VMA_DEBUG_LOG + #define VMA_DEBUG_LOG(str) VMA_DEBUG_LOG_FORMAT("%s", (str)) +#endif + +#ifndef VMA_LEAK_LOG_FORMAT + #define VMA_LEAK_LOG_FORMAT(format, ...) VMA_DEBUG_LOG_FORMAT(format, __VA_ARGS__) +#endif + +#ifndef VMA_CLASS_NO_COPY + #define VMA_CLASS_NO_COPY(className) \ + private: \ + className(const className&) = delete; \ + className& operator=(const className&) = delete; +#endif +#ifndef VMA_CLASS_NO_COPY_NO_MOVE + #define VMA_CLASS_NO_COPY_NO_MOVE(className) \ + private: \ + className(const className&) = delete; \ + className(className&&) = delete; \ + className& operator=(const className&) = delete; \ + className& operator=(className&&) = delete; +#endif + +// Define this macro to 1 to enable functions: vmaBuildStatsString, vmaFreeStatsString. +#if VMA_STATS_STRING_ENABLED + static inline void VmaUint32ToStr(char* VMA_NOT_NULL outStr, size_t strLen, uint32_t num) + { + snprintf(outStr, strLen, "%" PRIu32, num); + } + static inline void VmaUint64ToStr(char* VMA_NOT_NULL outStr, size_t strLen, uint64_t num) + { + snprintf(outStr, strLen, "%" PRIu64, num); + } + static inline void VmaPtrToStr(char* VMA_NOT_NULL outStr, size_t strLen, const void* ptr) + { + snprintf(outStr, strLen, "%p", ptr); + } +#endif + +#ifndef VMA_MUTEX + class VmaMutex + { + VMA_CLASS_NO_COPY_NO_MOVE(VmaMutex) + public: + VmaMutex() { } + void Lock() { m_Mutex.lock(); } + void Unlock() { m_Mutex.unlock(); } + bool TryLock() { return m_Mutex.try_lock(); } + private: + std::mutex m_Mutex; + }; + #define VMA_MUTEX VmaMutex +#endif + +// Read-write mutex, where "read" is shared access, "write" is exclusive access. +#ifndef VMA_RW_MUTEX + #if VMA_USE_STL_SHARED_MUTEX + // Use std::shared_mutex from C++17. + #include + class VmaRWMutex + { + public: + void LockRead() { m_Mutex.lock_shared(); } + void UnlockRead() { m_Mutex.unlock_shared(); } + bool TryLockRead() { return m_Mutex.try_lock_shared(); } + void LockWrite() { m_Mutex.lock(); } + void UnlockWrite() { m_Mutex.unlock(); } + bool TryLockWrite() { return m_Mutex.try_lock(); } + private: + std::shared_mutex m_Mutex; + }; + #define VMA_RW_MUTEX VmaRWMutex + #elif defined(_WIN32) && defined(WINVER) && defined(SRWLOCK_INIT) && WINVER >= 0x0600 + // Use SRWLOCK from WinAPI. + // Minimum supported client = Windows Vista, server = Windows Server 2008. + class VmaRWMutex + { + public: + VmaRWMutex() { InitializeSRWLock(&m_Lock); } + void LockRead() { AcquireSRWLockShared(&m_Lock); } + void UnlockRead() { ReleaseSRWLockShared(&m_Lock); } + bool TryLockRead() { return TryAcquireSRWLockShared(&m_Lock) != FALSE; } + void LockWrite() { AcquireSRWLockExclusive(&m_Lock); } + void UnlockWrite() { ReleaseSRWLockExclusive(&m_Lock); } + bool TryLockWrite() { return TryAcquireSRWLockExclusive(&m_Lock) != FALSE; } + private: + SRWLOCK m_Lock; + }; + #define VMA_RW_MUTEX VmaRWMutex + #else + // Less efficient fallback: Use normal mutex. + class VmaRWMutex + { + public: + void LockRead() { m_Mutex.Lock(); } + void UnlockRead() { m_Mutex.Unlock(); } + bool TryLockRead() { return m_Mutex.TryLock(); } + void LockWrite() { m_Mutex.Lock(); } + void UnlockWrite() { m_Mutex.Unlock(); } + bool TryLockWrite() { return m_Mutex.TryLock(); } + private: + VMA_MUTEX m_Mutex; + }; + #define VMA_RW_MUTEX VmaRWMutex + #endif // #if VMA_USE_STL_SHARED_MUTEX +#endif // #ifndef VMA_RW_MUTEX + +/* +If providing your own implementation, you need to implement a subset of std::atomic. +*/ +#ifndef VMA_ATOMIC_UINT32 + #include + #define VMA_ATOMIC_UINT32 std::atomic +#endif + +#ifndef VMA_ATOMIC_UINT64 + #include + #define VMA_ATOMIC_UINT64 std::atomic +#endif + +#ifndef VMA_DEBUG_ALWAYS_DEDICATED_MEMORY + /** + Every allocation will have its own memory block. + Define to 1 for debugging purposes only. + */ + #define VMA_DEBUG_ALWAYS_DEDICATED_MEMORY (0) +#endif + +#ifndef VMA_MIN_ALIGNMENT + /** + Minimum alignment of all allocations, in bytes. + Set to more than 1 for debugging purposes. Must be power of two. + */ + #ifdef VMA_DEBUG_ALIGNMENT // Old name + #define VMA_MIN_ALIGNMENT VMA_DEBUG_ALIGNMENT + #else + #define VMA_MIN_ALIGNMENT (1) + #endif +#endif + +#ifndef VMA_DEBUG_MARGIN + /** + Minimum margin after every allocation, in bytes. + Set nonzero for debugging purposes only. + */ + #define VMA_DEBUG_MARGIN (0) +#endif + +#ifndef VMA_DEBUG_INITIALIZE_ALLOCATIONS + /** + Define this macro to 1 to automatically fill new allocations and destroyed + allocations with some bit pattern. + */ + #define VMA_DEBUG_INITIALIZE_ALLOCATIONS (0) +#endif + +#ifndef VMA_DEBUG_DETECT_CORRUPTION + /** + Define this macro to 1 together with non-zero value of VMA_DEBUG_MARGIN to + enable writing magic value to the margin after every allocation and + validating it, so that memory corruptions (out-of-bounds writes) are detected. + */ + #define VMA_DEBUG_DETECT_CORRUPTION (0) +#endif + +#ifndef VMA_DEBUG_GLOBAL_MUTEX + /** + Set this to 1 for debugging purposes only, to enable single mutex protecting all + entry calls to the library. Can be useful for debugging multithreading issues. + */ + #define VMA_DEBUG_GLOBAL_MUTEX (0) +#endif + +#ifndef VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY + /** + Minimum value for VkPhysicalDeviceLimits::bufferImageGranularity. + Set to more than 1 for debugging purposes only. Must be power of two. + */ + #define VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY (1) +#endif + +#ifndef VMA_DEBUG_DONT_EXCEED_MAX_MEMORY_ALLOCATION_COUNT + /* + Set this to 1 to make VMA never exceed VkPhysicalDeviceLimits::maxMemoryAllocationCount + and return error instead of leaving up to Vulkan implementation what to do in such cases. + */ + #define VMA_DEBUG_DONT_EXCEED_MAX_MEMORY_ALLOCATION_COUNT (0) +#endif + +#ifndef VMA_SMALL_HEAP_MAX_SIZE + /// Maximum size of a memory heap in Vulkan to consider it "small". + #define VMA_SMALL_HEAP_MAX_SIZE (1024ull * 1024 * 1024) +#endif + +#ifndef VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE + /// Default size of a block allocated as single VkDeviceMemory from a "large" heap. + #define VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE (256ull * 1024 * 1024) +#endif + +/* +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 + +#define VMA_VALIDATE(cond) do { if(!(cond)) { \ + VMA_ASSERT(0 && "Validation failed: " #cond); \ + return false; \ + } } while(false) + +/******************************************************************************* +END OF CONFIGURATION +*/ +#endif // _VMA_CONFIGURATION + + +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; + +// 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) + + +#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; + +#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 intrinsics 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 VMA_CPP20 + 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 >> 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 VMA_CPP20 + if(mask) + return static_cast(std::countr_zero(mask)); + 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 VMA_CPP20 + if(mask) + return static_cast(std::countr_zero(mask)); + 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 VMA_CPP20 + if(mask) + return 63 - static_cast(std::countl_zero(mask)); +#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 VMA_CPP20 + if(mask) + return 31 - static_cast(std::countl_zero(mask)); +#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; +} + +/* +Returns true if given number is a power of two. +T must be unsigned integer number or signed integer but always nonnegative. +For 0 returns true. +*/ +template +inline bool VmaIsPow2(T x) +{ + return (x & (x - 1)) == 0; +} + +// Aligns given value up to nearest multiply of align value. For example: VmaAlignUp(11, 8) = 16. +// Use types like uint32_t, uint64_t as T. +template +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: VmaAlignDown(11, 8) = 8. +// Use types like uint32_t, uint64_t as T. +template +static inline T VmaAlignDown(T val, T alignment) +{ + VMA_HEAVY_ASSERT(VmaIsPow2(alignment)); + return val & ~(alignment - 1); +} + +// Division with mathematical rounding to nearest number. +template +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) +{ + v--; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v++; + return v; +} + +static inline uint64_t VmaNextPow2(uint64_t v) +{ + v--; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + v++; + return v; +} + +// Returns largest power of 2 less or equal to v. +static inline uint32_t VmaPrevPow2(uint32_t v) +{ + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v = v ^ (v >> 1); + return v; +} + +static inline uint64_t VmaPrevPow2(uint64_t v) +{ + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + v = v ^ (v >> 1); + return v; +} + +static inline bool VmaStrIsEmpty(const char* pStr) +{ + return pStr == VMA_NULL || *pStr == '\0'; +} + +/* +Returns true if two memory blocks occupy overlapping pages. +ResourceA must be in less memory offset than ResourceB. + +Algorithm is based on "Vulkan 1.0.39 - A Specification (with all registered Vulkan extensions)" +chapter 11.6 "Resource Memory Association", paragraph "Buffer-Image Granularity". +*/ +static inline bool VmaBlocksOnSamePage( + VkDeviceSize resourceAOffset, + VkDeviceSize resourceASize, + VkDeviceSize resourceBOffset, + VkDeviceSize pageSize) +{ + VMA_ASSERT(resourceAOffset + resourceASize <= resourceBOffset && resourceASize > 0 && pageSize > 0); + VkDeviceSize resourceAEnd = resourceAOffset + resourceASize - 1; + VkDeviceSize resourceAEndPage = resourceAEnd & ~(pageSize - 1); + VkDeviceSize resourceBStart = resourceBOffset; + VkDeviceSize resourceBStartPage = resourceBStart & ~(pageSize - 1); + return resourceAEndPage == resourceBStartPage; +} + +/* +Returns true if given suballocation types could conflict and must respect +VkPhysicalDeviceLimits::bufferImageGranularity. They conflict if one is buffer +or linear image and another one is optimal image. If type is unknown, behave +conservatively. +*/ +static inline bool VmaIsBufferImageGranularityConflict( + VmaSuballocationType suballocType1, + VmaSuballocationType suballocType2) +{ + if (suballocType1 > suballocType2) + { + std::swap(suballocType1, suballocType2); + } + + switch (suballocType1) + { + case VMA_SUBALLOCATION_TYPE_FREE: + return false; + case VMA_SUBALLOCATION_TYPE_UNKNOWN: + return true; + case VMA_SUBALLOCATION_TYPE_BUFFER: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR || + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR: + return + suballocType2 == VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL; + case VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL: + return false; + default: + VMA_ASSERT(0); + return true; + } +} + +static void VmaWriteMagicValue(void* pData, VkDeviceSize offset) +{ +#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) + { + *pDst = VMA_CORRUPTION_DETECTION_MAGIC_VALUE; + } +#else + // no-op +#endif +} + +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) + { + if (*pSrc != VMA_CORRUPTION_DETECTION_MAGIC_VALUE) + { + return false; + } + } +#endif + return true; +} + +/* +Fills structure with parameters of an example buffer to be used for transfers +during GPU memory defragmentation. +*/ +static void VmaFillGpuDefragmentationBufferCreateInfo(VkBufferCreateInfo& outBufCreateInfo) +{ + memset(&outBufCreateInfo, 0, sizeof(outBufCreateInfo)); + outBufCreateInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; + outBufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + outBufCreateInfo.size = (VkDeviceSize)VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE; // Example size. +} + + +/* +Performs binary search and returns iterator to first element that is greater or +equal to (key), according to comparison (cmp). + +Cmp should return true if first argument is less than second argument. + +Returned value is the found element, if present in the collection or place where +new element with value (key) should be inserted. +*/ +template +static IterT VmaBinaryFindFirstNotLess(IterT beg, IterT end, const KeyT& key, const CmpLess& cmp) +{ + size_t down = 0, up = size_t(end - beg); + while (down < up) + { + const size_t mid = down + (up - down) / 2; // Overflow-safe midpoint calculation + if (cmp(*(beg + mid), key)) + { + down = mid + 1; + } + else + { + up = mid; + } + } + return beg + down; +} + +template +IterT VmaBinaryFindSorted(const IterT& beg, const IterT& end, const KeyT& value, const CmpLess& cmp) +{ + IterT it = VmaBinaryFindFirstNotLess( + beg, end, value, cmp); + if (it == end || + (!cmp(*it, value) && !cmp(value, *it))) + { + return it; + } + return end; +} + +/* +Returns true if all pointers in the array are not-null and unique. +Warning! O(n^2) complexity. Use only inside VMA_HEAVY_ASSERT. +T must be pointer type, e.g. VmaAllocation, VmaPool. +*/ +template +static bool VmaValidatePointerArray(uint32_t count, const T* arr) +{ + for (uint32_t i = 0; i < count; ++i) + { + const T iPtr = arr[i]; + if (iPtr == VMA_NULL) + { + return false; + } + for (uint32_t j = i + 1; j < count; ++j) + { + if (iPtr == arr[j]) + { + return false; + } + } + } + return true; +} + +template +static inline void VmaPnextChainPushFront(MainT* mainStruct, NewT* newStruct) +{ + newStruct->pNext = mainStruct->pNext; + mainStruct->pNext = newStruct; +} +// Finds structure with s->sType == sType in mainStruct->pNext chain. +// Returns pointer to it. If not found, returns null. +template +static inline const FindT* VmaPnextChainFind(const MainT* mainStruct, VkStructureType sType) +{ + for(const VkBaseInStructure* s = (const VkBaseInStructure*)mainStruct->pNext; + s != VMA_NULL; s = s->pNext) + { + if(s->sType == sType) + { + return (const FindT*)s; + } + } + return VMA_NULL; +} + +// An abstraction over buffer or image `usage` flags, depending on available extensions. +struct VmaBufferImageUsage +{ +#if VMA_KHR_MAINTENANCE5 + typedef uint64_t BaseType; // VkFlags64 +#else + typedef uint32_t BaseType; // VkFlags32 +#endif + + static const VmaBufferImageUsage UNKNOWN; + + BaseType Value; + + VmaBufferImageUsage() { *this = UNKNOWN; } + explicit VmaBufferImageUsage(BaseType usage) : Value(usage) { } + VmaBufferImageUsage(const VkBufferCreateInfo &createInfo, bool useKhrMaintenance5); + explicit VmaBufferImageUsage(const VkImageCreateInfo &createInfo); + + bool operator==(const VmaBufferImageUsage& rhs) const { return Value == rhs.Value; } + bool operator!=(const VmaBufferImageUsage& rhs) const { return Value != rhs.Value; } + + bool Contains(BaseType flag) const { return (Value & flag) != 0; } + bool ContainsDeviceAccess() const + { + // This relies on values of VK_IMAGE_USAGE_TRANSFER* being the same as VK_BUFFER_IMAGE_TRANSFER*. + return (Value & ~BaseType(VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT)) != 0; + } +}; + +const VmaBufferImageUsage VmaBufferImageUsage::UNKNOWN = VmaBufferImageUsage(0); + +VmaBufferImageUsage::VmaBufferImageUsage(const VkBufferCreateInfo &createInfo, + bool useKhrMaintenance5) +{ +#if VMA_KHR_MAINTENANCE5 + if(useKhrMaintenance5) + { + // If VkBufferCreateInfo::pNext chain contains VkBufferUsageFlags2CreateInfoKHR, + // take usage from it and ignore VkBufferCreateInfo::usage, per specification + // of the VK_KHR_maintenance5 extension. + const VkBufferUsageFlags2CreateInfoKHR* const usageFlags2 = + VmaPnextChainFind(&createInfo, VK_STRUCTURE_TYPE_BUFFER_USAGE_FLAGS_2_CREATE_INFO_KHR); + if(usageFlags2) + { + this->Value = usageFlags2->usage; + return; + } + } +#endif + + this->Value = (BaseType)createInfo.usage; +} + +VmaBufferImageUsage::VmaBufferImageUsage(const VkImageCreateInfo &createInfo) +{ + // Maybe in the future there will be VK_KHR_maintenanceN extension with structure + // VkImageUsageFlags2CreateInfoKHR, like the one for buffers... + + this->Value = (BaseType)createInfo.usage; +} + +// 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, + VmaBufferImageUsage bufImgUsage, + 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 == VmaBufferImageUsage::UNKNOWN) + { + 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." + " Maybe you use VkBufferUsageFlags2CreateInfoKHR but forgot to use VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT?" ); + return false; + } + + const bool deviceAccess = bufImgUsage.ContainsDeviceAccess(); + 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) + { + // Prefer cached. Cannot require it, because some platforms don't have it (e.g. Raspberry Pi - see #362)! + outPreferredFlags |= VK_MEMORY_PROPERTY_HOST_CACHED_BIT; + + 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; + } + else + { + // Always CPU memory. + outRequiredFlags |= VK_MEMORY_PROPERTY_HOST_VISIBLE_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 + { + // if(deviceAccess) + // + // GPU access, no CPU access (e.g. a color attachment image) - prefer GPU memory, + // unless there is a clear preference from the user not to do so. + // + // else: + // + // 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) && + (pAllocationCallbacks->pfnAllocation != VMA_NULL)) + { + result = (*pAllocationCallbacks->pfnAllocation)( + pAllocationCallbacks->pUserData, + size, + alignment, + VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); + } + else + { + result = VMA_SYSTEM_ALIGNED_MALLOC(size, alignment); + } + VMA_ASSERT(result != VMA_NULL && "CPU memory allocation failed."); + return result; +} + +static void VmaFree(const VkAllocationCallbacks* pAllocationCallbacks, void* ptr) +{ + if ((pAllocationCallbacks != VMA_NULL) && + (pAllocationCallbacks->pfnFree != VMA_NULL)) + { + (*pAllocationCallbacks->pfnFree)(pAllocationCallbacks->pUserData, ptr); + } + else + { + VMA_SYSTEM_ALIGNED_FREE(ptr); + } +} + +template +static T* VmaAllocate(const VkAllocationCallbacks* pAllocationCallbacks) +{ + return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T), VMA_ALIGN_OF(T)); +} + +template +static T* VmaAllocateArray(const VkAllocationCallbacks* pAllocationCallbacks, size_t count) +{ + return (T*)VmaMalloc(pAllocationCallbacks, sizeof(T) * count, VMA_ALIGN_OF(T)); +} + +#define vma_new(allocator, type) new(VmaAllocate(allocator))(type) + +#define vma_new_array(allocator, type, count) new(VmaAllocateArray((allocator), (count)))(type) + +template +static void vma_delete(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr) +{ + ptr->~T(); + VmaFree(pAllocationCallbacks, ptr); +} + +template +static void vma_delete_array(const VkAllocationCallbacks* pAllocationCallbacks, T* ptr, size_t count) +{ + if (ptr != VMA_NULL) + { + for (size_t i = count; i--; ) + { + ptr[i].~T(); + } + VmaFree(pAllocationCallbacks, ptr); + } +} + +static char* VmaCreateStringCopy(const VkAllocationCallbacks* allocs, const char* srcStr) +{ + 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; + } + 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) + { + const size_t len = strlen(str); + vma_delete_array(allocs, str, len + 1); + } +} + +template +size_t VmaVectorInsertSorted(VectorT& vector, const typename VectorT::value_type& value) +{ + const size_t indexToInsert = VmaBinaryFindFirstNotLess( + vector.data(), + vector.data() + vector.size(), + value, + CmpLess()) - vector.data(); + VmaVectorInsert(vector, indexToInsert, value); + return indexToInsert; +} + +template +bool VmaVectorRemoveSorted(VectorT& vector, const typename VectorT::value_type& value) +{ + CmpLess comparator; + typename VectorT::iterator it = VmaBinaryFindFirstNotLess( + vector.begin(), + vector.end(), + value, + comparator); + if ((it != vector.end()) && !comparator(*it, value) && !comparator(value, *it)) + { + size_t indexToRemove = it - vector.begin(); + VmaVectorRemove(vector, indexToRemove); + return true; + } + return false; +} +#endif // _VMA_FUNCTIONS + +#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_NO_MOVE(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_NO_MOVE(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_NO_MOVE(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: + using T = decltype(AtomicT().load()); + + ~AtomicTransactionalIncrement() + { + if(m_Atomic) + --(*m_Atomic); + } + + void Commit() { m_Atomic = VMA_NULL; } + T Increment(AtomicT* atomic) + { + m_Atomic = atomic; + return m_Atomic->fetch_add(1); + } + +private: + AtomicT* m_Atomic = VMA_NULL; +}; +#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. + +It contains some number of elements in-place, which allows it to avoid heap allocation +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); + VmaSmallVector(size_t count, const AllocatorT& allocator); + template + VmaSmallVector(const VmaSmallVector&) = delete; + template + 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; } + 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 +}; + +#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 +allocator can create multiple blocks. +*/ +template +class VmaPoolAllocator +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaPoolAllocator) +public: + VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, uint32_t firstBlockCapacity); + ~VmaPoolAllocator(); + template T* Alloc(Types&&... args); + void Free(T* ptr); + +private: + union Item + { + uint32_t NextFreeIndex; + alignas(T) char Value[sizeof(T)]; + }; + struct ItemBlock + { + Item* pItems; + uint32_t Capacity; + uint32_t FirstFreeIndex; + }; + + const VkAllocationCallbacks* m_pAllocationCallbacks; + const uint32_t m_FirstBlockCapacity; + VmaVector> m_ItemBlocks; + + ItemBlock& CreateNewBlock(); +}; + +#ifndef _VMA_POOL_ALLOCATOR_FUNCTIONS +template +VmaPoolAllocator::VmaPoolAllocator(const VkAllocationCallbacks* pAllocationCallbacks, uint32_t firstBlockCapacity) + : m_pAllocationCallbacks(pAllocationCallbacks), + m_FirstBlockCapacity(firstBlockCapacity), + m_ItemBlocks(VmaStlAllocator(pAllocationCallbacks)) +{ + VMA_ASSERT(m_FirstBlockCapacity > 1); +} + +template +VmaPoolAllocator::~VmaPoolAllocator() +{ + for (size_t i = m_ItemBlocks.size(); i--;) + vma_delete_array(m_pAllocationCallbacks, m_ItemBlocks[i].pItems, m_ItemBlocks[i].Capacity); + m_ItemBlocks.clear(); +} + +template +template T* VmaPoolAllocator::Alloc(Types&&... args) +{ + for (size_t i = m_ItemBlocks.size(); i--; ) + { + ItemBlock& block = m_ItemBlocks[i]; + // This block has some free items: Use first one. + if (block.FirstFreeIndex != UINT32_MAX) + { + Item* const pItem = &block.pItems[block.FirstFreeIndex]; + block.FirstFreeIndex = pItem->NextFreeIndex; + T* result = (T*)&pItem->Value; + new(result)T(std::forward(args)...); // Explicit constructor call. + return result; + } + } + + // No block has free item: Create new one and use it. + ItemBlock& newBlock = CreateNewBlock(); + Item* const pItem = &newBlock.pItems[0]; + newBlock.FirstFreeIndex = pItem->NextFreeIndex; + T* result = (T*)&pItem->Value; + new(result) T(std::forward(args)...); // Explicit constructor call. + return result; +} + +template +void VmaPoolAllocator::Free(T* ptr) +{ + // Search all memory blocks to find ptr. + for (size_t i = m_ItemBlocks.size(); i--; ) + { + ItemBlock& block = m_ItemBlocks[i]; + + // Casting to union. + Item* pItemPtr; + memcpy(&pItemPtr, &ptr, sizeof(pItemPtr)); + + // Check if pItemPtr is in address range of this block. + if ((pItemPtr >= block.pItems) && (pItemPtr < block.pItems + block.Capacity)) + { + ptr->~T(); // Explicit destructor call. + const uint32_t index = static_cast(pItemPtr - block.pItems); + pItemPtr->NextFreeIndex = block.FirstFreeIndex; + block.FirstFreeIndex = index; + return; + } + } + VMA_ASSERT(0 && "Pointer doesn't belong to this memory pool."); +} + +template +typename VmaPoolAllocator::ItemBlock& VmaPoolAllocator::CreateNewBlock() +{ + const uint32_t newBlockCapacity = m_ItemBlocks.empty() ? + m_FirstBlockCapacity : m_ItemBlocks.back().Capacity * 3 / 2; + + const ItemBlock newBlock = + { + vma_new_array(m_pAllocationCallbacks, Item, newBlockCapacity), + newBlockCapacity, + 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) + 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 + +#ifndef _VMA_RAW_LIST +template +struct VmaListItem +{ + VmaListItem* pPrev; + VmaListItem* pNext; + T Value; +}; + +// Doubly linked list. +template +class VmaRawList +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaRawList) +public: + typedef VmaListItem ItemType; + + VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks); + // 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; } + ItemType* Back() { return m_pBack; } + const ItemType* Front() const { return m_pFront; } + const ItemType* Back() const { return m_pBack; } + + ItemType* PushFront(); + ItemType* PushBack(); + ItemType* PushFront(const T& value); + 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: + const VkAllocationCallbacks* const m_pAllocationCallbacks; + VmaPoolAllocator m_ItemAllocator; + ItemType* m_pFront; + ItemType* m_pBack; + size_t m_Count; +}; + +#ifndef _VMA_RAW_LIST_FUNCTIONS +template +VmaRawList::VmaRawList(const VkAllocationCallbacks* pAllocationCallbacks) + : m_pAllocationCallbacks(pAllocationCallbacks), + m_ItemAllocator(pAllocationCallbacks, 128), + m_pFront(VMA_NULL), + m_pBack(VMA_NULL), + m_Count(0) {} + +template +VmaListItem* VmaRawList::PushFront() +{ + ItemType* const pNewItem = m_ItemAllocator.Alloc(); + pNewItem->pPrev = VMA_NULL; + if (IsEmpty()) + { + pNewItem->pNext = VMA_NULL; + m_pFront = pNewItem; + m_pBack = pNewItem; + m_Count = 1; + } + else + { + pNewItem->pNext = m_pFront; + m_pFront->pPrev = pNewItem; + m_pFront = pNewItem; + ++m_Count; + } + return pNewItem; +} + +template +VmaListItem* VmaRawList::PushBack() +{ + ItemType* const pNewItem = m_ItemAllocator.Alloc(); + pNewItem->pNext = VMA_NULL; + if(IsEmpty()) + { + pNewItem->pPrev = VMA_NULL; + m_pFront = pNewItem; + m_pBack = pNewItem; + m_Count = 1; + } + else + { + pNewItem->pPrev = m_pBack; + m_pBack->pNext = pNewItem; + m_pBack = pNewItem; + ++m_Count; + } + return pNewItem; +} + +template +VmaListItem* VmaRawList::PushFront(const T& value) +{ + ItemType* const pNewItem = PushFront(); + pNewItem->Value = value; + return pNewItem; +} + +template +VmaListItem* VmaRawList::PushBack(const T& value) +{ + ItemType* const pNewItem = PushBack(); + pNewItem->Value = value; + return pNewItem; +} + +template +void VmaRawList::PopFront() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const pFrontItem = m_pFront; + ItemType* const pNextItem = pFrontItem->pNext; + if (pNextItem != VMA_NULL) + { + pNextItem->pPrev = VMA_NULL; + } + m_pFront = pNextItem; + m_ItemAllocator.Free(pFrontItem); + --m_Count; +} + +template +void VmaRawList::PopBack() +{ + VMA_HEAVY_ASSERT(m_Count > 0); + ItemType* const pBackItem = m_pBack; + ItemType* const pPrevItem = pBackItem->pPrev; + if(pPrevItem != VMA_NULL) + { + pPrevItem->pNext = VMA_NULL; + } + m_pBack = pPrevItem; + m_ItemAllocator.Free(pBackItem); + --m_Count; +} + +template +void VmaRawList::Clear() +{ + if (IsEmpty() == false) + { + ItemType* pItem = m_pBack; + while (pItem != VMA_NULL) + { + ItemType* const pPrevItem = pItem->pPrev; + m_ItemAllocator.Free(pItem); + pItem = pPrevItem; + } + m_pFront = VMA_NULL; + m_pBack = VMA_NULL; + m_Count = 0; + } +} + +template +void VmaRawList::Remove(ItemType* pItem) +{ + VMA_HEAVY_ASSERT(pItem != VMA_NULL); + VMA_HEAVY_ASSERT(m_Count > 0); + + if(pItem->pPrev != VMA_NULL) + { + pItem->pPrev->pNext = pItem->pNext; + } + else + { + VMA_HEAVY_ASSERT(m_pFront == pItem); + m_pFront = pItem->pNext; + } + + if(pItem->pNext != VMA_NULL) + { + pItem->pNext->pPrev = pItem->pPrev; + } + else + { + VMA_HEAVY_ASSERT(m_pBack == pItem); + m_pBack = pItem->pPrev; + } + + m_ItemAllocator.Free(pItem); + --m_Count; +} + +template +VmaListItem* VmaRawList::InsertBefore(ItemType* pItem) +{ + if(pItem != VMA_NULL) + { + ItemType* const prevItem = pItem->pPrev; + ItemType* const newItem = m_ItemAllocator.Alloc(); + newItem->pPrev = prevItem; + newItem->pNext = pItem; + pItem->pPrev = newItem; + if(prevItem != VMA_NULL) + { + prevItem->pNext = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_pFront == pItem); + m_pFront = newItem; + } + ++m_Count; + return newItem; + } + else + return PushBack(); +} + +template +VmaListItem* VmaRawList::InsertAfter(ItemType* pItem) +{ + if(pItem != VMA_NULL) + { + ItemType* const nextItem = pItem->pNext; + ItemType* const newItem = m_ItemAllocator.Alloc(); + newItem->pNext = nextItem; + newItem->pPrev = pItem; + pItem->pNext = newItem; + if(nextItem != VMA_NULL) + { + nextItem->pPrev = newItem; + } + else + { + VMA_HEAVY_ASSERT(m_pBack == pItem); + m_pBack = newItem; + } + ++m_Count; + return newItem; + } + else + return PushFront(); +} + +template +VmaListItem* VmaRawList::InsertBefore(ItemType* pItem, const T& value) +{ + ItemType* const newItem = InsertBefore(pItem); + newItem->Value = value; + return newItem; +} + +template +VmaListItem* VmaRawList::InsertAfter(ItemType* pItem, const T& value) +{ + ItemType* const newItem = InsertAfter(pItem); + newItem->Value = value; + return newItem; +} +#endif // _VMA_RAW_LIST_FUNCTIONS +#endif // _VMA_RAW_LIST + +#ifndef _VMA_LIST +template +class VmaList +{ + VMA_CLASS_NO_COPY_NO_MOVE(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(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; } + + 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++() { 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) {} + }; + class reverse_iterator + { + friend class const_reverse_iterator; + friend class VmaList; + public: + 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) {} + + 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; } + + 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; } + + reverse_iterator operator++(int) { reverse_iterator result = *this; ++* this; return result; } + reverse_iterator operator--(int) { reverse_iterator result = *this; --* this; return result; } + + reverse_iterator& operator++() { VMA_HEAVY_ASSERT(m_pItem != VMA_NULL); m_pItem = m_pItem->pPrev; return *this; } + reverse_iterator& operator--(); + + private: + 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) {} + + bool empty() const { return m_RawList.IsEmpty(); } + size_t size() const { return m_RawList.GetCount(); } + + iterator begin() { return iterator(&m_RawList, m_RawList.Front()); } + iterator end() { return iterator(&m_RawList, VMA_NULL); } + + const_iterator cbegin() const { return const_iterator(&m_RawList, m_RawList.Front()); } + const_iterator cend() const { return const_iterator(&m_RawList, VMA_NULL); } + + const_iterator begin() const { return cbegin(); } + const_iterator end() const { return cend(); } + + 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); } + 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; +}; + +#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; +} + +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 +{ + typedef MyItem ItemType; + static ItemType* GetPrev(const ItemType* item) { return item->myPrevPtr; } + static ItemType* GetNext(const ItemType* item) { return item->myNextPtr; } + static ItemType*& AccessPrev(ItemType* item) { return item->myPrevPtr; } + static ItemType*& AccessNext(ItemType* item) { return item->myNextPtr; } +}; +*/ +template +class VmaIntrusiveLinkedList +{ +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() = 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; } + ItemType* Back() { return m_Back; } + const ItemType* Front() const { return m_Front; } + const ItemType* Back() const { return m_Back; } + + 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); + // MyItem can be null - it means PushFront. + void InsertAfter(ItemType* existingItem, ItemType* newItem); + void Remove(ItemType* item); + void RemoveAll(); + +private: + ItemType* m_Front = VMA_NULL; + ItemType* m_Back = VMA_NULL; + size_t m_Count = 0; +}; + +#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 + +#if !defined(_VMA_STRING_BUILDER) && VMA_STATS_STRING_ENABLED +class VmaStringBuilder +{ +public: + VmaStringBuilder(const VkAllocationCallbacks* allocationCallbacks) : m_Data(VmaStlAllocator(allocationCallbacks)) {} + ~VmaStringBuilder() = default; + + 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); } + + 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' + (char)(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' + (char)(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_NO_MOVE(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); + // 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); + // 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; + + 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 ((uint8_t)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."); + } + } +} + +void VmaJsonWriter::ContinueString(uint32_t n) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::ContinueString(uint64_t n) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::ContinueString_Pointer(const void* ptr) +{ + VMA_ASSERT(m_InsideString); + m_SB.AddPointer(ptr); +} + +void VmaJsonWriter::EndString(const char* pStr) +{ + VMA_ASSERT(m_InsideString); + if (pStr != VMA_NULL && pStr[0] != '\0') + { + ContinueString(pStr); + } + m_SB.Add('"'); + m_InsideString = false; +} + +void VmaJsonWriter::WriteNumber(uint32_t n) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::WriteNumber(uint64_t n) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.AddNumber(n); +} + +void VmaJsonWriter::WriteBool(bool b) +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.Add(b ? "true" : "false"); +} + +void VmaJsonWriter::WriteNull() +{ + VMA_ASSERT(!m_InsideString); + BeginValue(false); + m_SB.Add("null"); +} + +void VmaJsonWriter::BeginValue(bool isString) +{ + if (!m_Stack.empty()) + { + StackItem& currItem = m_Stack.back(); + if (currItem.type == COLLECTION_TYPE_OBJECT && + currItem.valueCount % 2 == 0) + { + 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_NO_MOVE(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 + +#if VMA_EXTERNAL_MEMORY_WIN32 +class VmaWin32Handle +{ +public: + VmaWin32Handle() noexcept : m_hHandle(VMA_NULL) { } + explicit VmaWin32Handle(HANDLE hHandle) noexcept : m_hHandle(hHandle) { } + ~VmaWin32Handle() noexcept { if (m_hHandle != VMA_NULL) { ::CloseHandle(m_hHandle); } } + VMA_CLASS_NO_COPY_NO_MOVE(VmaWin32Handle) + +public: + // Strengthened + VkResult GetHandle(VkDevice device, VkDeviceMemory memory, PFN_vkGetMemoryWin32HandleKHR pvkGetMemoryWin32HandleKHR, HANDLE hTargetProcess, bool useMutex, HANDLE* pHandle) noexcept + { + *pHandle = VMA_NULL; + // Try to get handle first. + if (m_hHandle != VMA_NULL) + { + *pHandle = Duplicate(hTargetProcess); + return VK_SUCCESS; + } + + VkResult res = VK_SUCCESS; + // If failed, try to create it. + { + VmaMutexLockWrite lock(m_Mutex, useMutex); + if (m_hHandle == VMA_NULL) + { + res = Create(device, memory, pvkGetMemoryWin32HandleKHR, &m_hHandle); + } + } + + *pHandle = Duplicate(hTargetProcess); + return res; + } + + operator bool() const noexcept { return m_hHandle != VMA_NULL; } +private: + // Not atomic + static VkResult Create(VkDevice device, VkDeviceMemory memory, PFN_vkGetMemoryWin32HandleKHR pvkGetMemoryWin32HandleKHR, HANDLE* pHandle) noexcept + { + VkResult res = VK_ERROR_FEATURE_NOT_PRESENT; + if (pvkGetMemoryWin32HandleKHR != VMA_NULL) + { + VkMemoryGetWin32HandleInfoKHR handleInfo{ }; + handleInfo.sType = VK_STRUCTURE_TYPE_MEMORY_GET_WIN32_HANDLE_INFO_KHR; + handleInfo.memory = memory; + handleInfo.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT_KHR; + res = pvkGetMemoryWin32HandleKHR(device, &handleInfo, pHandle); + } + return res; + } + HANDLE Duplicate(HANDLE hTargetProcess = VMA_NULL) const noexcept + { + if (!m_hHandle) + return m_hHandle; + + HANDLE hCurrentProcess = ::GetCurrentProcess(); + HANDLE hDupHandle = VMA_NULL; + if (!::DuplicateHandle(hCurrentProcess, m_hHandle, hTargetProcess ? hTargetProcess : hCurrentProcess, &hDupHandle, 0, FALSE, DUPLICATE_SAME_ACCESS)) + { + VMA_ASSERT(0 && "Failed to duplicate handle."); + } + return hDupHandle; + } +private: + HANDLE m_hHandle; + VMA_RW_MUTEX m_Mutex; // Protects access m_Handle +}; +#else +class VmaWin32Handle +{ + // ABI compatibility + void* placeholder = VMA_NULL; + VMA_RW_MUTEX placeholder2; +}; +#endif // VMA_EXTERNAL_MEMORY_WIN32 + + +#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_NO_MOVE(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(VmaAllocator hAllocator); + 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); +#if VMA_EXTERNAL_MEMORY_WIN32 + VkResult CreateWin32Handle( + const VmaAllocator hAllocator, + PFN_vkGetMemoryWin32HandleKHR pvkGetMemoryWin32HandleKHR, + HANDLE hTargetProcess, + HANDLE* pHandle)noexcept; +#endif // VMA_EXTERNAL_MEMORY_WIN32 +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; + + VmaWin32Handle m_Handle; +}; +#endif // _VMA_DEVICE_MEMORY_BLOCK + +#ifndef _VMA_ALLOCATION_T +struct VmaAllocationExtraData +{ + void* m_pMappedData = VMA_NULL; // Not null means memory is mapped. + VmaWin32Handle m_Handle; +}; + +struct VmaAllocation_T +{ + friend struct VmaDedicatedAllocationListItemTraits; + + enum FLAGS + { + FLAG_PERSISTENT_MAP = 0x01, + FLAG_MAPPING_ALLOWED = 0x02, + }; + +public: + enum ALLOCATION_TYPE + { + ALLOCATION_TYPE_NONE, + ALLOCATION_TYPE_BLOCK, + ALLOCATION_TYPE_DEDICATED, + }; + + // This struct is allocated using VmaPoolAllocator. + VmaAllocation_T(bool mappingAllowed); + ~VmaAllocation_T(); + + void InitBlockAllocation( + VmaDeviceMemoryBlock* block, + VmaAllocHandle allocHandle, + VkDeviceSize alignment, + VkDeviceSize size, + uint32_t memoryTypeIndex, + VmaSuballocationType suballocationType, + bool mapped); + // pMappedData not null means allocation is created with MAPPED flag. + void InitDedicatedAllocation( + VmaAllocator allocator, + VmaPool hParentPool, + uint32_t memoryTypeIndex, + VkDeviceMemory hMemory, + VmaSuballocationType suballocationType, + void* pMappedData, + VkDeviceSize size); + void Destroy(VmaAllocator allocator); + + ALLOCATION_TYPE GetType() const { return (ALLOCATION_TYPE)m_Type; } + VkDeviceSize GetAlignment() const { return m_Alignment; } + VkDeviceSize GetSize() const { return m_Size; } + void* GetUserData() const { return m_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; } + uint32_t GetMemoryTypeIndex() const { return m_MemoryTypeIndex; } + 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; + + void BlockAllocMap(); + void BlockAllocUnmap(); + VkResult DedicatedAllocMap(VmaAllocator hAllocator, void** ppData); + void DedicatedAllocUnmap(VmaAllocator hAllocator); + +#if VMA_STATS_STRING_ENABLED + VmaBufferImageUsage GetBufferImageUsage() const { return m_BufferImageUsage; } + void InitBufferUsage(const VkBufferCreateInfo &createInfo, bool useKhrMaintenance5) + { + VMA_ASSERT(m_BufferImageUsage == VmaBufferImageUsage::UNKNOWN); + m_BufferImageUsage = VmaBufferImageUsage(createInfo, useKhrMaintenance5); + } + void InitImageUsage(const VkImageCreateInfo &createInfo) + { + VMA_ASSERT(m_BufferImageUsage == VmaBufferImageUsage::UNKNOWN); + m_BufferImageUsage = VmaBufferImageUsage(createInfo); + } + void PrintParameters(class VmaJsonWriter& json) const; +#endif + +#if VMA_EXTERNAL_MEMORY_WIN32 + VkResult GetWin32Handle(VmaAllocator hAllocator, HANDLE hTargetProcess, HANDLE* hHandle) noexcept; +#endif // VMA_EXTERNAL_MEMORY_WIN32 + +private: + // Allocation out of VmaDeviceMemoryBlock. + struct BlockAllocation + { + VmaDeviceMemoryBlock* m_Block; + 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; + VmaAllocationExtraData* m_ExtraData; + VmaAllocation_T* m_Prev; + VmaAllocation_T* m_Next; + }; + union + { + // Allocation out of VmaDeviceMemoryBlock. + BlockAllocation m_BlockAllocation; + // Allocation for an object that has its own private VkDeviceMemory. + 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 + VmaBufferImageUsage m_BufferImageUsage; // 0 if unknown. +#endif + + void EnsureExtraData(VmaAllocator hAllocator); +}; +#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); + return item->m_DedicatedAllocation.m_Prev; + } + static ItemType* GetNext(const ItemType* item) + { + VMA_HEAVY_ASSERT(item->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); + return item->m_DedicatedAllocation.m_Next; + } + static ItemType*& AccessPrev(ItemType* item) + { + VMA_HEAVY_ASSERT(item->GetType() == VmaAllocation_T::ALLOCATION_TYPE_DEDICATED); + return item->m_DedicatedAllocation.m_Prev; + } + 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 +{ + VMA_CLASS_NO_COPY_NO_MOVE(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_LEAK(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 != VMA_NULL; 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 != VMA_NULL; 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. +*/ +struct VmaSuballocation +{ + VkDeviceSize offset; + VkDeviceSize size; + void* userData; + VmaSuballocationType type; +}; + +// Comparator for offsets. +struct VmaSuballocationOffsetLess +{ + bool operator()(const VmaSuballocation& lhs, const VmaSuballocation& rhs) const + { + return lhs.offset < rhs.offset; + } +}; + +struct VmaSuballocationOffsetGreater +{ + bool operator()(const VmaSuballocation& lhs, const VmaSuballocation& rhs) const + { + return lhs.offset > rhs.offset; + } +}; + +struct VmaSuballocationItemSizeLess +{ + bool operator()(const VmaSuballocationList::iterator lhs, + const VmaSuballocationList::iterator rhs) const + { + return lhs->size < rhs->size; + } + + bool operator()(const VmaSuballocationList::iterator lhs, + VkDeviceSize rhsSize) const + { + return lhs->size < rhsSize; + } +}; +#endif // _VMA_SUBALLOCATION + +#ifndef _VMA_ALLOCATION_REQUEST +/* +Parameters of planned allocation inside a VmaDeviceMemoryBlock. +item points to a FREE suballocation. +*/ +struct VmaAllocationRequest +{ + VmaAllocHandle allocHandle; + VkDeviceSize size; + VmaSuballocationList::iterator item; + void* customData; + uint64_t algorithmData; + VmaAllocationRequestType type; +}; +#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. +*/ +class VmaBlockMetadata +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaBlockMetadata) +public: + // 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; + virtual size_t GetAllocationCount() const = 0; + virtual size_t GetFreeRegionsCount() const = 0; + virtual VkDeviceSize GetSumFreeSize() 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; + + // Shouldn't modify blockCount. + 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; +#endif + + // Tries to find a place for suballocation with given parameters inside this block. + // If succeeded, fills pAllocationRequest and returns true. + // If failed, returns false. + virtual bool CreateAllocationRequest( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + bool upperAddress, + VmaSuballocationType allocType, + // Always one of VMA_ALLOCATION_CREATE_STRATEGY_* or VMA_ALLOCATION_INTERNAL_STRATEGY_* flags. + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest) = 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, + void* userData) = 0; + + // Frees suballocation assigned to given memory region. + 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 VkDeviceSize(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, VkDeviceSize size, void* userData) const; + void PrintDetailedMap_UnusedRange(class VmaJsonWriter& json, + VkDeviceSize offset, + VkDeviceSize size) const; + void PrintDetailedMap_End(class VmaJsonWriter& json) const; +#endif + +private: + VkDeviceSize m_Size; + const VkAllocationCallbacks* m_pAllocationCallbacks; + const VkDeviceSize m_BufferImageGranularity; + const bool m_IsVirtual; +}; + +#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) {} + +void VmaBlockMetadata::DebugLogAllocation(VkDeviceSize offset, VkDeviceSize size, void* userData) const +{ + if (IsVirtual()) + { + VMA_LEAK_LOG_FORMAT("UNFREED VIRTUAL ALLOCATION; Offset: %" PRIu64 "; Size: %" PRIu64 "; UserData: %p", offset, size, userData); + } + else + { + VMA_ASSERT(userData != VMA_NULL); + VmaAllocation allocation = reinterpret_cast(userData); + + userData = allocation->GetUserData(); + const char* name = allocation->GetName(); + +#if VMA_STATS_STRING_ENABLED + VMA_LEAK_LOG_FORMAT("UNFREED ALLOCATION; Offset: %" PRIu64 "; Size: %" PRIu64 "; UserData: %p; Name: %s; Type: %s; Usage: %" PRIu64, + offset, size, userData, name ? name : "vma_empty", + VMA_SUBALLOCATION_TYPE_NAMES[allocation->GetSuballocationType()], + (uint64_t)allocation->GetBufferImageUsage().Value); +#else + VMA_LEAK_LOG_FORMAT("UNFREED ALLOCATION; Offset: %" PRIu64 "; Size: %" PRIu64 "; UserData: %p; Name: %s; Type: %u", + offset, size, userData, name ? name : "vma_empty", + (unsigned)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.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, 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 overridden 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 + +#ifndef _VMA_BLOCK_METADATA_LINEAR +/* +Allocations and their references in internal data structure look like this: + +if(m_2ndVectorMode == SECOND_VECTOR_EMPTY): + + 0 +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | + | | + | | +GetSize() +-------+ + +if(m_2ndVectorMode == SECOND_VECTOR_RING_BUFFER): + + 0 +-------+ + | Alloc | 2nd[0] + +-------+ + | Alloc | 2nd[1] + +-------+ + | ... | + +-------+ + | Alloc | 2nd[2nd.size() - 1] + +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | +GetSize() +-------+ + +if(m_2ndVectorMode == SECOND_VECTOR_DOUBLE_STACK): + + 0 +-------+ + | | + | | + | | + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount] + +-------+ + | Alloc | 1st[m_1stNullItemsBeginCount + 1] + +-------+ + | ... | + +-------+ + | Alloc | 1st[1st.size() - 1] + +-------+ + | | + | | + | | + +-------+ + | Alloc | 2nd[2nd.size() - 1] + +-------+ + | ... | + +-------+ + | Alloc | 2nd[1] + +-------+ + | Alloc | 2nd[0] +GetSize() +-------+ + +*/ +class VmaBlockMetadata_Linear : public VmaBlockMetadata +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaBlockMetadata_Linear) +public: + VmaBlockMetadata_Linear(const VkAllocationCallbacks* pAllocationCallbacks, + VkDeviceSize bufferImageGranularity, bool isVirtual); + virtual ~VmaBlockMetadata_Linear() = default; + + VkDeviceSize GetSumFreeSize() const override { return m_SumFreeSize; } + bool IsEmpty() const override { return GetAllocationCount() == 0; } + VkDeviceSize GetAllocationOffset(VmaAllocHandle allocHandle) const override { return (VkDeviceSize)allocHandle - 1; } + + 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 + 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: + /* + There are two suballocation vectors, used in ping-pong way. + The one with index m_1stVectorIndex is called 1st. + The one with index (m_1stVectorIndex ^ 1) is called 2nd. + 2nd can be non-empty only when 1st is not empty. + When 2nd is not empty, m_2ndVectorMode indicates its mode of operation. + */ + typedef VmaVector> SuballocationVectorType; + + enum SECOND_VECTOR_MODE + { + SECOND_VECTOR_EMPTY, + /* + Suballocations in 2nd vector are created later than the ones in 1st, but they + all have smaller offset. + */ + SECOND_VECTOR_RING_BUFFER, + /* + Suballocations in 2nd vector are upper side of double stack. + They all have offsets higher than those in 1st vector. + Top of this stack means smaller offsets, but higher indices in this vector. + */ + SECOND_VECTOR_DOUBLE_STACK, + }; + + VkDeviceSize m_SumFreeSize; + SuballocationVectorType m_Suballocations0, m_Suballocations1; + uint32_t m_1stVectorIndex; + SECOND_VECTOR_MODE m_2ndVectorMode; + // Number of items in 1st vector with hAllocation = null at the beginning. + size_t m_1stNullItemsBeginCount; + // Number of other items in 1st vector with hAllocation = null somewhere in the middle. + size_t m_1stNullItemsMiddleCount; + // Number of items in 2nd vector with hAllocation = null. + size_t m_2ndNullItemsCount; + + 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( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + uint32_t strategy, + VmaAllocationRequest* pAllocationRequest); + bool CreateAllocationRequest_UpperAddress( + VkDeviceSize allocSize, + VkDeviceSize allocAlignment, + VmaSuballocationType allocType, + 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]; + + // Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc2ndIndex; + } + // We are at the end. + else + { + // 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]; + + // Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + ++nextAlloc1stIndex; + } + // We are at the end. + else + { + // 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]; + + // Process this allocation. + // There is allocation with suballoc.offset, suballoc.size. + ++inoutStats.allocationCount; + + // Prepare for next iteration. + lastOffset = suballoc.offset + suballoc.size; + --nextAlloc2ndIndex; + } + // We are at the end. + else + { + // 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 < freeSpace1stTo2ndEnd) + { + // 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()); + + if(allocSize > GetSize()) + return false; + + 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 + +#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_NO_MOVE(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: + // 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; + /* + * 0: 0-3 lists for small buffers + * 1+: 0-(2^SLI-1) lists for normal buffers + */ + 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); +}; + +#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() +{ + if (m_FreeList) + vma_delete_array(GetAllocationCallbacks(), m_FreeList, m_ListsCount); + m_GranularityHandler.Destroy(GetAllocationCallbacks()); +} + +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 + uint8_t(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 = VkDeviceSize(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 = m_ListsCount; + uint32_t prevListIndex = m_ListsCount; + 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 missingAlignment = offset - currentBlock->offset; + + // Append missing alignment to prev block or create new one + if (missingAlignment) + { + 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 += missingAlignment; + // Check if new size crosses list bucket + if (oldList != GetListIndex(prevBlock->size)) + { + prevBlock->size -= missingAlignment; + RemoveFreeBlock(prevBlock); + prevBlock->size += missingAlignment; + InsertFreeBlock(prevBlock); + } + else + m_BlocksFreeSize += missingAlignment; + } + else + { + Block* newBlock = m_BlockAllocator.Alloc(); + currentBlock->prevPhysical = newBlock; + prevBlock->nextPhysical = newBlock; + newBlock->prevPhysical = prevBlock; + newBlock->nextPhysical = currentBlock; + newBlock->size = missingAlignment; + newBlock->offset = currentBlock->offset; + newBlock->MarkTaken(); + + InsertFreeBlock(newBlock); + } + + currentBlock->size -= missingAlignment; + currentBlock->offset += missingAlignment; + } + + 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 uint8_t(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 separate 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 available blocks + uint32_t freeMap = m_IsFreeBitmap & (~0UL << (memoryClass + 1)); + if (!freeMap) + return VMA_NULL; // No more memory available + + // 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. +*/ +class VmaBlockVector +{ + friend struct VmaDefragmentationContext_T; + VMA_CLASS_NO_COPY_NO_MOVE(VmaBlockVector) +public: + VmaBlockVector( + VmaAllocator hAllocator, + VmaPool hParentPool, + uint32_t memoryTypeIndex, + VkDeviceSize preferredBlockSize, + size_t minBlockCount, + size_t maxBlockCount, + VkDeviceSize bufferImageGranularity, + bool explicitBlockSize, + uint32_t algorithm, + float priority, + VkDeviceSize minAllocationAlignment, + void* pMemoryAllocateNext); + ~VmaBlockVector(); + + 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 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; } + + VkResult CreateMinBlocks(); + void AddStatistics(VmaStatistics& inoutStats); + void AddDetailedStatistics(VmaDetailedStatistics& inoutStats); + bool IsEmpty(); + bool IsCorruptionDetectionEnabled() const; + + VkResult Allocate( + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations); + + void Free(const VmaAllocation hAllocation); + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json); +#endif + + VkResult CheckCorruption(); + +private: + const VmaAllocator m_hAllocator; + const VmaPool m_hParentPool; + const uint32_t m_MemoryTypeIndex; + const VkDeviceSize m_PreferredBlockSize; + const size_t m_MinBlockCount; + const size_t m_MaxBlockCount; + const VkDeviceSize m_BufferImageGranularity; + 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; + // Incrementally sorted by sumFreeSize, ascending. + 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( + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation); + + VkResult AllocateFromBlock( + VmaDeviceMemoryBlock* pBlock, + VkDeviceSize size, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + 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); + bool HasEmptyBlock(); +}; +#endif // _VMA_BLOCK_VECTOR + +#ifndef _VMA_DEFRAGMENTATION_CONTEXT +struct VmaDefragmentationContext_T +{ + VMA_CLASS_NO_COPY_NO_MOVE(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; + const PFN_vmaCheckDefragmentationBreakFunction m_BreakCallback; + void* m_BreakCallbackUserData; + + 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_NO_MOVE(VmaPool_T) +public: + VmaBlockVector m_BlockVector; + VmaDedicatedAllocationList m_DedicatedAllocations; + + VmaPool_T( + VmaAllocator hAllocator, + const VmaPoolCreateInfo& createInfo, + VkDeviceSize preferredBlockSize); + ~VmaPool_T(); + + uint32_t GetId() const { return m_Id; } + void SetId(uint32_t id) { VMA_ASSERT(m_Id == 0); m_Id = id; } + + const char* GetName() const { return m_Name; } + void SetName(const char* pName); + +#if VMA_STATS_STRING_ENABLED + //void PrintDetailedMap(class VmaStringBuilder& sb); +#endif + +private: + uint32_t m_Id; + char* m_Name; + VmaPool_T* m_PrevPool = VMA_NULL; + VmaPool_T* m_NextPool = VMA_NULL; +}; + +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 + +#ifndef _VMA_CURRENT_BUDGET_DATA +struct VmaCurrentBudgetData +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaCurrentBudgetData) +public: + + 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]; + +#if VMA_MEMORY_BUDGET + VMA_ATOMIC_UINT32 m_OperationsSinceBudgetFetch; + VMA_RW_MUTEX m_BudgetMutex; + 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 // VMA_MEMORY_BUDGET + + VmaCurrentBudgetData(); + + 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_NO_MOVE(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_NO_MOVE(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) + { + 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; + default: + VMA_ASSERT(0); + m_Metadata = vma_new(GetAllocationCallbacks(), VmaBlockMetadata_TLSF)(VK_NULL_HANDLE, 1, true); + } + + m_Metadata->Init(createInfo.size); +} + +VmaVirtualBlock_T::~VmaVirtualBlock_T() +{ + // Define macro VMA_DEBUG_LOG_FORMAT or more specialized VMA_LEAK_LOG_FORMAT + // 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_LEAK(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 +{ + VMA_CLASS_NO_COPY_NO_MOVE(VmaAllocator_T) +public: + const bool m_UseMutex; + const uint32_t m_VulkanApiVersion; + bool m_UseKhrDedicatedAllocation; // Can be set only if m_VulkanApiVersion < VK_MAKE_VERSION(1, 1, 0). + bool m_UseKhrBindMemory2; // Can be set only if m_VulkanApiVersion < VK_MAKE_VERSION(1, 1, 0). + bool m_UseExtMemoryBudget; + bool m_UseAmdDeviceCoherentMemory; + bool m_UseKhrBufferDeviceAddress; + bool m_UseExtMemoryPriority; + bool m_UseKhrMaintenance4; + bool m_UseKhrMaintenance5; + bool m_UseKhrExternalMemoryWin32; + const VkDevice m_hDevice; + const VkInstance m_hInstance; + const bool m_AllocationCallbacksSpecified; + const VkAllocationCallbacks m_AllocationCallbacks; + VmaDeviceMemoryCallbacks m_DeviceMemoryCallbacks; + VmaAllocationObjectAllocator m_AllocationObjectAllocator; + + // Each bit (1 << i) is set if HeapSizeLimit is enabled for that heap, so cannot allocate more than the heap size. + uint32_t m_HeapSizeLimitMask; + + VkPhysicalDeviceProperties m_PhysicalDeviceProperties; + VkPhysicalDeviceMemoryProperties m_MemProps; + + // Default pools. + VmaBlockVector* m_pBlockVectors[VK_MAX_MEMORY_TYPES]; + VmaDedicatedAllocationList m_DedicatedAllocations[VK_MAX_MEMORY_TYPES]; + + VmaCurrentBudgetData m_Budget; + VMA_ATOMIC_UINT32 m_DeviceMemoryCount; // Total number of VkDeviceMemory objects. + + VmaAllocator_T(const VmaAllocatorCreateInfo* pCreateInfo); + VkResult Init(const VmaAllocatorCreateInfo* pCreateInfo); + ~VmaAllocator_T(); + + const VkAllocationCallbacks* GetAllocationCallbacks() const + { + return m_AllocationCallbacksSpecified ? &m_AllocationCallbacks : VMA_NULL; + } + const VmaVulkanFunctions& GetVulkanFunctions() const + { + return m_VulkanFunctions; + } + + VkPhysicalDevice GetPhysicalDevice() const { return m_PhysicalDevice; } + + VkDeviceSize GetBufferImageGranularity() const + { + return VMA_MAX( + static_cast(VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY), + m_PhysicalDeviceProperties.limits.bufferImageGranularity); + } + + uint32_t GetMemoryHeapCount() const { return m_MemProps.memoryHeapCount; } + uint32_t GetMemoryTypeCount() const { return m_MemProps.memoryTypeCount; } + + uint32_t MemoryTypeIndexToHeapIndex(uint32_t memTypeIndex) const + { + VMA_ASSERT(memTypeIndex < m_MemProps.memoryTypeCount); + return m_MemProps.memoryTypes[memTypeIndex].heapIndex; + } + // True when specific memory type is HOST_VISIBLE but not HOST_COHERENT. + bool IsMemoryTypeNonCoherent(uint32_t memTypeIndex) const + { + return (m_MemProps.memoryTypes[memTypeIndex].propertyFlags & (VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT)) == + VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; + } + // Minimum alignment for all allocations in specific memory type. + VkDeviceSize GetMemoryTypeMinAlignment(uint32_t memTypeIndex) const + { + return IsMemoryTypeNonCoherent(memTypeIndex) ? + VMA_MAX((VkDeviceSize)VMA_MIN_ALIGNMENT, m_PhysicalDeviceProperties.limits.nonCoherentAtomSize) : + (VkDeviceSize)VMA_MIN_ALIGNMENT; + } + + bool IsIntegratedGpu() const + { + return m_PhysicalDeviceProperties.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU; + } + + uint32_t GetGlobalMemoryTypeBits() const { return m_GlobalMemoryTypeBits; } + + void GetBufferMemoryRequirements( + VkBuffer hBuffer, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const; + void GetImageMemoryRequirements( + VkImage hImage, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const; + VkResult FindMemoryTypeIndex( + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VmaBufferImageUsage bufImgUsage, + uint32_t* pMemoryTypeIndex) const; + + // Main allocation function. + VkResult AllocateMemory( + const VkMemoryRequirements& vkMemReq, + bool requiresDedicatedAllocation, + bool prefersDedicatedAllocation, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + VmaBufferImageUsage dedicatedBufferImageUsage, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations); + + // Main deallocation function. + void FreeMemory( + size_t allocationCount, + const VmaAllocation* pAllocations); + + void CalculateStatistics(VmaTotalStatistics* pStats); + + void GetHeapBudgets( + VmaBudget* outBudgets, uint32_t firstHeap, uint32_t heapCount); + +#if VMA_STATS_STRING_ENABLED + void PrintDetailedMap(class VmaJsonWriter& json); +#endif + + void GetAllocationInfo(VmaAllocation hAllocation, VmaAllocationInfo* pAllocationInfo); + void GetAllocationInfo2(VmaAllocation hAllocation, VmaAllocationInfo2* pAllocationInfo); + + VkResult CreatePool(const VmaPoolCreateInfo* pCreateInfo, VmaPool* pPool); + void DestroyPool(VmaPool pool); + 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(); } + + VkResult CheckPoolCorruption(VmaPool hPool); + VkResult CheckCorruption(uint32_t memoryTypeBits); + + // Call to Vulkan function vkAllocateMemory with accompanying bookkeeping. + VkResult AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory); + // Call to Vulkan function vkFreeMemory with accompanying bookkeeping. + void FreeVulkanMemory(uint32_t memoryType, VkDeviceSize size, VkDeviceMemory hMemory); + // Call to Vulkan function vkBindBufferMemory or vkBindBufferMemory2KHR. + VkResult BindVulkanBuffer( + VkDeviceMemory memory, + VkDeviceSize memoryOffset, + VkBuffer buffer, + const void* pNext); + // Call to Vulkan function vkBindImageMemory or vkBindImageMemory2KHR. + VkResult BindVulkanImage( + VkDeviceMemory memory, + VkDeviceSize memoryOffset, + VkImage image, + const void* pNext); + + VkResult Map(VmaAllocation hAllocation, void** ppData); + void Unmap(VmaAllocation hAllocation); + + VkResult BindBufferMemory( + VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkBuffer hBuffer, + const void* pNext); + VkResult BindImageMemory( + VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkImage hImage, + const void* pNext); + + VkResult FlushOrInvalidateAllocation( + VmaAllocation hAllocation, + VkDeviceSize offset, VkDeviceSize size, + VMA_CACHE_OPERATION op); + VkResult FlushOrInvalidateAllocations( + uint32_t allocationCount, + const VmaAllocation* allocations, + const VkDeviceSize* offsets, const VkDeviceSize* sizes, + VMA_CACHE_OPERATION op); + + VkResult CopyMemoryToAllocation( + const void* pSrcHostPointer, + VmaAllocation dstAllocation, + VkDeviceSize dstAllocationLocalOffset, + VkDeviceSize size); + VkResult CopyAllocationToMemory( + VmaAllocation srcAllocation, + VkDeviceSize srcAllocationLocalOffset, + void* pDstHostPointer, + VkDeviceSize size); + + void FillAllocation(const VmaAllocation hAllocation, uint8_t pattern); + + /* + Returns bit mask of memory types that can support defragmentation on GPU as + they support creation of required buffer for copy operations. + */ + uint32_t GetGpuDefragmentationMemoryTypeBits(); + +#if VMA_EXTERNAL_MEMORY + VkExternalMemoryHandleTypeFlagsKHR GetExternalMemoryHandleTypeFlags(uint32_t memTypeIndex) const + { + return m_TypeExternalMemoryHandleTypes[memTypeIndex]; + } +#endif // #if VMA_EXTERNAL_MEMORY + +private: + VkDeviceSize m_PreferredLargeHeapBlockSize; + + VkPhysicalDevice m_PhysicalDevice; + VMA_ATOMIC_UINT32 m_CurrentFrameIndex; + VMA_ATOMIC_UINT32 m_GpuDefragmentationMemoryTypeBits; // UINT32_MAX means uninitialized. +#if VMA_EXTERNAL_MEMORY + VkExternalMemoryHandleTypeFlagsKHR m_TypeExternalMemoryHandleTypes[VK_MAX_MEMORY_TYPES]; +#endif // #if VMA_EXTERNAL_MEMORY + + VMA_RW_MUTEX m_PoolsMutex; + typedef VmaIntrusiveLinkedList PoolList; + // Protected by m_PoolsMutex. + PoolList m_Pools; + uint32_t m_NextPoolId; + + VmaVulkanFunctions m_VulkanFunctions; + + // Global bit mask AND-ed with any memoryTypeBits to disallow certain memory types. + uint32_t m_GlobalMemoryTypeBits; + + void ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunctions); + +#if VMA_STATIC_VULKAN_FUNCTIONS == 1 + void ImportVulkanFunctions_Static(); +#endif + + void ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVulkanFunctions); + +#if VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 + void ImportVulkanFunctions_Dynamic(); +#endif + + void ValidateVulkanFunctions(); + + VkDeviceSize CalcPreferredBlockSize(uint32_t memTypeIndex); + + VkResult AllocateMemoryOfType( + VmaPool pool, + VkDeviceSize size, + VkDeviceSize alignment, + bool dedicatedPreferred, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + VmaBufferImageUsage 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 map, + bool isUserDataString, + bool isMappingAllowed, + bool canAliasMemory, + void* pUserData, + float priority, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + VmaBufferImageUsage dedicatedBufferImageUsage, + size_t allocationCount, + VmaAllocation* pAllocations, + const void* pNextChain = VMA_NULL); + + 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( + VmaAllocation allocation, + VkDeviceSize offset, VkDeviceSize size, + VkMappedMemoryRange& outRange) const; + +#if VMA_MEMORY_BUDGET + void UpdateVulkanBudget(); +#endif // #if VMA_MEMORY_BUDGET +}; + + +#ifndef _VMA_MEMORY_FUNCTIONS +static void* VmaMalloc(VmaAllocator hAllocator, size_t size, size_t alignment) +{ + return VmaMalloc(&hAllocator->m_AllocationCallbacks, size, alignment); +} + +static void VmaFree(VmaAllocator hAllocator, void* ptr) +{ + VmaFree(&hAllocator->m_AllocationCallbacks, ptr); +} + +template +static T* VmaAllocate(VmaAllocator hAllocator) +{ + return (T*)VmaMalloc(hAllocator, sizeof(T), VMA_ALIGN_OF(T)); +} + +template +static T* VmaAllocateArray(VmaAllocator hAllocator, size_t count) +{ + return (T*)VmaMalloc(hAllocator, sizeof(T) * count, VMA_ALIGN_OF(T)); +} + +template +static void vma_delete(VmaAllocator hAllocator, T* ptr) +{ + if(ptr != VMA_NULL) + { + ptr->~T(); + VmaFree(hAllocator, ptr); + } +} + +template +static void vma_delete_array(VmaAllocator hAllocator, T* ptr, size_t count) +{ + if(ptr != VMA_NULL) + { + for(size_t i = count; i--; ) + ptr[i].~T(); + VmaFree(hAllocator, ptr); + } +} +#endif // _VMA_MEMORY_FUNCTIONS + +#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){} + +VmaDeviceMemoryBlock::~VmaDeviceMemoryBlock() +{ + VMA_ASSERT_LEAK(m_MapCount == 0 && "VkDeviceMemory block is being destroyed while it is still mapped."); + VMA_ASSERT_LEAK(m_hMemory == VK_NULL_HANDLE); +} + +void VmaDeviceMemoryBlock::Init( + VmaAllocator hAllocator, + VmaPool hParentPool, + uint32_t newMemoryTypeIndex, + VkDeviceMemory newMemory, + VkDeviceSize newSize, + uint32_t id, + uint32_t algorithm, + VkDeviceSize bufferImageGranularity) +{ + VMA_ASSERT(m_hMemory == VK_NULL_HANDLE); + + m_hParentPool = hParentPool; + m_MemoryTypeIndex = newMemoryTypeIndex; + m_Id = id; + m_hMemory = newMemory; + + switch (algorithm) + { + case 0: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_TLSF)(hAllocator->GetAllocationCallbacks(), + bufferImageGranularity, false); // isVirtual + break; + case VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT: + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_Linear)(hAllocator->GetAllocationCallbacks(), + bufferImageGranularity, false); // isVirtual + break; + default: + VMA_ASSERT(0); + m_pMetadata = vma_new(hAllocator, VmaBlockMetadata_TLSF)(hAllocator->GetAllocationCallbacks(), + bufferImageGranularity, false); // isVirtual + } + m_pMetadata->Init(newSize); +} + +void VmaDeviceMemoryBlock::Destroy(VmaAllocator allocator) +{ + // Define macro VMA_DEBUG_LOG_FORMAT or more specialized VMA_LEAK_LOG_FORMAT + // 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_LEAK(m_pMetadata->IsEmpty() && "Some allocations were not freed before destruction of this memory block!"); + + VMA_ASSERT_LEAK(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::PostAlloc(VmaAllocator hAllocator) +{ + VmaMutexLock lock(m_MapAndBindMutex, hAllocator->m_UseMutex); + m_MappingHysteresis.PostAlloc(); +} + +void VmaDeviceMemoryBlock::PostFree(VmaAllocator hAllocator) +{ + VmaMutexLock lock(m_MapAndBindMutex, hAllocator->m_UseMutex); + if(m_MappingHysteresis.PostFree()) + { + VMA_ASSERT(m_MappingHysteresis.GetExtraMapping() == 0); + if (m_MapCount == 0) + { + m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)(hAllocator->m_hDevice, m_hMemory); + } + } +} + +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 = VMA_NULL; + 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_MapAndBindMutex, hAllocator->m_UseMutex); + const uint32_t oldTotalMapCount = m_MapCount + m_MappingHysteresis.GetExtraMapping(); + if (oldTotalMapCount != 0) + { + VMA_ASSERT(m_pMappedData != VMA_NULL); + m_MappingHysteresis.PostMap(); + m_MapCount += count; + 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) + { + VMA_ASSERT(m_pMappedData != VMA_NULL); + m_MappingHysteresis.PostMap(); + m_MapCount = count; + if (ppData != VMA_NULL) + { + *ppData = m_pMappedData; + } + } + return result; + } +} + +void VmaDeviceMemoryBlock::Unmap(VmaAllocator hAllocator, uint32_t count) +{ + if (count == 0) + { + return; + } + + 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."); + } +} + +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); +} + +#if VMA_EXTERNAL_MEMORY_WIN32 +VkResult VmaDeviceMemoryBlock::CreateWin32Handle(const VmaAllocator hAllocator, PFN_vkGetMemoryWin32HandleKHR pvkGetMemoryWin32HandleKHR, HANDLE hTargetProcess, HANDLE* pHandle) noexcept +{ + VMA_ASSERT(pHandle); + return m_Handle.GetHandle(hAllocator->m_hDevice, m_hMemory, pvkGetMemoryWin32HandleKHR, hTargetProcess, hAllocator->m_UseMutex, pHandle); +} +#endif // VMA_EXTERNAL_MEMORY_WIN32 +#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; +} + +VmaAllocation_T::~VmaAllocation_T() +{ + VMA_ASSERT_LEAK(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( + VmaAllocator allocator, + 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_DedicatedAllocation.m_ExtraData = VMA_NULL; + m_DedicatedAllocation.m_hParentPool = hParentPool; + m_DedicatedAllocation.m_hMemory = hMemory; + m_DedicatedAllocation.m_Prev = VMA_NULL; + m_DedicatedAllocation.m_Next = VMA_NULL; + + 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; + EnsureExtraData(allocator); + m_DedicatedAllocation.m_ExtraData->m_pMappedData = pMappedData; + } +} + +void VmaAllocation_T::Destroy(VmaAllocator allocator) +{ + FreeName(allocator); + + if (GetType() == ALLOCATION_TYPE_DEDICATED) + { + vma_delete(allocator, m_DedicatedAllocation.m_ExtraData); + } +} + +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); + 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); + std::swap(m_BlockAllocation, allocation->m_BlockAllocation); + m_BlockAllocation.m_Block->m_pMetadata->SetAllocationUserData(m_BlockAllocation.m_AllocHandle, this); + +#if VMA_STATS_STRING_ENABLED + std::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) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->m_pMetadata->GetAllocationOffset(m_BlockAllocation.m_AllocHandle); + case ALLOCATION_TYPE_DEDICATED: + return 0; + default: + VMA_ASSERT(0); + return 0; + } +} + +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) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->GetDeviceMemory(); + case ALLOCATION_TYPE_DEDICATED: + return m_DedicatedAllocation.m_hMemory; + default: + VMA_ASSERT(0); + return VK_NULL_HANDLE; + } +} + +void* VmaAllocation_T::GetMappedData() const +{ + switch (m_Type) + { + case ALLOCATION_TYPE_BLOCK: + if (m_MapCount != 0 || IsPersistentMap()) + { + void* pBlockData = m_BlockAllocation.m_Block->GetMappedData(); + VMA_ASSERT(pBlockData != VMA_NULL); + return (char*)pBlockData + GetOffset(); + } + else + { + return VMA_NULL; + } + break; + case ALLOCATION_TYPE_DEDICATED: + VMA_ASSERT((m_DedicatedAllocation.m_ExtraData != VMA_NULL && m_DedicatedAllocation.m_ExtraData->m_pMappedData != VMA_NULL) == + (m_MapCount != 0 || IsPersistentMap())); + return m_DedicatedAllocation.m_ExtraData != VMA_NULL ? m_DedicatedAllocation.m_ExtraData->m_pMappedData : VMA_NULL; + default: + VMA_ASSERT(0); + return 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 < 0xFF) + { + ++m_MapCount; + } + else + { + VMA_ASSERT(0 && "Allocation mapped too many times simultaneously."); + } +} + +void VmaAllocation_T::BlockAllocUnmap() +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_BLOCK); + + if (m_MapCount > 0) + { + --m_MapCount; + } + else + { + VMA_ASSERT(0 && "Unmapping allocation not previously mapped."); + } +} + +VkResult VmaAllocation_T::DedicatedAllocMap(VmaAllocator hAllocator, void** ppData) +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); + 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."); + + EnsureExtraData(hAllocator); + + if (m_MapCount != 0 || IsPersistentMap()) + { + if (m_MapCount < 0xFF) + { + VMA_ASSERT(m_DedicatedAllocation.m_ExtraData->m_pMappedData != VMA_NULL); + *ppData = m_DedicatedAllocation.m_ExtraData->m_pMappedData; + ++m_MapCount; + return VK_SUCCESS; + } + else + { + VMA_ASSERT(0 && "Dedicated allocation mapped too many times simultaneously."); + return VK_ERROR_MEMORY_MAP_FAILED; + } + } + else + { + VkResult result = (*hAllocator->GetVulkanFunctions().vkMapMemory)( + hAllocator->m_hDevice, + m_DedicatedAllocation.m_hMemory, + 0, // offset + VK_WHOLE_SIZE, + 0, // flags + ppData); + if (result == VK_SUCCESS) + { + m_DedicatedAllocation.m_ExtraData->m_pMappedData = *ppData; + m_MapCount = 1; + } + return result; + } +} + +void VmaAllocation_T::DedicatedAllocUnmap(VmaAllocator hAllocator) +{ + VMA_ASSERT(GetType() == ALLOCATION_TYPE_DEDICATED); + + if (m_MapCount > 0) + { + --m_MapCount; + if (m_MapCount == 0 && !IsPersistentMap()) + { + VMA_ASSERT(m_DedicatedAllocation.m_ExtraData != VMA_NULL); + m_DedicatedAllocation.m_ExtraData->m_pMappedData = VMA_NULL; + (*hAllocator->GetVulkanFunctions().vkUnmapMemory)( + hAllocator->m_hDevice, + m_DedicatedAllocation.m_hMemory); + } + } + else + { + VMA_ASSERT(0 && "Unmapping dedicated allocation not previously mapped."); + } +} + +#if VMA_STATS_STRING_ENABLED +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); + json.WriteString("Usage"); + json.WriteNumber(m_BufferImageUsage.Value); // It may be uint32_t or uint64_t. + + if (m_pUserData != VMA_NULL) + { + json.WriteString("CustomData"); + json.BeginString(); + json.ContinueString_Pointer(m_pUserData); + json.EndString(); + } + if (m_pName != VMA_NULL) + { + json.WriteString("Name"); + json.WriteString(m_pName); + } +} +#if VMA_EXTERNAL_MEMORY_WIN32 +VkResult VmaAllocation_T::GetWin32Handle(VmaAllocator hAllocator, HANDLE hTargetProcess, HANDLE* pHandle) noexcept +{ + auto pvkGetMemoryWin32HandleKHR = hAllocator->GetVulkanFunctions().vkGetMemoryWin32HandleKHR; + switch (m_Type) + { + case ALLOCATION_TYPE_BLOCK: + return m_BlockAllocation.m_Block->CreateWin32Handle(hAllocator, pvkGetMemoryWin32HandleKHR, hTargetProcess, pHandle); + case ALLOCATION_TYPE_DEDICATED: + EnsureExtraData(hAllocator); + return m_DedicatedAllocation.m_ExtraData->m_Handle.GetHandle(hAllocator->m_hDevice, m_DedicatedAllocation.m_hMemory, pvkGetMemoryWin32HandleKHR, hTargetProcess, hAllocator->m_UseMutex, pHandle); + default: + VMA_ASSERT(0); + return VK_ERROR_FEATURE_NOT_PRESENT; + } +} +#endif // VMA_EXTERNAL_MEMORY_WIN32 +#endif // VMA_STATS_STRING_ENABLED + +void VmaAllocation_T::EnsureExtraData(VmaAllocator hAllocator) +{ + if (m_DedicatedAllocation.m_ExtraData == VMA_NULL) + { + m_DedicatedAllocation.m_ExtraData = vma_new(hAllocator, VmaAllocationExtraData)(); + } +} + +void VmaAllocation_T::FreeName(VmaAllocator hAllocator) +{ + if(m_pName) + { + VmaFreeString(hAllocator->GetAllocationCallbacks(), m_pName); + m_pName = VMA_NULL; + } +} +#endif // _VMA_ALLOCATION_T_FUNCTIONS + +#ifndef _VMA_BLOCK_VECTOR_FUNCTIONS +VmaBlockVector::VmaBlockVector( + VmaAllocator hAllocator, + VmaPool hParentPool, + uint32_t memoryTypeIndex, + VkDeviceSize preferredBlockSize, + size_t minBlockCount, + size_t maxBlockCount, + VkDeviceSize bufferImageGranularity, + bool explicitBlockSize, + uint32_t algorithm, + float priority, + VkDeviceSize minAllocationAlignment, + void* pMemoryAllocateNext) + : m_hAllocator(hAllocator), + m_hParentPool(hParentPool), + m_MemoryTypeIndex(memoryTypeIndex), + m_PreferredBlockSize(preferredBlockSize), + m_MinBlockCount(minBlockCount), + m_MaxBlockCount(maxBlockCount), + m_BufferImageGranularity(bufferImageGranularity), + m_ExplicitBlockSize(explicitBlockSize), + m_Algorithm(algorithm), + m_Priority(priority), + m_MinAllocationAlignment(minAllocationAlignment), + m_pMemoryAllocateNext(pMemoryAllocateNext), + m_Blocks(VmaStlAllocator(hAllocator->GetAllocationCallbacks())), + m_NextBlockId(0) {} + +VmaBlockVector::~VmaBlockVector() +{ + for (size_t i = m_Blocks.size(); i--; ) + { + m_Blocks[i]->Destroy(m_hAllocator); + vma_delete(m_hAllocator, m_Blocks[i]); + } +} + +VkResult VmaBlockVector::CreateMinBlocks() +{ + for (size_t i = 0; i < m_MinBlockCount; ++i) + { + VkResult res = CreateBlock(m_PreferredBlockSize, VMA_NULL); + if (res != VK_SUCCESS) + { + return res; + } + } + return VK_SUCCESS; +} + +void VmaBlockVector::AddStatistics(VmaStatistics& 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->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); + } +} + +bool VmaBlockVector::IsEmpty() +{ + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + return m_Blocks.empty(); +} + +bool VmaBlockVector::IsCorruptionDetectionEnabled() const +{ + const uint32_t requiredMemFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; + return (VMA_DEBUG_DETECT_CORRUPTION != 0) && + (VMA_DEBUG_MARGIN > 0) && + (m_Algorithm == 0 || m_Algorithm == VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) && + (m_hAllocator->m_MemProps.memoryTypes[m_MemoryTypeIndex].propertyFlags & requiredMemFlags) == requiredMemFlags; +} + +VkResult VmaBlockVector::Allocate( + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + size_t allocIndex; + VkResult res = VK_SUCCESS; + + alignment = VMA_MAX(alignment, m_MinAllocationAlignment); + + if (IsCorruptionDetectionEnabled()) + { + size = VmaAlignUp(size, sizeof(VMA_CORRUPTION_DETECTION_MAGIC_VALUE)); + alignment = VmaAlignUp(alignment, sizeof(VMA_CORRUPTION_DETECTION_MAGIC_VALUE)); + } + + { + VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); + for (allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + res = AllocatePage( + size, + alignment, + createInfo, + suballocType, + pAllocations + allocIndex); + if (res != VK_SUCCESS) + { + break; + } + } + } + + if (res != VK_SUCCESS) + { + // Free all already created allocations. + while (allocIndex--) + Free(pAllocations[allocIndex]); + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + } + + return res; +} + +VkResult VmaBlockVector::AllocatePage( + VkDeviceSize size, + VkDeviceSize alignment, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + VmaAllocation* pAllocation) +{ + const bool isUpperAddress = (createInfo.flags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0; + + VkDeviceSize freeMemory; + { + const uint32_t heapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex); + VmaBudget heapBudget = {}; + m_hAllocator->GetHeapBudgets(&heapBudget, heapIndex, 1); + freeMemory = (heapBudget.usage < heapBudget.budget) ? (heapBudget.budget - heapBudget.usage) : 0; + } + + 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; + + // Upper address can only be used with linear allocator and within single memory block. + if (isUpperAddress && + (m_Algorithm != VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT || m_MaxBlockCount > 1)) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + // Early reject: requested allocation size is larger that maximum block size for this block vector. + if (size + VMA_DEBUG_MARGIN > m_PreferredBlockSize) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + + // 1. Search existing allocations. Try to allocate. + if (m_Algorithm == VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT) + { + // Use only last block. + if (!m_Blocks.empty()) + { + VmaDeviceMemoryBlock* const pCurrBlock = m_Blocks.back(); + VMA_ASSERT(pCurrBlock); + VkResult res = AllocateFromBlock( + pCurrBlock, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) + { + VMA_DEBUG_LOG_FORMAT(" Returned from last block #%" PRIu32, pCurrBlock->GetId()); + IncrementallySortBlocks(); + return VK_SUCCESS; + } + } + } + else + { + if (strategy != VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT) // MIN_MEMORY or default + { + 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_FORMAT(" Returned from existing block #%" PRIu32, 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) + { + 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) + { + VMA_DEBUG_LOG_FORMAT(" Returned from existing block #%" PRIu32, pCurrBlock->GetId()); + IncrementallySortBlocks(); + return VK_SUCCESS; + } + } + } + } + else // VMA_ALLOCATION_CREATE_STRATEGY_MIN_TIME_BIT + { + // 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, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) + { + VMA_DEBUG_LOG_FORMAT(" Returned from existing block #%" PRIu32, pCurrBlock->GetId()); + IncrementallySortBlocks(); + return VK_SUCCESS; + } + } + } + } + + // 2. Try to create new block. + if (canCreateNewBlock) + { + // Calculate optimal size for new block. + VkDeviceSize newBlockSize = m_PreferredBlockSize; + uint32_t newBlockSizeShift = 0; + const uint32_t NEW_BLOCK_SIZE_SHIFT_MAX = 3; + + if (!m_ExplicitBlockSize) + { + // Allocate 1/8, 1/4, 1/2 as first blocks. + const VkDeviceSize maxExistingBlockSize = CalcMaxBlockSize(); + for (uint32_t i = 0; i < NEW_BLOCK_SIZE_SHIFT_MAX; ++i) + { + const VkDeviceSize smallerNewBlockSize = newBlockSize / 2; + if (smallerNewBlockSize > maxExistingBlockSize && smallerNewBlockSize >= size * 2) + { + newBlockSize = smallerNewBlockSize; + ++newBlockSizeShift; + } + else + { + break; + } + } + } + + size_t newBlockIndex = 0; + VkResult res = (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, size, alignment, createInfo.flags, createInfo.pUserData, suballocType, strategy, pAllocation); + if (res == VK_SUCCESS) + { + VMA_DEBUG_LOG_FORMAT(" Created new block #%" PRIu32 " Size=%" PRIu64, pBlock->GetId(), newBlockSize); + 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; + } + } + } + + return VK_ERROR_OUT_OF_DEVICE_MEMORY; +} + +void VmaBlockVector::Free(const VmaAllocation hAllocation) +{ + VmaDeviceMemoryBlock* pBlockToDelete = VMA_NULL; + + bool budgetExceeded = false; + { + const uint32_t heapIndex = m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex); + VmaBudget heapBudget = {}; + m_hAllocator->GetHeapBudgets(&heapBudget, heapIndex, 1); + budgetExceeded = heapBudget.usage >= heapBudget.budget; + } + + // Scope for lock. + { + VmaMutexLockWrite lock(m_Mutex, m_hAllocator->m_UseMutex); + + VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); + + if (IsCorruptionDetectionEnabled()) + { + 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()) + { + pBlock->Unmap(m_hAllocator, 1); + } + + const bool hadEmptyBlockBeforeFree = HasEmptyBlock(); + pBlock->m_pMetadata->Free(hAllocation->GetAllocHandle()); + pBlock->PostFree(m_hAllocator); + VMA_HEAVY_ASSERT(pBlock->Validate()); + + VMA_DEBUG_LOG_FORMAT(" Freed from MemoryTypeIndex=%" PRIu32, m_MemoryTypeIndex); + + const bool canDeleteBlock = m_Blocks.size() > m_MinBlockCount; + // pBlock became empty after this deallocation. + if (pBlock->m_pMetadata->IsEmpty()) + { + // 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 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 (hadEmptyBlockBeforeFree && canDeleteBlock) + { + VmaDeviceMemoryBlock* pLastBlock = m_Blocks.back(); + if (pLastBlock->m_pMetadata->IsEmpty()) + { + pBlockToDelete = pLastBlock; + m_Blocks.pop_back(); + } + } + + IncrementallySortBlocks(); + + m_hAllocator->m_Budget.RemoveAllocation(m_hAllocator->MemoryTypeIndexToHeapIndex(m_MemoryTypeIndex), hAllocation->GetSize()); + hAllocation->Destroy(m_hAllocator); + m_hAllocator->m_AllocationObjectAllocator.Free(hAllocation); + } + + // Destruction of a free block. Deferred until this point, outside of mutex + // lock, for performance reason. + if (pBlockToDelete != VMA_NULL) + { + VMA_DEBUG_LOG_FORMAT(" Deleted empty block #%" PRIu32, pBlockToDelete->GetId()); + pBlockToDelete->Destroy(m_hAllocator); + vma_delete(m_hAllocator, pBlockToDelete); + } +} + +VkDeviceSize VmaBlockVector::CalcMaxBlockSize() const +{ + VkDeviceSize result = 0; + for (size_t i = m_Blocks.size(); i--; ) + { + result = VMA_MAX(result, m_Blocks[i]->m_pMetadata->GetSize()); + if (result >= m_PreferredBlockSize) + { + break; + } + } + return result; +} + +void VmaBlockVector::Remove(VmaDeviceMemoryBlock* pBlock) +{ + for (uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + if (m_Blocks[blockIndex] == pBlock) + { + VmaVectorRemove(m_Blocks, blockIndex); + return; + } + } + VMA_ASSERT(0); +} + +void VmaBlockVector::IncrementallySortBlocks() +{ + if (!m_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) + { + if (m_Blocks[i - 1]->m_pMetadata->GetSumFreeSize() > m_Blocks[i]->m_pMetadata->GetSumFreeSize()) + { + std::swap(m_Blocks[i - 1], m_Blocks[i]); + return; + } + } + } +} + +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, + VkDeviceSize size, + VkDeviceSize alignment, + VmaAllocationCreateFlags allocFlags, + void* pUserData, + VmaSuballocationType suballocType, + uint32_t strategy, + VmaAllocation* pAllocation) +{ + const bool isUpperAddress = (allocFlags & VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT) != 0; + + VmaAllocationRequest currRequest = {}; + if (pBlock->m_pMetadata->CreateAllocationRequest( + size, + alignment, + isUpperAddress, + suballocType, + strategy, + &currRequest)) + { + 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(m_hAllocator); + // 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 }; + allocInfo.pNext = m_pMemoryAllocateNext; + allocInfo.memoryTypeIndex = m_MemoryTypeIndex; + allocInfo.allocationSize = blockSize; + +#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) + { + allocFlagsInfo.flags = VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT_KHR; + VmaPnextChainPushFront(&allocInfo, &allocFlagsInfo); + } +#endif // VMA_BUFFER_DEVICE_ADDRESS + +#if VMA_MEMORY_PRIORITY + VkMemoryPriorityAllocateInfoEXT priorityInfo = { VK_STRUCTURE_TYPE_MEMORY_PRIORITY_ALLOCATE_INFO_EXT }; + if (m_hAllocator->m_UseExtMemoryPriority) + { + VMA_ASSERT(m_Priority >= 0.f && m_Priority <= 1.f); + priorityInfo.priority = m_Priority; + VmaPnextChainPushFront(&allocInfo, &priorityInfo); + } +#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) + { + VmaPnextChainPushFront(&allocInfo, &exportMemoryAllocInfo); + } +#endif // VMA_EXTERNAL_MEMORY + + VkDeviceMemory mem = VK_NULL_HANDLE; + VkResult res = m_hAllocator->AllocateVulkanMemory(&allocInfo, &mem); + if (res < 0) + { + return res; + } + + // New VkDeviceMemory successfully created. + + // Create new Allocation for it. + VmaDeviceMemoryBlock* const pBlock = vma_new(m_hAllocator, VmaDeviceMemoryBlock)(m_hAllocator); + pBlock->Init( + m_hAllocator, + m_hParentPool, + m_MemoryTypeIndex, + mem, + allocInfo.allocationSize, + m_NextBlockId++, + m_Algorithm, + m_BufferImageGranularity); + + m_Blocks.push_back(pBlock); + if (pNewBlockIndex != VMA_NULL) + { + *pNewBlockIndex = m_Blocks.size() - 1; + } + + return VK_SUCCESS; +} + +bool VmaBlockVector::HasEmptyBlock() +{ + for (size_t index = 0, count = m_Blocks.size(); index < count; ++index) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[index]; + if (pBlock->m_pMetadata->IsEmpty()) + { + return true; + } + } + return false; +} + +#if VMA_STATS_STRING_ENABLED +void VmaBlockVector::PrintDetailedMap(class VmaJsonWriter& json) +{ + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + + + json.BeginObject(); + for (size_t i = 0; i < m_Blocks.size(); ++i) + { + 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()) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + + VmaMutexLockRead lock(m_Mutex, m_hAllocator->m_UseMutex); + for (uint32_t blockIndex = 0; blockIndex < m_Blocks.size(); ++blockIndex) + { + VmaDeviceMemoryBlock* const pBlock = m_Blocks[blockIndex]; + VMA_ASSERT(pBlock); + VkResult res = pBlock->CheckCorruption(m_hAllocator); + if (res != VK_SUCCESS) + { + return res; + } + } + return VK_SUCCESS; +} + +#endif // _VMA_BLOCK_VECTOR_FUNCTIONS + +#ifndef _VMA_DEFRAGMENTATION_CONTEXT_FUNCTIONS +VmaDefragmentationContext_T::VmaDefragmentationContext_T( + VmaAllocator hAllocator, + const VmaDefragmentationInfo& info) + : m_MaxPassBytes(info.maxBytesPerPass == 0 ? VK_WHOLE_SIZE : info.maxBytesPerPass), + m_MaxPassAllocations(info.maxAllocationsPerPass == 0 ? UINT32_MAX : info.maxAllocationsPerPass), + m_BreakCallback(info.pfnBreakCallback), + m_BreakCallbackUserData(info.pBreakCallbackUserData), + m_MoveAllocator(hAllocator->GetAllocationCallbacks()), + m_Moves(m_MoveAllocator) +{ + m_Algorithm = info.flags & VMA_DEFRAGMENTATION_FLAG_ALGORITHM_MASK; + + if (info.pool != VMA_NULL) + { + m_BlockVectorCount = 1; + m_PoolBlockVector = &info.pool->m_BlockVector; + m_pBlockVectors = &m_PoolBlockVector; + m_PoolBlockVector->SetIncrementalSort(false); + m_PoolBlockVector->SortByFreeSize(); + } + else + { + m_BlockVectorCount = hAllocator->GetMemoryTypeCount(); + m_PoolBlockVector = VMA_NULL; + m_pBlockVectors = hAllocator->m_pBlockVectors; + for (uint32_t i = 0; i < m_BlockVectorCount; ++i) + { + 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; + m_AlgorithmState = vma_new_array(hAllocator, StateBalanced, m_BlockVectorCount); + break; + 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); + } + + if (prevCount > currentCount) + { + size_t freedBlocks = prevCount - currentCount; + m_PassStats.deviceMemoryBlocksFreed += static_cast(freedBlocks); + m_PassStats.bytesFreed += freedBlockSize; + } + + if(m_Algorithm == VMA_DEFRAGMENTATION_FLAG_ALGORITHM_EXTENSIVE_BIT && + m_AlgorithmState != VMA_NULL) + { + // Avoid unnecessary tries to allocate when new free block is available + StateExtensive& state = reinterpret_cast(m_AlgorithmState)[vectorIndex]; + if (state.firstFreeBlock != SIZE_MAX) + { + 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(); + + // 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 }; + + // Move blocks with immovable allocations according to algorithm + if (immovableBlocks.size() > 0) + { + do + { + if(m_Algorithm == 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) + { + 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) + { + std::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) + std::swap(vector->m_Blocks[i], vector->m_Blocks[--state.firstFreeBlock]); + else + --state.firstFreeBlock; + } + } + swapped = true; + break; + } + } + } + } + if (swapped) + result = VK_INCOMPLETE; + break; + } + } + + // Move to the beginning + for (const FragmentedBlock& block : immovableBlocks) + { + VmaBlockVector* vector = m_pBlockVectors[block.data]; + VmaMutexLockWrite lock(vector->GetMutex(), vector->GetAllocator()->m_UseMutex); + + for (size_t i = m_ImmovableBlockCount; i < vector->GetBlockCount(); ++i) + { + if (vector->GetBlock(i) == block.block) + { + std::swap(vector->m_Blocks[i], vector->m_Blocks[m_ImmovableBlockCount++]); + break; + } + } + } + } while (false); + } + + // Bulk-map destination blocks + for (const FragmentedBlock& block : mappedBlocks) + { + VkResult res = block.block->Map(allocator, block.data, VMA_NULL); + VMA_ASSERT(res == VK_SUCCESS); + } + return result; +} + +bool VmaDefragmentationContext_T::ComputeDefragmentation(VmaBlockVector& vector, size_t index) +{ + switch (m_Algorithm) + { + case VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT: + return ComputeDefragmentation_Fast(vector); + 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); + default: + VMA_ASSERT(0); + return ComputeDefragmentation_Balanced(vector, index, true); + } +} + +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) +{ + // Check custom criteria if exists + if (m_BreakCallback && m_BreakCallback(m_BreakCallbackUserData)) + return CounterStatus::End; + + // Ignore allocation if will exceed max size for copy + if (m_PassStats.bytesMoved + bytes > m_MaxPassBytes) + { + if (++m_IgnoredAllocs < MAX_ALLOCS_TO_IGNORE) + return CounterStatus::Ignore; + else + return CounterStatus::End; + } + else + m_IgnoredAllocs = 0; + 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; +} + +bool VmaDefragmentationContext_T::ReallocWithinBlock(VmaBlockVector& vector, VmaDeviceMemoryBlock* block) +{ + 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; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + 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)) + { + 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; + } + } + } + return false; +} + +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())) + { + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + // 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 noticeable 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())) + { + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + // 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) + { + 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) + { + 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; + } + } + } + } + } + prevFreeRegionSize = nextFreeRegionSize; + } + } + + // No moves performed, 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; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + // 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) + { + 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::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())) + { + case CounterStatus::Ignore: + continue; + case CounterStatus::End: + return true; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + // 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) + vectorState.firstFreeBlock = 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) + { + if (ReallocWithinBlock(vector, vector.GetBlock(i))) + return true; + } + } + + if (prevMoveCount == m_Moves.size()) + { + // No possible reallocs within blocks, try to move them around fast + return ComputeDefragmentation_Fast(vector); + } + } + else + { + switch (vectorState.operation) + { + case StateExtensive::Operation::FindFreeBlockBuffer: + vectorState.operation = StateExtensive::Operation::MoveBuffers; + break; + case StateExtensive::Operation::FindFreeBlockTexture: + vectorState.operation = StateExtensive::Operation::MoveTextures; + break; + case StateExtensive::Operation::FindFreeBlockAll: + vectorState.operation = StateExtensive::Operation::MoveAll; + break; + default: + VMA_ASSERT(0); + vectorState.operation = StateExtensive::Operation::MoveTextures; + } + 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; + VMA_FALLTHROUGH; // Fallthrough + } + 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; + VMA_FALLTHROUGH; // Fallthrough + } + 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; + case CounterStatus::Pass: + break; + default: + VMA_ASSERT(0); + } + + // 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); + + const VkAllocationCallbacks* allocs = m_BlockVector.GetAllocator()->GetAllocationCallbacks(); + VmaFreeString(allocs, m_Name); +} + +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_Name = VMA_NULL; + } +} +#endif // _VMA_POOL_T_FUNCTIONS + +#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), + m_UseKhrDedicatedAllocation((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT) != 0), + m_UseKhrBindMemory2((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT) != 0), + m_UseExtMemoryBudget((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT) != 0), + m_UseAmdDeviceCoherentMemory((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_AMD_DEVICE_COHERENT_MEMORY_BIT) != 0), + m_UseKhrBufferDeviceAddress((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT) != 0), + m_UseExtMemoryPriority((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT) != 0), + m_UseKhrMaintenance4((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE4_BIT) != 0), + m_UseKhrMaintenance5((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT) != 0), + m_UseKhrExternalMemoryWin32((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_EXTERNAL_MEMORY_WIN32_BIT) != 0), + m_hDevice(pCreateInfo->device), + m_hInstance(pCreateInfo->instance), + m_AllocationCallbacksSpecified(pCreateInfo->pAllocationCallbacks != VMA_NULL), + m_AllocationCallbacks(pCreateInfo->pAllocationCallbacks ? + *pCreateInfo->pAllocationCallbacks : VmaEmptyAllocationCallbacks), + m_AllocationObjectAllocator(&m_AllocationCallbacks), + m_HeapSizeLimitMask(0), + m_DeviceMemoryCount(0), + m_PreferredLargeHeapBlockSize(0), + m_PhysicalDevice(pCreateInfo->physicalDevice), + m_GpuDefragmentationMemoryTypeBits(UINT32_MAX), + m_NextPoolId(0), + m_GlobalMemoryTypeBits(UINT32_MAX) +{ + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + m_UseKhrDedicatedAllocation = false; + m_UseKhrBindMemory2 = false; + } + + if(VMA_DEBUG_DETECT_CORRUPTION) + { + // Needs to be multiply of uint32_t size because we are going to write VMA_CORRUPTION_DETECTION_MAGIC_VALUE to it. + VMA_ASSERT(VMA_DEBUG_MARGIN % sizeof(uint32_t) == 0); + } + + VMA_ASSERT(pCreateInfo->physicalDevice && pCreateInfo->device && pCreateInfo->instance); + + if(m_VulkanApiVersion < VK_MAKE_VERSION(1, 1, 0)) + { +#if !(VMA_DEDICATED_ALLOCATION) + if((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT) != 0) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT set but required extensions are disabled by preprocessor macros."); + } +#endif +#if !(VMA_BIND_MEMORY2) + if((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT) != 0) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT set but required extension is disabled by preprocessor macros."); + } +#endif + } +#if !(VMA_MEMORY_BUDGET) + if((pCreateInfo->flags & VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT) != 0) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT set but required extension is disabled by preprocessor macros."); + } +#endif +#if !(VMA_BUFFER_DEVICE_ADDRESS) + if(m_UseKhrBufferDeviceAddress) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT is set but required extension or Vulkan 1.2 is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif +#if VMA_VULKAN_VERSION < 1004000 + VMA_ASSERT(m_VulkanApiVersion < VK_MAKE_VERSION(1, 4, 0) && "vulkanApiVersion >= VK_API_VERSION_1_4 but required Vulkan version is disabled by preprocessor macros."); +#endif +#if VMA_VULKAN_VERSION < 1003000 + VMA_ASSERT(m_VulkanApiVersion < VK_MAKE_VERSION(1, 3, 0) && "vulkanApiVersion >= VK_API_VERSION_1_3 but required Vulkan version is disabled by preprocessor macros."); +#endif +#if VMA_VULKAN_VERSION < 1002000 + VMA_ASSERT(m_VulkanApiVersion < VK_MAKE_VERSION(1, 2, 0) && "vulkanApiVersion >= VK_API_VERSION_1_2 but required Vulkan version is disabled by preprocessor macros."); +#endif +#if VMA_VULKAN_VERSION < 1001000 + VMA_ASSERT(m_VulkanApiVersion < VK_MAKE_VERSION(1, 1, 0) && "vulkanApiVersion >= VK_API_VERSION_1_1 but required Vulkan version is disabled by preprocessor macros."); +#endif +#if !(VMA_MEMORY_PRIORITY) + if(m_UseExtMemoryPriority) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT is set but required extension is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif +#if !(VMA_KHR_MAINTENANCE4) + if(m_UseKhrMaintenance4) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE4_BIT is set but required extension is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif +#if !(VMA_KHR_MAINTENANCE5) + if(m_UseKhrMaintenance5) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT is set but required extension is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif +#if !(VMA_KHR_MAINTENANCE5) + if(m_UseKhrMaintenance5) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT is set but required extension is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif + +#if !(VMA_EXTERNAL_MEMORY_WIN32) + if(m_UseKhrExternalMemoryWin32) + { + VMA_ASSERT(0 && "VMA_ALLOCATOR_CREATE_KHR_EXTERNAL_MEMORY_WIN32_BIT is set but required extension is not available in your Vulkan header or its support in VMA has been disabled by a preprocessor macro."); + } +#endif + + memset(&m_DeviceMemoryCallbacks, 0 ,sizeof(m_DeviceMemoryCallbacks)); + memset(&m_PhysicalDeviceProperties, 0, sizeof(m_PhysicalDeviceProperties)); + memset(&m_MemProps, 0, sizeof(m_MemProps)); + + memset(&m_pBlockVectors, 0, sizeof(m_pBlockVectors)); + memset(&m_VulkanFunctions, 0, sizeof(m_VulkanFunctions)); + +#if VMA_EXTERNAL_MEMORY + memset(&m_TypeExternalMemoryHandleTypes, 0, sizeof(m_TypeExternalMemoryHandleTypes)); +#endif // #if VMA_EXTERNAL_MEMORY + + if(pCreateInfo->pDeviceMemoryCallbacks != VMA_NULL) + { + m_DeviceMemoryCallbacks.pUserData = pCreateInfo->pDeviceMemoryCallbacks->pUserData; + m_DeviceMemoryCallbacks.pfnAllocate = pCreateInfo->pDeviceMemoryCallbacks->pfnAllocate; + m_DeviceMemoryCallbacks.pfnFree = pCreateInfo->pDeviceMemoryCallbacks->pfnFree; + } + + ImportVulkanFunctions(pCreateInfo->pVulkanFunctions); + + (*m_VulkanFunctions.vkGetPhysicalDeviceProperties)(m_PhysicalDevice, &m_PhysicalDeviceProperties); + (*m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties)(m_PhysicalDevice, &m_MemProps); + + VMA_ASSERT(VmaIsPow2(VMA_MIN_ALIGNMENT)); + VMA_ASSERT(VmaIsPow2(VMA_DEBUG_MIN_BUFFER_IMAGE_GRANULARITY)); + VMA_ASSERT(VmaIsPow2(m_PhysicalDeviceProperties.limits.bufferImageGranularity)); + VMA_ASSERT(VmaIsPow2(m_PhysicalDeviceProperties.limits.nonCoherentAtomSize)); + + m_PreferredLargeHeapBlockSize = (pCreateInfo->preferredLargeHeapBlockSize != 0) ? + pCreateInfo->preferredLargeHeapBlockSize : static_cast(VMA_DEFAULT_LARGE_HEAP_BLOCK_SIZE); + + m_GlobalMemoryTypeBits = CalculateGlobalMemoryTypeBits(); + +#if VMA_EXTERNAL_MEMORY + if(pCreateInfo->pTypeExternalMemoryHandleTypes != VMA_NULL) + { + memcpy(m_TypeExternalMemoryHandleTypes, pCreateInfo->pTypeExternalMemoryHandleTypes, + sizeof(VkExternalMemoryHandleTypeFlagsKHR) * GetMemoryTypeCount()); + } +#endif // #if VMA_EXTERNAL_MEMORY + + if(pCreateInfo->pHeapSizeLimit != VMA_NULL) + { + for(uint32_t heapIndex = 0; heapIndex < GetMemoryHeapCount(); ++heapIndex) + { + const VkDeviceSize limit = pCreateInfo->pHeapSizeLimit[heapIndex]; + if(limit != VK_WHOLE_SIZE) + { + m_HeapSizeLimitMask |= 1u << heapIndex; + if(limit < m_MemProps.memoryHeaps[heapIndex].size) + { + m_MemProps.memoryHeaps[heapIndex].size = limit; + } + } + } + } + + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + // 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, + // because minBlockCount is 0. + } + } +} + +VkResult VmaAllocator_T::Init(const VmaAllocatorCreateInfo* pCreateInfo) +{ + VkResult res = VK_SUCCESS; + +#if VMA_MEMORY_BUDGET + if(m_UseExtMemoryBudget) + { + UpdateVulkanBudget(); + } +#endif // #if VMA_MEMORY_BUDGET + + return res; +} + +VmaAllocator_T::~VmaAllocator_T() +{ + VMA_ASSERT(m_Pools.IsEmpty()); + + for(size_t memTypeIndex = GetMemoryTypeCount(); memTypeIndex--; ) + { + vma_delete(this, m_pBlockVectors[memTypeIndex]); + } +} + +void VmaAllocator_T::ImportVulkanFunctions(const VmaVulkanFunctions* pVulkanFunctions) +{ +#if VMA_STATIC_VULKAN_FUNCTIONS == 1 + ImportVulkanFunctions_Static(); +#endif + + if(pVulkanFunctions != VMA_NULL) + { + ImportVulkanFunctions_Custom(pVulkanFunctions); + } + +#if VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 + ImportVulkanFunctions_Dynamic(); +#endif + + ValidateVulkanFunctions(); +} + +#if VMA_STATIC_VULKAN_FUNCTIONS == 1 + +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; + m_VulkanFunctions.vkFreeMemory = (PFN_vkFreeMemory)vkFreeMemory; + m_VulkanFunctions.vkMapMemory = (PFN_vkMapMemory)vkMapMemory; + m_VulkanFunctions.vkUnmapMemory = (PFN_vkUnmapMemory)vkUnmapMemory; + m_VulkanFunctions.vkFlushMappedMemoryRanges = (PFN_vkFlushMappedMemoryRanges)vkFlushMappedMemoryRanges; + m_VulkanFunctions.vkInvalidateMappedMemoryRanges = (PFN_vkInvalidateMappedMemoryRanges)vkInvalidateMappedMemoryRanges; + m_VulkanFunctions.vkBindBufferMemory = (PFN_vkBindBufferMemory)vkBindBufferMemory; + m_VulkanFunctions.vkBindImageMemory = (PFN_vkBindImageMemory)vkBindImageMemory; + m_VulkanFunctions.vkGetBufferMemoryRequirements = (PFN_vkGetBufferMemoryRequirements)vkGetBufferMemoryRequirements; + m_VulkanFunctions.vkGetImageMemoryRequirements = (PFN_vkGetImageMemoryRequirements)vkGetImageMemoryRequirements; + m_VulkanFunctions.vkCreateBuffer = (PFN_vkCreateBuffer)vkCreateBuffer; + m_VulkanFunctions.vkDestroyBuffer = (PFN_vkDestroyBuffer)vkDestroyBuffer; + m_VulkanFunctions.vkCreateImage = (PFN_vkCreateImage)vkCreateImage; + m_VulkanFunctions.vkDestroyImage = (PFN_vkDestroyImage)vkDestroyImage; + m_VulkanFunctions.vkCmdCopyBuffer = (PFN_vkCmdCopyBuffer)vkCmdCopyBuffer; + + // Vulkan 1.1 +#if VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR = (PFN_vkGetBufferMemoryRequirements2)vkGetBufferMemoryRequirements2; + m_VulkanFunctions.vkGetImageMemoryRequirements2KHR = (PFN_vkGetImageMemoryRequirements2)vkGetImageMemoryRequirements2; + m_VulkanFunctions.vkBindBufferMemory2KHR = (PFN_vkBindBufferMemory2)vkBindBufferMemory2; + m_VulkanFunctions.vkBindImageMemory2KHR = (PFN_vkBindImageMemory2)vkBindImageMemory2; + } +#endif + +#if VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + 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 // VMA_STATIC_VULKAN_FUNCTIONS == 1 + +void VmaAllocator_T::ImportVulkanFunctions_Custom(const VmaVulkanFunctions* pVulkanFunctions) +{ + VMA_ASSERT(pVulkanFunctions != VMA_NULL); + +#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); + VMA_COPY_IF_NOT_NULL(vkFreeMemory); + VMA_COPY_IF_NOT_NULL(vkMapMemory); + VMA_COPY_IF_NOT_NULL(vkUnmapMemory); + VMA_COPY_IF_NOT_NULL(vkFlushMappedMemoryRanges); + VMA_COPY_IF_NOT_NULL(vkInvalidateMappedMemoryRanges); + VMA_COPY_IF_NOT_NULL(vkBindBufferMemory); + VMA_COPY_IF_NOT_NULL(vkBindImageMemory); + VMA_COPY_IF_NOT_NULL(vkGetBufferMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkGetImageMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkCreateBuffer); + VMA_COPY_IF_NOT_NULL(vkDestroyBuffer); + VMA_COPY_IF_NOT_NULL(vkCreateImage); + VMA_COPY_IF_NOT_NULL(vkDestroyImage); + VMA_COPY_IF_NOT_NULL(vkCmdCopyBuffer); + +#if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + VMA_COPY_IF_NOT_NULL(vkGetBufferMemoryRequirements2KHR); + VMA_COPY_IF_NOT_NULL(vkGetImageMemoryRequirements2KHR); +#endif + +#if VMA_BIND_MEMORY2 || VMA_VULKAN_VERSION >= 1001000 + VMA_COPY_IF_NOT_NULL(vkBindBufferMemory2KHR); + VMA_COPY_IF_NOT_NULL(vkBindImageMemory2KHR); +#endif + +#if VMA_MEMORY_BUDGET || VMA_VULKAN_VERSION >= 1001000 + VMA_COPY_IF_NOT_NULL(vkGetPhysicalDeviceMemoryProperties2KHR); +#endif + +#if VMA_KHR_MAINTENANCE4 || VMA_VULKAN_VERSION >= 1003000 + VMA_COPY_IF_NOT_NULL(vkGetDeviceBufferMemoryRequirements); + VMA_COPY_IF_NOT_NULL(vkGetDeviceImageMemoryRequirements); +#endif +#if VMA_EXTERNAL_MEMORY_WIN32 + VMA_COPY_IF_NOT_NULL(vkGetMemoryWin32HandleKHR); +#endif +#undef VMA_COPY_IF_NOT_NULL +} + +#if VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 + +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)m_VulkanFunctions.vkGetInstanceProcAddr(m_hInstance, functionNameString); +#define VMA_FETCH_DEVICE_FUNC(memberName, functionPointerType, functionNameString) \ + if(m_VulkanFunctions.memberName == VMA_NULL) \ + m_VulkanFunctions.memberName = \ + (functionPointerType)m_VulkanFunctions.vkGetDeviceProcAddr(m_hDevice, functionNameString); + + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceProperties, PFN_vkGetPhysicalDeviceProperties, "vkGetPhysicalDeviceProperties"); + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties, PFN_vkGetPhysicalDeviceMemoryProperties, "vkGetPhysicalDeviceMemoryProperties"); + VMA_FETCH_DEVICE_FUNC(vkAllocateMemory, PFN_vkAllocateMemory, "vkAllocateMemory"); + VMA_FETCH_DEVICE_FUNC(vkFreeMemory, PFN_vkFreeMemory, "vkFreeMemory"); + VMA_FETCH_DEVICE_FUNC(vkMapMemory, PFN_vkMapMemory, "vkMapMemory"); + VMA_FETCH_DEVICE_FUNC(vkUnmapMemory, PFN_vkUnmapMemory, "vkUnmapMemory"); + VMA_FETCH_DEVICE_FUNC(vkFlushMappedMemoryRanges, PFN_vkFlushMappedMemoryRanges, "vkFlushMappedMemoryRanges"); + VMA_FETCH_DEVICE_FUNC(vkInvalidateMappedMemoryRanges, PFN_vkInvalidateMappedMemoryRanges, "vkInvalidateMappedMemoryRanges"); + VMA_FETCH_DEVICE_FUNC(vkBindBufferMemory, PFN_vkBindBufferMemory, "vkBindBufferMemory"); + VMA_FETCH_DEVICE_FUNC(vkBindImageMemory, PFN_vkBindImageMemory, "vkBindImageMemory"); + VMA_FETCH_DEVICE_FUNC(vkGetBufferMemoryRequirements, PFN_vkGetBufferMemoryRequirements, "vkGetBufferMemoryRequirements"); + VMA_FETCH_DEVICE_FUNC(vkGetImageMemoryRequirements, PFN_vkGetImageMemoryRequirements, "vkGetImageMemoryRequirements"); + VMA_FETCH_DEVICE_FUNC(vkCreateBuffer, PFN_vkCreateBuffer, "vkCreateBuffer"); + VMA_FETCH_DEVICE_FUNC(vkDestroyBuffer, PFN_vkDestroyBuffer, "vkDestroyBuffer"); + VMA_FETCH_DEVICE_FUNC(vkCreateImage, PFN_vkCreateImage, "vkCreateImage"); + VMA_FETCH_DEVICE_FUNC(vkDestroyImage, PFN_vkDestroyImage, "vkDestroyImage"); + VMA_FETCH_DEVICE_FUNC(vkCmdCopyBuffer, PFN_vkCmdCopyBuffer, "vkCmdCopyBuffer"); + +#if VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VMA_FETCH_DEVICE_FUNC(vkGetBufferMemoryRequirements2KHR, PFN_vkGetBufferMemoryRequirements2, "vkGetBufferMemoryRequirements2"); + VMA_FETCH_DEVICE_FUNC(vkGetImageMemoryRequirements2KHR, PFN_vkGetImageMemoryRequirements2, "vkGetImageMemoryRequirements2"); + VMA_FETCH_DEVICE_FUNC(vkBindBufferMemory2KHR, PFN_vkBindBufferMemory2, "vkBindBufferMemory2"); + VMA_FETCH_DEVICE_FUNC(vkBindImageMemory2KHR, PFN_vkBindImageMemory2, "vkBindImageMemory2"); + } +#endif + +#if VMA_MEMORY_BUDGET || VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2"); + // Try to fetch the pointer from the other name, based on suspected driver bug - see issue #410. + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2KHR"); + } + else if(m_UseExtMemoryBudget) + { + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2KHR"); + // Try to fetch the pointer from the other name, based on suspected driver bug - see issue #410. + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2"); + } +#endif + +#if VMA_DEDICATED_ALLOCATION + if(m_UseKhrDedicatedAllocation) + { + VMA_FETCH_DEVICE_FUNC(vkGetBufferMemoryRequirements2KHR, PFN_vkGetBufferMemoryRequirements2KHR, "vkGetBufferMemoryRequirements2KHR"); + VMA_FETCH_DEVICE_FUNC(vkGetImageMemoryRequirements2KHR, PFN_vkGetImageMemoryRequirements2KHR, "vkGetImageMemoryRequirements2KHR"); + } +#endif + +#if VMA_BIND_MEMORY2 + if(m_UseKhrBindMemory2) + { + VMA_FETCH_DEVICE_FUNC(vkBindBufferMemory2KHR, PFN_vkBindBufferMemory2KHR, "vkBindBufferMemory2KHR"); + VMA_FETCH_DEVICE_FUNC(vkBindImageMemory2KHR, PFN_vkBindImageMemory2KHR, "vkBindImageMemory2KHR"); + } +#endif // #if VMA_BIND_MEMORY2 + +#if VMA_MEMORY_BUDGET || VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2"); + } + else if(m_UseExtMemoryBudget) + { + VMA_FETCH_INSTANCE_FUNC(vkGetPhysicalDeviceMemoryProperties2KHR, PFN_vkGetPhysicalDeviceMemoryProperties2KHR, "vkGetPhysicalDeviceMemoryProperties2KHR"); + } +#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 +#if VMA_KHR_MAINTENANCE4 + if(m_UseKhrMaintenance4) + { + VMA_FETCH_DEVICE_FUNC(vkGetDeviceBufferMemoryRequirements, PFN_vkGetDeviceBufferMemoryRequirementsKHR, "vkGetDeviceBufferMemoryRequirementsKHR"); + VMA_FETCH_DEVICE_FUNC(vkGetDeviceImageMemoryRequirements, PFN_vkGetDeviceImageMemoryRequirementsKHR, "vkGetDeviceImageMemoryRequirementsKHR"); + } +#endif +#if VMA_EXTERNAL_MEMORY_WIN32 + if (m_UseKhrExternalMemoryWin32) + { + VMA_FETCH_DEVICE_FUNC(vkGetMemoryWin32HandleKHR, PFN_vkGetMemoryWin32HandleKHR, "vkGetMemoryWin32HandleKHR"); + } +#endif +#undef VMA_FETCH_DEVICE_FUNC +#undef VMA_FETCH_INSTANCE_FUNC +} + +#endif // VMA_DYNAMIC_VULKAN_FUNCTIONS == 1 + +void VmaAllocator_T::ValidateVulkanFunctions() +{ + VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceProperties != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkAllocateMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkFreeMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkMapMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkUnmapMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkFlushMappedMemoryRanges != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkInvalidateMappedMemoryRanges != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkBindBufferMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkBindImageMemory != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetBufferMemoryRequirements != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetImageMemoryRequirements != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCreateBuffer != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkDestroyBuffer != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCreateImage != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkDestroyImage != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkCmdCopyBuffer != VMA_NULL); + +#if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0) || m_UseKhrDedicatedAllocation) + { + VMA_ASSERT(m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkGetImageMemoryRequirements2KHR != VMA_NULL); + } +#endif + +#if VMA_BIND_MEMORY2 || VMA_VULKAN_VERSION >= 1001000 + if(m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0) || m_UseKhrBindMemory2) + { + VMA_ASSERT(m_VulkanFunctions.vkBindBufferMemory2KHR != VMA_NULL); + VMA_ASSERT(m_VulkanFunctions.vkBindImageMemory2KHR != VMA_NULL); + } +#endif + +#if VMA_MEMORY_BUDGET || VMA_VULKAN_VERSION >= 1001000 + if(m_UseExtMemoryBudget || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VMA_ASSERT(m_VulkanFunctions.vkGetPhysicalDeviceMemoryProperties2KHR != VMA_NULL); + } +#endif +#if VMA_EXTERNAL_MEMORY_WIN32 + if (m_UseKhrExternalMemoryWin32) + { + VMA_ASSERT(m_VulkanFunctions.vkGetMemoryWin32HandleKHR != VMA_NULL); + } +#endif + + // Not validating these due to suspected driver bugs with these function + // pointers being null despite correct extension or Vulkan version is enabled. + // See issue #397. Their usage in VMA is optional anyway. + // + // VMA_ASSERT(m_VulkanFunctions.vkGetDeviceBufferMemoryRequirements != VMA_NULL); + // VMA_ASSERT(m_VulkanFunctions.vkGetDeviceImageMemoryRequirements != VMA_NULL); +} + +VkDeviceSize VmaAllocator_T::CalcPreferredBlockSize(uint32_t memTypeIndex) +{ + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memTypeIndex); + const VkDeviceSize heapSize = m_MemProps.memoryHeaps[heapIndex].size; + const bool isSmallHeap = heapSize <= VMA_SMALL_HEAP_MAX_SIZE; + return VmaAlignUp(isSmallHeap ? (heapSize / 8) : m_PreferredLargeHeapBlockSize, (VkDeviceSize)32); +} + +VkResult VmaAllocator_T::AllocateMemoryOfType( + VmaPool pool, + VkDeviceSize size, + VkDeviceSize alignment, + bool dedicatedPreferred, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + VmaBufferImageUsage dedicatedBufferImageUsage, + const VmaAllocationCreateInfo& createInfo, + uint32_t memTypeIndex, + VmaSuballocationType suballocType, + VmaDedicatedAllocationList& dedicatedAllocations, + VmaBlockVector& blockVector, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + VMA_ASSERT(pAllocations != VMA_NULL); + VMA_DEBUG_LOG_FORMAT(" AllocateMemory: MemoryTypeIndex=%" PRIu32 ", AllocationCount=%zu, Size=%" PRIu64, memTypeIndex, allocationCount, size); + + VmaAllocationCreateInfo finalCreateInfo = createInfo; + VkResult res = CalcMemTypeParams( + finalCreateInfo, + memTypeIndex, + size, + allocationCount); + if(res != VK_SUCCESS) + return res; + + if((finalCreateInfo.flags & VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT) != 0) + { + 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 + { + 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_PhysicalDeviceProperties.limits.maxMemoryAllocationCount < UINT32_MAX / 4 && + 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 filled pMemory, nothing more to do here. + VMA_DEBUG_LOG(" Allocated as DedicatedMemory"); + return VK_SUCCESS; + } + } + } + + res = blockVector.Allocate( + size, + alignment, + finalCreateInfo, + suballocType, + allocationCount, + pAllocations); + if(res == VK_SUCCESS) + return VK_SUCCESS; + + // Try dedicated memory. + if(canAllocateDedicated && !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 filled 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 map, + bool isUserDataString, + bool isMappingAllowed, + bool canAliasMemory, + void* pUserData, + float priority, + VkBuffer dedicatedBuffer, + VkImage dedicatedImage, + VmaBufferImageUsage dedicatedBufferImageUsage, + size_t allocationCount, + VmaAllocation* pAllocations, + const void* pNextChain) +{ + VMA_ASSERT(allocationCount > 0 && pAllocations); + + 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(!canAliasMemory) + { + if(m_UseKhrDedicatedAllocation || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + 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 + +#if VMA_BUFFER_DEVICE_ADDRESS + VkMemoryAllocateFlagsInfoKHR allocFlagsInfo = { VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO_KHR }; + if(m_UseKhrBufferDeviceAddress) + { + bool canContainBufferWithDeviceAddress = true; + if(dedicatedBuffer != VK_NULL_HANDLE) + { + canContainBufferWithDeviceAddress = dedicatedBufferImageUsage == VmaBufferImageUsage::UNKNOWN || + dedicatedBufferImageUsage.Contains(VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_EXT); + } + else if(dedicatedImage != VK_NULL_HANDLE) + { + canContainBufferWithDeviceAddress = false; + } + if(canContainBufferWithDeviceAddress) + { + allocFlagsInfo.flags = VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT_KHR; + VmaPnextChainPushFront(&allocInfo, &allocFlagsInfo); + } + } +#endif // #if VMA_BUFFER_DEVICE_ADDRESS + +#if VMA_MEMORY_PRIORITY + 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); + } +#endif // #if VMA_MEMORY_PRIORITY + +#if VMA_EXTERNAL_MEMORY + // Attach VkExportMemoryAllocateInfoKHR if necessary. + VkExportMemoryAllocateInfoKHR exportMemoryAllocInfo = { VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO_KHR }; + exportMemoryAllocInfo.handleTypes = GetExternalMemoryHandleTypeFlags(memTypeIndex); + if(exportMemoryAllocInfo.handleTypes != 0) + { + VmaPnextChainPushFront(&allocInfo, &exportMemoryAllocInfo); + } +#endif // #if VMA_EXTERNAL_MEMORY + + size_t allocIndex; + VkResult res = VK_SUCCESS; + for(allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + res = AllocateDedicatedMemoryPage( + pool, + size, + suballocType, + memTypeIndex, + allocInfo, + map, + isUserDataString, + isMappingAllowed, + pUserData, + pAllocations + allocIndex); + if(res != VK_SUCCESS) + { + break; + } + } + + if(res == VK_SUCCESS) + { + for (allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + dedicatedAllocations.Register(pAllocations[allocIndex]); + } + VMA_DEBUG_LOG_FORMAT(" Allocated DedicatedMemory Count=%zu, MemoryTypeIndex=#%" PRIu32, allocationCount, memTypeIndex); + } + else + { + // Free all already created allocations. + while(allocIndex--) + { + VmaAllocation currAlloc = pAllocations[allocIndex]; + VkDeviceMemory hMemory = currAlloc->GetMemory(); + + /* + There is no need to call this, because Vulkan spec allows to skip vkUnmapMemory + before vkFreeMemory. + + if(currAlloc->GetMappedData() != VMA_NULL) + { + (*m_VulkanFunctions.vkUnmapMemory)(m_hDevice, hMemory); + } + */ + + FreeVulkanMemory(memTypeIndex, currAlloc->GetSize(), hMemory); + m_Budget.RemoveAllocation(MemoryTypeIndexToHeapIndex(memTypeIndex), currAlloc->GetSize()); + m_AllocationObjectAllocator.Free(currAlloc); + } + + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + } + + return res; +} + +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) +{ + VkDeviceMemory hMemory = VK_NULL_HANDLE; + VkResult res = AllocateVulkanMemory(&allocInfo, &hMemory); + if(res < 0) + { + VMA_DEBUG_LOG(" vkAllocateMemory FAILED"); + return res; + } + + void* pMappedData = VMA_NULL; + if(map) + { + res = (*m_VulkanFunctions.vkMapMemory)( + m_hDevice, + hMemory, + 0, + VK_WHOLE_SIZE, + 0, + &pMappedData); + if(res < 0) + { + VMA_DEBUG_LOG(" vkMapMemory FAILED"); + FreeVulkanMemory(memTypeIndex, size, hMemory); + return res; + } + } + + *pAllocation = m_AllocationObjectAllocator.Allocate(isMappingAllowed); + (*pAllocation)->InitDedicatedAllocation(this, 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) + { + FillAllocation(*pAllocation, VMA_ALLOCATION_FILL_PATTERN_CREATED); + } + + return VK_SUCCESS; +} + +void VmaAllocator_T::GetBufferMemoryRequirements( + VkBuffer hBuffer, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const +{ +#if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + if(m_UseKhrDedicatedAllocation || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VkBufferMemoryRequirementsInfo2KHR memReqInfo = { VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2_KHR }; + memReqInfo.buffer = hBuffer; + + VkMemoryDedicatedRequirementsKHR memDedicatedReq = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR }; + + VkMemoryRequirements2KHR memReq2 = { VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR }; + VmaPnextChainPushFront(&memReq2, &memDedicatedReq); + + (*m_VulkanFunctions.vkGetBufferMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); + + memReq = memReq2.memoryRequirements; + requiresDedicatedAllocation = (memDedicatedReq.requiresDedicatedAllocation != VK_FALSE); + prefersDedicatedAllocation = (memDedicatedReq.prefersDedicatedAllocation != VK_FALSE); + } + else +#endif // #if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + { + (*m_VulkanFunctions.vkGetBufferMemoryRequirements)(m_hDevice, hBuffer, &memReq); + requiresDedicatedAllocation = false; + prefersDedicatedAllocation = false; + } +} + +void VmaAllocator_T::GetImageMemoryRequirements( + VkImage hImage, + VkMemoryRequirements& memReq, + bool& requiresDedicatedAllocation, + bool& prefersDedicatedAllocation) const +{ +#if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + if(m_UseKhrDedicatedAllocation || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) + { + VkImageMemoryRequirementsInfo2KHR memReqInfo = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2_KHR }; + memReqInfo.image = hImage; + + VkMemoryDedicatedRequirementsKHR memDedicatedReq = { VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR }; + + VkMemoryRequirements2KHR memReq2 = { VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2_KHR }; + VmaPnextChainPushFront(&memReq2, &memDedicatedReq); + + (*m_VulkanFunctions.vkGetImageMemoryRequirements2KHR)(m_hDevice, &memReqInfo, &memReq2); + + memReq = memReq2.memoryRequirements; + requiresDedicatedAllocation = (memDedicatedReq.requiresDedicatedAllocation != VK_FALSE); + prefersDedicatedAllocation = (memDedicatedReq.prefersDedicatedAllocation != VK_FALSE); + } + else +#endif // #if VMA_DEDICATED_ALLOCATION || VMA_VULKAN_VERSION >= 1001000 + { + (*m_VulkanFunctions.vkGetImageMemoryRequirements)(m_hDevice, hImage, &memReq); + requiresDedicatedAllocation = false; + prefersDedicatedAllocation = false; + } +} + +VkResult VmaAllocator_T::FindMemoryTypeIndex( + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VmaBufferImageUsage 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, + VkImage dedicatedImage, + VmaBufferImageUsage dedicatedBufferImageUsage, + const VmaAllocationCreateInfo& createInfo, + VmaSuballocationType suballocType, + size_t allocationCount, + VmaAllocation* pAllocations) +{ + memset(pAllocations, 0, sizeof(VmaAllocation) * allocationCount); + + VMA_ASSERT(VmaIsPow2(vkMemReq.alignment)); + + if(vkMemReq.size == 0) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + + VmaAllocationCreateInfo createInfoFinal = createInfo; + VkResult res = CalcAllocationParams(createInfoFinal, requiresDedicatedAllocation, prefersDedicatedAllocation); + if(res != VK_SUCCESS) + return res; + + if(createInfoFinal.pool != VK_NULL_HANDLE) + { + VmaBlockVector& blockVector = createInfoFinal.pool->m_BlockVector; + return AllocateMemoryOfType( + createInfoFinal.pool, + vkMemReq.size, + vkMemReq.alignment, + prefersDedicatedAllocation, + dedicatedBuffer, + dedicatedImage, + dedicatedBufferImageUsage, + createInfoFinal, + blockVector.GetMemoryTypeIndex(), + suballocType, + createInfoFinal.pool->m_DedicatedAllocations, + blockVector, + allocationCount, + pAllocations); + } + else + { + // Bit mask of memory Vulkan types acceptable for this allocation. + uint32_t memoryTypeBits = vkMemReq.memoryTypeBits; + uint32_t memTypeIndex = UINT32_MAX; + 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, + dedicatedImage, + dedicatedBufferImageUsage, + createInfoFinal, + memTypeIndex, + suballocType, + m_DedicatedAllocations[memTypeIndex], + *blockVector, + allocationCount, + pAllocations); + // Allocation succeeded + if(res == VK_SUCCESS) + 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; + } +} + +void VmaAllocator_T::FreeMemory( + size_t allocationCount, + const VmaAllocation* pAllocations) +{ + VMA_ASSERT(pAllocations); + + for(size_t allocIndex = allocationCount; allocIndex--; ) + { + VmaAllocation allocation = pAllocations[allocIndex]; + + if(allocation != VK_NULL_HANDLE) + { + if(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->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::CalculateStatistics(VmaTotalStatistics* pStats) +{ + // Initialize. + 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]; + if (pBlockVector != VMA_NULL) + pBlockVector->AddDetailedStatistics(pStats->memoryType[memTypeIndex]); + } + + // Process custom pools. + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + for(VmaPool pool = m_Pools.Front(); pool != VMA_NULL; pool = m_Pools.GetNext(pool)) + { + 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) + { + m_DedicatedAllocations[memTypeIndex].AddDetailedStatistics(pStats->memoryType[memTypeIndex]); + } + + // 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::GetHeapBudgets(VmaBudget* outBudgets, uint32_t firstHeap, uint32_t heapCount) +{ +#if VMA_MEMORY_BUDGET + if(m_UseExtMemoryBudget) + { + if(m_Budget.m_OperationsSinceBudgetFetch < 30) + { + VmaMutexLockRead lockRead(m_Budget.m_BudgetMutex, m_UseMutex); + for(uint32_t i = 0; i < heapCount; ++i, ++outBudgets) + { + const uint32_t heapIndex = firstHeap + i; + + 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] + outBudgets->statistics.blockBytes > m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]) + { + outBudgets->usage = m_Budget.m_VulkanUsage[heapIndex] + + outBudgets->statistics.blockBytes - m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]; + } + else + { + outBudgets->usage = 0; + } + + // Have to take MIN with heap size because explicit HeapSizeLimit is included in it. + outBudgets->budget = VMA_MIN( + m_Budget.m_VulkanBudget[heapIndex], m_MemProps.memoryHeaps[heapIndex].size); + } + } + else + { + UpdateVulkanBudget(); // Outside of mutex lock + GetHeapBudgets(outBudgets, firstHeap, heapCount); // Recursion + } + } + else +#endif + { + for(uint32_t i = 0; i < heapCount; ++i, ++outBudgets) + { + const uint32_t heapIndex = firstHeap + i; + + 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]; + + outBudgets->usage = outBudgets->statistics.blockBytes; + outBudgets->budget = m_MemProps.memoryHeaps[heapIndex].size * 8 / 10; // 80% heuristics. + } + } +} + +void VmaAllocator_T::GetAllocationInfo(VmaAllocation hAllocation, VmaAllocationInfo* pAllocationInfo) +{ + 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(); +} + +void VmaAllocator_T::GetAllocationInfo2(VmaAllocation hAllocation, VmaAllocationInfo2* pAllocationInfo) +{ + GetAllocationInfo(hAllocation, &pAllocationInfo->allocationInfo); + + switch (hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + pAllocationInfo->blockSize = hAllocation->GetBlock()->m_pMetadata->GetSize(); + pAllocationInfo->dedicatedMemory = VK_FALSE; + break; + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + pAllocationInfo->blockSize = pAllocationInfo->allocationInfo.size; + pAllocationInfo->dedicatedMemory = VK_TRUE; + break; + default: + VMA_ASSERT(0); + } +} + +VkResult VmaAllocator_T::CreatePool(const VmaPoolCreateInfo* pCreateInfo, VmaPool* pPool) +{ + VMA_DEBUG_LOG_FORMAT(" CreatePool: MemoryTypeIndex=%" PRIu32 ", flags=%" PRIu32, pCreateInfo->memoryTypeIndex, pCreateInfo->flags); + + VmaPoolCreateInfo newCreateInfo = *pCreateInfo; + + // Protection against uninitialized new structure member. If garbage data are left there, this pointer dereference would crash. + if(pCreateInfo->pMemoryAllocateNext) + { + VMA_ASSERT(((const VkBaseInStructure*)pCreateInfo->pMemoryAllocateNext)->sType != 0); + } + + if(newCreateInfo.maxBlockCount == 0) + { + newCreateInfo.maxBlockCount = SIZE_MAX; + } + if(newCreateInfo.minBlockCount > newCreateInfo.maxBlockCount) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + // Memory type index out of range or forbidden. + if(pCreateInfo->memoryTypeIndex >= GetMemoryTypeCount() || + ((1u << pCreateInfo->memoryTypeIndex) & m_GlobalMemoryTypeBits) == 0) + { + return VK_ERROR_FEATURE_NOT_PRESENT; + } + if(newCreateInfo.minAllocationAlignment > 0) + { + VMA_ASSERT(VmaIsPow2(newCreateInfo.minAllocationAlignment)); + } + + const VkDeviceSize preferredBlockSize = CalcPreferredBlockSize(newCreateInfo.memoryTypeIndex); + + *pPool = vma_new(this, VmaPool_T)(this, newCreateInfo, preferredBlockSize); + + VkResult res = (*pPool)->m_BlockVector.CreateMinBlocks(); + if(res != VK_SUCCESS) + { + vma_delete(this, *pPool); + *pPool = VMA_NULL; + return res; + } + + // Add to m_Pools. + { + VmaMutexLockWrite lock(m_PoolsMutex, m_UseMutex); + (*pPool)->SetId(m_NextPoolId++); + m_Pools.PushBack(*pPool); + } + + return VK_SUCCESS; +} + +void VmaAllocator_T::DestroyPool(VmaPool pool) +{ + // Remove from m_Pools. + { + VmaMutexLockWrite lock(m_PoolsMutex, m_UseMutex); + m_Pools.Remove(pool); + } + + vma_delete(this, pool); +} + +void VmaAllocator_T::GetPoolStatistics(VmaPool pool, VmaStatistics* 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) +{ + m_CurrentFrameIndex.store(frameIndex); + +#if VMA_MEMORY_BUDGET + if(m_UseExtMemoryBudget) + { + UpdateVulkanBudget(); + } +#endif // #if VMA_MEMORY_BUDGET +} + +VkResult VmaAllocator_T::CheckPoolCorruption(VmaPool hPool) +{ + return hPool->m_BlockVector.CheckCorruption(); +} + +VkResult VmaAllocator_T::CheckCorruption(uint32_t memoryTypeBits) +{ + VkResult finalRes = VK_ERROR_FEATURE_NOT_PRESENT; + + // Process default pools. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + VmaBlockVector* const pBlockVector = m_pBlockVectors[memTypeIndex]; + if(pBlockVector != VMA_NULL) + { + VkResult localRes = pBlockVector->CheckCorruption(); + switch(localRes) + { + case VK_ERROR_FEATURE_NOT_PRESENT: + break; + case VK_SUCCESS: + finalRes = VK_SUCCESS; + break; + default: + return localRes; + } + } + } + + // Process custom pools. + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + for(VmaPool pool = m_Pools.Front(); pool != VMA_NULL; pool = m_Pools.GetNext(pool)) + { + if(((1u << pool->m_BlockVector.GetMemoryTypeIndex()) & memoryTypeBits) != 0) + { + VkResult localRes = pool->m_BlockVector.CheckCorruption(); + switch(localRes) + { + case VK_ERROR_FEATURE_NOT_PRESENT: + break; + case VK_SUCCESS: + finalRes = VK_SUCCESS; + break; + default: + return localRes; + } + } + } + } + + return finalRes; +} + +VkResult VmaAllocator_T::AllocateVulkanMemory(const VkMemoryAllocateInfo* pAllocateInfo, VkDeviceMemory* pMemory) +{ + AtomicTransactionalIncrement deviceMemoryCountIncrement; + const uint64_t prevDeviceMemoryCount = deviceMemoryCountIncrement.Increment(&m_DeviceMemoryCount); +#if VMA_DEBUG_DONT_EXCEED_MAX_MEMORY_ALLOCATION_COUNT + if(prevDeviceMemoryCount >= m_PhysicalDeviceProperties.limits.maxMemoryAllocationCount) + { + return VK_ERROR_TOO_MANY_OBJECTS; + } +#endif + + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(pAllocateInfo->memoryTypeIndex); + + // HeapSizeLimit is in effect for this heap. + if((m_HeapSizeLimitMask & (1u << heapIndex)) != 0) + { + const VkDeviceSize heapSize = m_MemProps.memoryHeaps[heapIndex].size; + VkDeviceSize blockBytes = m_Budget.m_BlockBytes[heapIndex]; + for(;;) + { + const VkDeviceSize blockBytesAfterAllocation = blockBytes + pAllocateInfo->allocationSize; + if(blockBytesAfterAllocation > heapSize) + { + return VK_ERROR_OUT_OF_DEVICE_MEMORY; + } + if(m_Budget.m_BlockBytes[heapIndex].compare_exchange_strong(blockBytes, blockBytesAfterAllocation)) + { + break; + } + } + } + else + { + 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); + + if(res == VK_SUCCESS) + { +#if VMA_MEMORY_BUDGET + ++m_Budget.m_OperationsSinceBudgetFetch; +#endif + + // Informative callback. + if(m_DeviceMemoryCallbacks.pfnAllocate != VMA_NULL) + { + (*m_DeviceMemoryCallbacks.pfnAllocate)(this, pAllocateInfo->memoryTypeIndex, *pMemory, pAllocateInfo->allocationSize, m_DeviceMemoryCallbacks.pUserData); + } + + deviceMemoryCountIncrement.Commit(); + } + else + { + --m_Budget.m_BlockCount[heapIndex]; + m_Budget.m_BlockBytes[heapIndex] -= pAllocateInfo->allocationSize; + } + + return res; +} + +void VmaAllocator_T::FreeVulkanMemory(uint32_t memoryType, VkDeviceSize size, VkDeviceMemory hMemory) +{ + // Informative callback. + if(m_DeviceMemoryCallbacks.pfnFree != VMA_NULL) + { + (*m_DeviceMemoryCallbacks.pfnFree)(this, memoryType, hMemory, size, m_DeviceMemoryCallbacks.pUserData); + } + + // VULKAN CALL vkFreeMemory. + (*m_VulkanFunctions.vkFreeMemory)(m_hDevice, hMemory, GetAllocationCallbacks()); + + const uint32_t heapIndex = MemoryTypeIndexToHeapIndex(memoryType); + --m_Budget.m_BlockCount[heapIndex]; + m_Budget.m_BlockBytes[heapIndex] -= size; + + --m_DeviceMemoryCount; +} + +VkResult VmaAllocator_T::BindVulkanBuffer( + VkDeviceMemory memory, + VkDeviceSize memoryOffset, + VkBuffer buffer, + const void* pNext) +{ + if(pNext != VMA_NULL) + { +#if VMA_VULKAN_VERSION >= 1001000 || VMA_BIND_MEMORY2 + if((m_UseKhrBindMemory2 || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) && + m_VulkanFunctions.vkBindBufferMemory2KHR != VMA_NULL) + { + VkBindBufferMemoryInfoKHR bindBufferMemoryInfo = { VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO_KHR }; + bindBufferMemoryInfo.pNext = pNext; + bindBufferMemoryInfo.buffer = buffer; + bindBufferMemoryInfo.memory = memory; + bindBufferMemoryInfo.memoryOffset = memoryOffset; + return (*m_VulkanFunctions.vkBindBufferMemory2KHR)(m_hDevice, 1, &bindBufferMemoryInfo); + } + else +#endif // #if VMA_VULKAN_VERSION >= 1001000 || VMA_BIND_MEMORY2 + { + return VK_ERROR_EXTENSION_NOT_PRESENT; + } + } + else + { + return (*m_VulkanFunctions.vkBindBufferMemory)(m_hDevice, buffer, memory, memoryOffset); + } +} + +VkResult VmaAllocator_T::BindVulkanImage( + VkDeviceMemory memory, + VkDeviceSize memoryOffset, + VkImage image, + const void* pNext) +{ + if(pNext != VMA_NULL) + { +#if VMA_VULKAN_VERSION >= 1001000 || VMA_BIND_MEMORY2 + if((m_UseKhrBindMemory2 || m_VulkanApiVersion >= VK_MAKE_VERSION(1, 1, 0)) && + m_VulkanFunctions.vkBindImageMemory2KHR != VMA_NULL) + { + VkBindImageMemoryInfoKHR bindBufferMemoryInfo = { VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO_KHR }; + bindBufferMemoryInfo.pNext = pNext; + bindBufferMemoryInfo.image = image; + bindBufferMemoryInfo.memory = memory; + bindBufferMemoryInfo.memoryOffset = memoryOffset; + return (*m_VulkanFunctions.vkBindImageMemory2KHR)(m_hDevice, 1, &bindBufferMemoryInfo); + } + else +#endif // #if VMA_BIND_MEMORY2 + { + return VK_ERROR_EXTENSION_NOT_PRESENT; + } + } + else + { + return (*m_VulkanFunctions.vkBindImageMemory)(m_hDevice, image, memory, memoryOffset); + } +} + +VkResult VmaAllocator_T::Map(VmaAllocation hAllocation, void** ppData) +{ + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* const pBlock = hAllocation->GetBlock(); + char *pBytes = VMA_NULL; + VkResult res = pBlock->Map(this, 1, (void**)&pBytes); + if(res == VK_SUCCESS) + { + *ppData = pBytes + (ptrdiff_t)hAllocation->GetOffset(); + hAllocation->BlockAllocMap(); + } + return res; + } + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + return hAllocation->DedicatedAllocMap(this, ppData); + default: + VMA_ASSERT(0); + return VK_ERROR_MEMORY_MAP_FAILED; + } +} + +void VmaAllocator_T::Unmap(VmaAllocation hAllocation) +{ + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* const pBlock = hAllocation->GetBlock(); + hAllocation->BlockAllocUnmap(); + pBlock->Unmap(this, 1); + } + break; + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + hAllocation->DedicatedAllocUnmap(this); + break; + default: + VMA_ASSERT(0); + } +} + +VkResult VmaAllocator_T::BindBufferMemory( + VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkBuffer hBuffer, + const void* pNext) +{ + VkResult res = VK_ERROR_UNKNOWN_COPY; + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + res = BindVulkanBuffer(hAllocation->GetMemory(), allocationLocalOffset, hBuffer, pNext); + break; + 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."); + res = pBlock->BindBufferMemory(this, hAllocation, allocationLocalOffset, hBuffer, pNext); + break; + } + default: + VMA_ASSERT(0); + } + return res; +} + +VkResult VmaAllocator_T::BindImageMemory( + VmaAllocation hAllocation, + VkDeviceSize allocationLocalOffset, + VkImage hImage, + const void* pNext) +{ + VkResult res = VK_ERROR_UNKNOWN_COPY; + switch(hAllocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + res = BindVulkanImage(hAllocation->GetMemory(), allocationLocalOffset, hImage, pNext); + break; + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + VmaDeviceMemoryBlock* pBlock = hAllocation->GetBlock(); + VMA_ASSERT(pBlock && "Binding image to allocation that doesn't belong to any block."); + res = pBlock->BindImageMemory(this, hAllocation, allocationLocalOffset, hImage, pNext); + break; + } + default: + VMA_ASSERT(0); + } + return res; +} + +VkResult VmaAllocator_T::FlushOrInvalidateAllocation( + VmaAllocation hAllocation, + VkDeviceSize offset, VkDeviceSize size, + VMA_CACHE_OPERATION op) +{ + VkResult res = VK_SUCCESS; + + VkMappedMemoryRange memRange = {}; + if(GetFlushOrInvalidateRange(hAllocation, offset, size, memRange)) + { + switch(op) + { + case VMA_CACHE_FLUSH: + res = (*GetVulkanFunctions().vkFlushMappedMemoryRanges)(m_hDevice, 1, &memRange); + break; + case VMA_CACHE_INVALIDATE: + res = (*GetVulkanFunctions().vkInvalidateMappedMemoryRanges)(m_hDevice, 1, &memRange); + break; + default: + VMA_ASSERT(0); + } + } + // else: Just ignore this call. + return res; +} + +VkResult VmaAllocator_T::FlushOrInvalidateAllocations( + uint32_t allocationCount, + const VmaAllocation* allocations, + const VkDeviceSize* offsets, const VkDeviceSize* sizes, + VMA_CACHE_OPERATION op) +{ + typedef VmaStlAllocator RangeAllocator; + typedef VmaSmallVector RangeVector; + RangeVector ranges = RangeVector(RangeAllocator(GetAllocationCallbacks())); + + for(uint32_t allocIndex = 0; allocIndex < allocationCount; ++allocIndex) + { + const VmaAllocation alloc = allocations[allocIndex]; + const VkDeviceSize offset = offsets != VMA_NULL ? offsets[allocIndex] : 0; + const VkDeviceSize size = sizes != VMA_NULL ? sizes[allocIndex] : VK_WHOLE_SIZE; + VkMappedMemoryRange newRange; + if(GetFlushOrInvalidateRange(alloc, offset, size, newRange)) + { + ranges.push_back(newRange); + } + } + + VkResult res = VK_SUCCESS; + if(!ranges.empty()) + { + switch(op) + { + case VMA_CACHE_FLUSH: + res = (*GetVulkanFunctions().vkFlushMappedMemoryRanges)(m_hDevice, (uint32_t)ranges.size(), ranges.data()); + break; + case VMA_CACHE_INVALIDATE: + res = (*GetVulkanFunctions().vkInvalidateMappedMemoryRanges)(m_hDevice, (uint32_t)ranges.size(), ranges.data()); + break; + default: + VMA_ASSERT(0); + } + } + // else: Just ignore this call. + return res; +} + +VkResult VmaAllocator_T::CopyMemoryToAllocation( + const void* pSrcHostPointer, + VmaAllocation dstAllocation, + VkDeviceSize dstAllocationLocalOffset, + VkDeviceSize size) +{ + void* dstMappedData = VMA_NULL; + VkResult res = Map(dstAllocation, &dstMappedData); + if(res == VK_SUCCESS) + { + memcpy((char*)dstMappedData + dstAllocationLocalOffset, pSrcHostPointer, (size_t)size); + Unmap(dstAllocation); + res = FlushOrInvalidateAllocation(dstAllocation, dstAllocationLocalOffset, size, VMA_CACHE_FLUSH); + } + return res; +} + +VkResult VmaAllocator_T::CopyAllocationToMemory( + VmaAllocation srcAllocation, + VkDeviceSize srcAllocationLocalOffset, + void* pDstHostPointer, + VkDeviceSize size) +{ + void* srcMappedData = VMA_NULL; + VkResult res = Map(srcAllocation, &srcMappedData); + if(res == VK_SUCCESS) + { + res = FlushOrInvalidateAllocation(srcAllocation, srcAllocationLocalOffset, size, VMA_CACHE_INVALIDATE); + if(res == VK_SUCCESS) + { + memcpy(pDstHostPointer, (const char*)srcMappedData + srcAllocationLocalOffset, (size_t)size); + Unmap(srcAllocation); + } + } + return res; +} + +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) + { + // Default pool + m_DedicatedAllocations[memTypeIndex].Unregister(allocation); + } + else + { + // Custom pool + parentPool->m_DedicatedAllocations.Unregister(allocation); + } + + VkDeviceMemory hMemory = allocation->GetMemory(); + + /* + There is no need to call this, because Vulkan spec allows to skip vkUnmapMemory + before vkFreeMemory. + + if(allocation->GetMappedData() != VMA_NULL) + { + (*m_VulkanFunctions.vkUnmapMemory)(m_hDevice, hMemory); + } + */ + + FreeVulkanMemory(memTypeIndex, allocation->GetSize(), hMemory); + + m_Budget.RemoveAllocation(MemoryTypeIndexToHeapIndex(allocation->GetMemoryTypeIndex()), allocation->GetSize()); + allocation->Destroy(this); + m_AllocationObjectAllocator.Free(allocation); + + VMA_DEBUG_LOG_FORMAT(" Freed DedicatedMemory MemoryTypeIndex=%" PRIu32, memTypeIndex); +} + +uint32_t VmaAllocator_T::CalculateGpuDefragmentationMemoryTypeBits() const +{ + VkBufferCreateInfo dummyBufCreateInfo; + VmaFillGpuDefragmentationBufferCreateInfo(dummyBufCreateInfo); + + uint32_t memoryTypeBits = 0; + + // Create buffer. + VkBuffer buf = VK_NULL_HANDLE; + VkResult res = (*GetVulkanFunctions().vkCreateBuffer)( + m_hDevice, &dummyBufCreateInfo, GetAllocationCallbacks(), &buf); + if(res == VK_SUCCESS) + { + // Query for supported memory types. + VkMemoryRequirements memReq; + (*GetVulkanFunctions().vkGetBufferMemoryRequirements)(m_hDevice, buf, &memReq); + memoryTypeBits = memReq.memoryTypeBits; + + // Destroy buffer. + (*GetVulkanFunctions().vkDestroyBuffer)(m_hDevice, buf, GetAllocationCallbacks()); + } + + return memoryTypeBits; +} + +uint32_t VmaAllocator_T::CalculateGlobalMemoryTypeBits() const +{ + // Make sure memory information is already fetched. + VMA_ASSERT(GetMemoryTypeCount() > 0); + + uint32_t memoryTypeBits = UINT32_MAX; + + if(!m_UseAmdDeviceCoherentMemory) + { + // Exclude memory types that have VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD. + for(uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + if((m_MemProps.memoryTypes[memTypeIndex].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD_COPY) != 0) + { + memoryTypeBits &= ~(1u << memTypeIndex); + } + } + } + + return memoryTypeBits; +} + +bool VmaAllocator_T::GetFlushOrInvalidateRange( + VmaAllocation allocation, + VkDeviceSize offset, VkDeviceSize size, + VkMappedMemoryRange& outRange) const +{ + const uint32_t memTypeIndex = allocation->GetMemoryTypeIndex(); + if(size > 0 && IsMemoryTypeNonCoherent(memTypeIndex)) + { + const VkDeviceSize nonCoherentAtomSize = m_PhysicalDeviceProperties.limits.nonCoherentAtomSize; + const VkDeviceSize allocationSize = allocation->GetSize(); + VMA_ASSERT(offset <= allocationSize); + + outRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE; + outRange.pNext = VMA_NULL; + outRange.memory = allocation->GetMemory(); + + switch(allocation->GetType()) + { + case VmaAllocation_T::ALLOCATION_TYPE_DEDICATED: + outRange.offset = VmaAlignDown(offset, nonCoherentAtomSize); + if(size == VK_WHOLE_SIZE) + { + outRange.size = allocationSize - outRange.offset; + } + else + { + VMA_ASSERT(offset + size <= allocationSize); + outRange.size = VMA_MIN( + VmaAlignUp(size + (offset - outRange.offset), nonCoherentAtomSize), + allocationSize - outRange.offset); + } + break; + case VmaAllocation_T::ALLOCATION_TYPE_BLOCK: + { + // 1. Still within this allocation. + outRange.offset = VmaAlignDown(offset, nonCoherentAtomSize); + if(size == VK_WHOLE_SIZE) + { + size = allocationSize - offset; + } + else + { + VMA_ASSERT(offset + size <= allocationSize); + } + outRange.size = VmaAlignUp(size + (offset - outRange.offset), nonCoherentAtomSize); + + // 2. Adjust to whole block. + const VkDeviceSize allocationOffset = allocation->GetOffset(); + VMA_ASSERT(allocationOffset % nonCoherentAtomSize == 0); + const VkDeviceSize blockSize = allocation->GetBlock()->m_pMetadata->GetSize(); + outRange.offset += allocationOffset; + outRange.size = VMA_MIN(outRange.size, blockSize - outRange.offset); + + break; + } + default: + VMA_ASSERT(0); + } + return true; + } + return false; +} + +#if VMA_MEMORY_BUDGET +void VmaAllocator_T::UpdateVulkanBudget() +{ + VMA_ASSERT(m_UseExtMemoryBudget); + + VkPhysicalDeviceMemoryProperties2KHR memProps = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_PROPERTIES_2_KHR }; + + VkPhysicalDeviceMemoryBudgetPropertiesEXT budgetProps = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT }; + VmaPnextChainPushFront(&memProps, &budgetProps); + + GetVulkanFunctions().vkGetPhysicalDeviceMemoryProperties2KHR(m_PhysicalDevice, &memProps); + + { + VmaMutexLockWrite lockWrite(m_Budget.m_BudgetMutex, m_UseMutex); + + for(uint32_t heapIndex = 0; heapIndex < GetMemoryHeapCount(); ++heapIndex) + { + m_Budget.m_VulkanUsage[heapIndex] = budgetProps.heapUsage[heapIndex]; + m_Budget.m_VulkanBudget[heapIndex] = budgetProps.heapBudget[heapIndex]; + m_Budget.m_BlockBytesAtBudgetFetch[heapIndex] = m_Budget.m_BlockBytes[heapIndex].load(); + + // Some bugged drivers return the budget incorrectly, e.g. 0 or much bigger than heap size. + if(m_Budget.m_VulkanBudget[heapIndex] == 0) + { + m_Budget.m_VulkanBudget[heapIndex] = m_MemProps.memoryHeaps[heapIndex].size * 8 / 10; // 80% heuristics. + } + else if(m_Budget.m_VulkanBudget[heapIndex] > m_MemProps.memoryHeaps[heapIndex].size) + { + m_Budget.m_VulkanBudget[heapIndex] = m_MemProps.memoryHeaps[heapIndex].size; + } + if(m_Budget.m_VulkanUsage[heapIndex] == 0 && m_Budget.m_BlockBytesAtBudgetFetch[heapIndex] > 0) + { + m_Budget.m_VulkanUsage[heapIndex] = m_Budget.m_BlockBytesAtBudgetFetch[heapIndex]; + } + } + m_Budget.m_OperationsSinceBudgetFetch = 0; + } +} +#endif // VMA_MEMORY_BUDGET + +void VmaAllocator_T::FillAllocation(const VmaAllocation hAllocation, uint8_t pattern) +{ + if(VMA_DEBUG_INITIALIZE_ALLOCATIONS && + hAllocation->IsMappingAllowed() && + (m_MemProps.memoryTypes[hAllocation->GetMemoryTypeIndex()].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) != 0) + { + void* pData = VMA_NULL; + VkResult res = Map(hAllocation, &pData); + if(res == VK_SUCCESS) + { + memset(pData, (int)pattern, (size_t)hAllocation->GetSize()); + FlushOrInvalidateAllocation(hAllocation, 0, VK_WHOLE_SIZE, VMA_CACHE_FLUSH); + Unmap(hAllocation); + } + else + { + VMA_ASSERT(0 && "VMA_DEBUG_INITIALIZE_ALLOCATIONS is enabled, but couldn't map memory to fill allocation."); + } + } +} + +uint32_t VmaAllocator_T::GetGpuDefragmentationMemoryTypeBits() +{ + uint32_t memoryTypeBits = m_GpuDefragmentationMemoryTypeBits.load(); + if(memoryTypeBits == UINT32_MAX) + { + memoryTypeBits = CalculateGpuDefragmentationMemoryTypeBits(); + m_GpuDefragmentationMemoryTypeBits.store(memoryTypeBits); + } + return memoryTypeBits; +} + +#if VMA_STATS_STRING_ENABLED +void VmaAllocator_T::PrintDetailedMap(VmaJsonWriter& json) +{ + json.WriteString("DefaultPools"); + json.BeginObject(); + { + for (uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + VmaBlockVector* pBlockVector = m_pBlockVectors[memTypeIndex]; + VmaDedicatedAllocationList& dedicatedAllocList = m_DedicatedAllocations[memTypeIndex]; + if (pBlockVector != VMA_NULL) + { + json.BeginString("Type "); + json.ContinueString(memTypeIndex); + json.EndString(); + json.BeginObject(); + { + json.WriteString("PreferredBlockSize"); + json.WriteNumber(pBlockVector->GetPreferredBlockSize()); + + json.WriteString("Blocks"); + pBlockVector->PrintDetailedMap(json); + + json.WriteString("DedicatedAllocations"); + dedicatedAllocList.BuildStatsString(json); + } + json.EndObject(); + } + } + } + json.EndObject(); + + json.WriteString("CustomPools"); + json.BeginObject(); + { + VmaMutexLockRead lock(m_PoolsMutex, m_UseMutex); + if (!m_Pools.IsEmpty()) + { + for (uint32_t memTypeIndex = 0; memTypeIndex < GetMemoryTypeCount(); ++memTypeIndex) + { + 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; + } + + json.BeginObject(); + { + json.WriteString("Name"); + json.BeginString(); + json.ContinueString((uint64_t)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(); +} +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_ALLOCATOR_T_FUNCTIONS + + +#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) <= 4)); + VMA_DEBUG_LOG("vmaCreateAllocator"); + *pAllocator = vma_new(pCreateInfo->pAllocationCallbacks, VmaAllocator_T)(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( + VmaAllocator allocator) +{ + if(allocator != VK_NULL_HANDLE) + { + VMA_DEBUG_LOG("vmaDestroyAllocator"); + VkAllocationCallbacks allocationCallbacks = allocator->m_AllocationCallbacks; // Have to copy the callbacks when destroying. + vma_delete(&allocationCallbacks, allocator); + } +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocatorInfo(VmaAllocator allocator, VmaAllocatorInfo* pAllocatorInfo) +{ + VMA_ASSERT(allocator && pAllocatorInfo); + pAllocatorInfo->instance = allocator->m_hInstance; + pAllocatorInfo->physicalDevice = allocator->GetPhysicalDevice(); + pAllocatorInfo->device = allocator->m_hDevice; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetPhysicalDeviceProperties( + VmaAllocator allocator, + const VkPhysicalDeviceProperties **ppPhysicalDeviceProperties) +{ + VMA_ASSERT(allocator && ppPhysicalDeviceProperties); + *ppPhysicalDeviceProperties = &allocator->m_PhysicalDeviceProperties; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryProperties( + VmaAllocator allocator, + const VkPhysicalDeviceMemoryProperties** ppPhysicalDeviceMemoryProperties) +{ + VMA_ASSERT(allocator && ppPhysicalDeviceMemoryProperties); + *ppPhysicalDeviceMemoryProperties = &allocator->m_MemProps; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetMemoryTypeProperties( + VmaAllocator allocator, + uint32_t memoryTypeIndex, + VkMemoryPropertyFlags* pFlags) +{ + VMA_ASSERT(allocator && pFlags); + VMA_ASSERT(memoryTypeIndex < allocator->GetMemoryTypeCount()); + *pFlags = allocator->m_MemProps.memoryTypes[memoryTypeIndex].propertyFlags; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaSetCurrentFrameIndex( + VmaAllocator allocator, + uint32_t frameIndex) +{ + VMA_ASSERT(allocator); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->SetCurrentFrameIndex(frameIndex); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaCalculateStatistics( + VmaAllocator allocator, + VmaTotalStatistics* pStats) +{ + VMA_ASSERT(allocator && pStats); + VMA_DEBUG_GLOBAL_MUTEX_LOCK + allocator->CalculateStatistics(pStats); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetHeapBudgets( + VmaAllocator allocator, + VmaBudget* pBudgets) +{ + VMA_ASSERT(allocator && pBudgets); + VMA_DEBUG_GLOBAL_MUTEX_LOCK + allocator->GetHeapBudgets(pBudgets, 0, allocator->GetMemoryHeapCount()); +} + +#if VMA_STATS_STRING_ENABLED + +VMA_CALL_PRE void VMA_CALL_POST vmaBuildStatsString( + VmaAllocator allocator, + char** ppStatsString, + VkBool32 detailedMap) +{ + VMA_ASSERT(allocator && ppStatsString); + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + 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(); + { + json.WriteString("General"); + json.BeginObject(); + { + 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_VERSION_MAJOR(deviceProperties.apiVersion)); + json.ContinueString("."); + json.ContinueString(VK_VERSION_MINOR(deviceProperties.apiVersion)); + json.ContinueString("."); + json.ContinueString(VK_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(); + } + { + json.WriteString("Total"); + VmaPrintDetailedStatistics(json, stats.total); + } + { + json.WriteString("MemoryInfo"); + json.BeginObject(); + { + for (uint32_t heapIndex = 0; heapIndex < allocator->GetMemoryHeapCount(); ++heapIndex) + { + 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 + + 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(); + + json.WriteString("Stats"); + 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) + allocator->PrintDetailedMap(json); + + json.EndObject(); + } + + *ppStatsString = VmaCreateStringCopy(allocator->GetAllocationCallbacks(), sb.GetData(), sb.GetLength()); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaFreeStatsString( + VmaAllocator allocator, + char* pStatsString) +{ + if(pStatsString != VMA_NULL) + { + VMA_ASSERT(allocator); + VmaFreeString(allocator->GetAllocationCallbacks(), pStatsString); + } +} + +#endif // VMA_STATS_STRING_ENABLED + +/* +This function is not protected by any mutex because it just reads immutable data. +*/ +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndex( + VmaAllocator allocator, + uint32_t memoryTypeBits, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + return allocator->FindMemoryTypeIndex(memoryTypeBits, pAllocationCreateInfo, VmaBufferImageUsage::UNKNOWN, pMemoryTypeIndex); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForBufferInfo( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pBufferCreateInfo != VMA_NULL); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + const VkDevice hDev = allocator->m_hDevice; + const VmaVulkanFunctions* funcs = &allocator->GetVulkanFunctions(); + VkResult res; + +#if VMA_KHR_MAINTENANCE4 || VMA_VULKAN_VERSION >= 1003000 + if(funcs->vkGetDeviceBufferMemoryRequirements) + { + // Can query straight from VkBufferCreateInfo :) + VkDeviceBufferMemoryRequirementsKHR devBufMemReq = {VK_STRUCTURE_TYPE_DEVICE_BUFFER_MEMORY_REQUIREMENTS_KHR}; + devBufMemReq.pCreateInfo = pBufferCreateInfo; + + VkMemoryRequirements2 memReq = {VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2}; + (*funcs->vkGetDeviceBufferMemoryRequirements)(hDev, &devBufMemReq, &memReq); + + res = allocator->FindMemoryTypeIndex( + memReq.memoryRequirements.memoryTypeBits, pAllocationCreateInfo, + VmaBufferImageUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5), pMemoryTypeIndex); + } + else +#endif // VMA_KHR_MAINTENANCE4 || 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, + VmaBufferImageUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5), pMemoryTypeIndex); + + funcs->vkDestroyBuffer( + hDev, hBuffer, allocator->GetAllocationCallbacks()); + } + } + return res; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFindMemoryTypeIndexForImageInfo( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + uint32_t* pMemoryTypeIndex) +{ + VMA_ASSERT(allocator != VK_NULL_HANDLE); + VMA_ASSERT(pImageCreateInfo != VMA_NULL); + VMA_ASSERT(pAllocationCreateInfo != VMA_NULL); + VMA_ASSERT(pMemoryTypeIndex != VMA_NULL); + + const VkDevice hDev = allocator->m_hDevice; + const VmaVulkanFunctions* funcs = &allocator->GetVulkanFunctions(); + VkResult res; + +#if VMA_KHR_MAINTENANCE4 || VMA_VULKAN_VERSION >= 1003000 + if(funcs->vkGetDeviceImageMemoryRequirements) + { + // Can query straight from VkImageCreateInfo :) + VkDeviceImageMemoryRequirementsKHR devImgMemReq = {VK_STRUCTURE_TYPE_DEVICE_IMAGE_MEMORY_REQUIREMENTS_KHR}; + 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."); + + VkMemoryRequirements2 memReq = {VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2}; + (*funcs->vkGetDeviceImageMemoryRequirements)(hDev, &devImgMemReq, &memReq); + + res = allocator->FindMemoryTypeIndex( + memReq.memoryRequirements.memoryTypeBits, pAllocationCreateInfo, + VmaBufferImageUsage(*pImageCreateInfo), pMemoryTypeIndex); + } + else +#endif // VMA_KHR_MAINTENANCE4 || 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, + VmaBufferImageUsage(*pImageCreateInfo), pMemoryTypeIndex); + + funcs->vkDestroyImage( + hDev, hImage, allocator->GetAllocationCallbacks()); + } + } + return res; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreatePool( + VmaAllocator allocator, + const VmaPoolCreateInfo* pCreateInfo, + VmaPool* pPool) +{ + VMA_ASSERT(allocator && pCreateInfo && pPool); + + VMA_DEBUG_LOG("vmaCreatePool"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->CreatePool(pCreateInfo, pPool); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaDestroyPool( + VmaAllocator allocator, + VmaPool pool) +{ + VMA_ASSERT(allocator); + + if(pool == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyPool"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->DestroyPool(pool); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolStatistics( + VmaAllocator allocator, + VmaPool pool, + VmaStatistics* pPoolStats) +{ + VMA_ASSERT(allocator && pool && pPoolStats); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->GetPoolStatistics(pool, pPoolStats); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaCalculatePoolStatistics( + VmaAllocator allocator, + VmaPool pool, + VmaDetailedStatistics* pPoolStats) +{ + VMA_ASSERT(allocator && pool && pPoolStats); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->CalculatePoolStatistics(pool, pPoolStats); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckPoolCorruption(VmaAllocator allocator, VmaPool pool) +{ + VMA_ASSERT(allocator && pool); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VMA_DEBUG_LOG("vmaCheckPoolCorruption"); + + return allocator->CheckPoolCorruption(pool); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetPoolName( + VmaAllocator allocator, + VmaPool pool, + const char** ppName) +{ + VMA_ASSERT(allocator && pool && ppName); + + VMA_DEBUG_LOG("vmaGetPoolName"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *ppName = pool->GetName(); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaSetPoolName( + VmaAllocator allocator, + VmaPool pool, + const char* pName) +{ + VMA_ASSERT(allocator && pool); + + VMA_DEBUG_LOG("vmaSetPoolName"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + pool->SetName(pName); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemory( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pVkMemoryRequirements && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult result = allocator->AllocateMemory( + *pVkMemoryRequirements, + false, // requiresDedicatedAllocation + false, // prefersDedicatedAllocation + VK_NULL_HANDLE, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + VmaBufferImageUsage::UNKNOWN, // dedicatedBufferImageUsage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_UNKNOWN, + 1, // allocationCount + pAllocation); + + if(pAllocationInfo != VMA_NULL && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryPages( + VmaAllocator allocator, + const VkMemoryRequirements* pVkMemoryRequirements, + const VmaAllocationCreateInfo* pCreateInfo, + size_t allocationCount, + VmaAllocation* pAllocations, + VmaAllocationInfo* pAllocationInfo) +{ + if(allocationCount == 0) + { + return VK_SUCCESS; + } + + VMA_ASSERT(allocator && pVkMemoryRequirements && pCreateInfo && pAllocations); + + VMA_DEBUG_LOG("vmaAllocateMemoryPages"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkResult result = allocator->AllocateMemory( + *pVkMemoryRequirements, + false, // requiresDedicatedAllocation + false, // prefersDedicatedAllocation + VK_NULL_HANDLE, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + VmaBufferImageUsage::UNKNOWN, // dedicatedBufferImageUsage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_UNKNOWN, + allocationCount, + pAllocations); + + if(pAllocationInfo != VMA_NULL && result == VK_SUCCESS) + { + for(size_t i = 0; i < allocationCount; ++i) + { + allocator->GetAllocationInfo(pAllocations[i], pAllocationInfo + i); + } + } + + return result; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForBuffer( + VmaAllocator allocator, + VkBuffer buffer, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && buffer != VK_NULL_HANDLE && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemoryForBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetBufferMemoryRequirements(buffer, vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation); + + VkResult result = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + buffer, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + VmaBufferImageUsage::UNKNOWN, // dedicatedBufferImageUsage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_BUFFER, + 1, // allocationCount + pAllocation); + + if(pAllocationInfo && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaAllocateMemoryForImage( + VmaAllocator allocator, + VkImage image, + const VmaAllocationCreateInfo* pCreateInfo, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && image != VK_NULL_HANDLE && pCreateInfo && pAllocation); + + VMA_DEBUG_LOG("vmaAllocateMemoryForImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetImageMemoryRequirements(image, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + VkResult result = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + VK_NULL_HANDLE, // dedicatedBuffer + image, // dedicatedImage + VmaBufferImageUsage::UNKNOWN, // dedicatedBufferImageUsage + *pCreateInfo, + VMA_SUBALLOCATION_TYPE_IMAGE_UNKNOWN, + 1, // allocationCount + pAllocation); + + if(pAllocationInfo && result == VK_SUCCESS) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return result; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaFreeMemory( + VmaAllocator allocator, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator); + + if(allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaFreeMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->FreeMemory( + 1, // allocationCount + &allocation); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaFreeMemoryPages( + VmaAllocator allocator, + size_t allocationCount, + const VmaAllocation* pAllocations) +{ + if(allocationCount == 0) + { + return; + } + + VMA_ASSERT(allocator); + + VMA_DEBUG_LOG("vmaFreeMemoryPages"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->FreeMemory(allocationCount, pAllocations); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationInfo( + VmaAllocator allocator, + VmaAllocation allocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && allocation && pAllocationInfo); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->GetAllocationInfo(allocation, pAllocationInfo); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaGetAllocationInfo2( + VmaAllocator allocator, + VmaAllocation allocation, + VmaAllocationInfo2* pAllocationInfo) +{ + VMA_ASSERT(allocator && allocation && pAllocationInfo); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->GetAllocationInfo2(allocation, pAllocationInfo); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationUserData( + VmaAllocator allocator, + VmaAllocation allocation, + void* pUserData) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocation->SetUserData(allocator, pUserData); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaSetAllocationName( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + const char* VMA_NULLABLE pName) +{ + allocation->SetName(allocator, pName); +} + +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( + VmaAllocator allocator, + VmaAllocation allocation, + void** ppData) +{ + VMA_ASSERT(allocator && allocation && ppData); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->Map(allocation, ppData); +} + +VMA_CALL_PRE void VMA_CALL_POST vmaUnmapMemory( + VmaAllocator allocator, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + allocator->Unmap(allocation); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize offset, + VkDeviceSize size) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_LOG("vmaFlushAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->FlushOrInvalidateAllocation(allocation, offset, size, VMA_CACHE_FLUSH); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocation( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize offset, + VkDeviceSize size) +{ + VMA_ASSERT(allocator && allocation); + + VMA_DEBUG_LOG("vmaInvalidateAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->FlushOrInvalidateAllocation(allocation, offset, size, VMA_CACHE_INVALIDATE); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaFlushAllocations( + VmaAllocator allocator, + uint32_t allocationCount, + const VmaAllocation* allocations, + const VkDeviceSize* offsets, + const VkDeviceSize* sizes) +{ + VMA_ASSERT(allocator); + + if(allocationCount == 0) + { + return VK_SUCCESS; + } + + VMA_ASSERT(allocations); + + VMA_DEBUG_LOG("vmaFlushAllocations"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->FlushOrInvalidateAllocations(allocationCount, allocations, offsets, sizes, VMA_CACHE_FLUSH); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaInvalidateAllocations( + VmaAllocator allocator, + uint32_t allocationCount, + const VmaAllocation* allocations, + const VkDeviceSize* offsets, + const VkDeviceSize* sizes) +{ + VMA_ASSERT(allocator); + + if(allocationCount == 0) + { + return VK_SUCCESS; + } + + VMA_ASSERT(allocations); + + VMA_DEBUG_LOG("vmaInvalidateAllocations"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->FlushOrInvalidateAllocations(allocationCount, allocations, offsets, sizes, VMA_CACHE_INVALIDATE); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCopyMemoryToAllocation( + VmaAllocator allocator, + const void* pSrcHostPointer, + VmaAllocation dstAllocation, + VkDeviceSize dstAllocationLocalOffset, + VkDeviceSize size) +{ + VMA_ASSERT(allocator && pSrcHostPointer && dstAllocation); + + if(size == 0) + { + return VK_SUCCESS; + } + + VMA_DEBUG_LOG("vmaCopyMemoryToAllocation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->CopyMemoryToAllocation(pSrcHostPointer, dstAllocation, dstAllocationLocalOffset, size); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCopyAllocationToMemory( + VmaAllocator allocator, + VmaAllocation srcAllocation, + VkDeviceSize srcAllocationLocalOffset, + void* pDstHostPointer, + VkDeviceSize size) +{ + VMA_ASSERT(allocator && srcAllocation && pDstHostPointer); + + if(size == 0) + { + return VK_SUCCESS; + } + + VMA_DEBUG_LOG("vmaCopyAllocationToMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->CopyAllocationToMemory(srcAllocation, srcAllocationLocalOffset, pDstHostPointer, size); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCheckCorruption( + VmaAllocator allocator, + uint32_t memoryTypeBits) +{ + VMA_ASSERT(allocator); + + VMA_DEBUG_LOG("vmaCheckCorruption"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->CheckCorruption(memoryTypeBits); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentation( + VmaAllocator allocator, + const VmaDefragmentationInfo* pInfo, + VmaDefragmentationContext* pContext) +{ + VMA_ASSERT(allocator && pInfo && pContext); + + VMA_DEBUG_LOG("vmaBeginDefragmentation"); + + if (pInfo->pool != VMA_NULL) + { + // 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_DEBUG_GLOBAL_MUTEX_LOCK + + *pContext = vma_new(allocator, VmaDefragmentationContext_T)(allocator, *pInfo); + return VK_SUCCESS; +} + +VMA_CALL_PRE void VMA_CALL_POST vmaEndDefragmentation( + VmaAllocator allocator, + VmaDefragmentationContext context, + VmaDefragmentationStats* pStats) +{ + VMA_ASSERT(allocator && context); + + VMA_DEBUG_LOG("vmaEndDefragmentation"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + if (pStats) + context->GetStats(*pStats); + vma_delete(allocator, context); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBeginDefragmentationPass( + VmaAllocator VMA_NOT_NULL allocator, + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo) +{ + VMA_ASSERT(context && pPassInfo); + + VMA_DEBUG_LOG("vmaBeginDefragmentationPass"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return context->DefragmentPassBegin(*pPassInfo); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaEndDefragmentationPass( + VmaAllocator VMA_NOT_NULL allocator, + VmaDefragmentationContext VMA_NOT_NULL context, + VmaDefragmentationPassMoveInfo* VMA_NOT_NULL pPassInfo) +{ + VMA_ASSERT(context && pPassInfo); + + VMA_DEBUG_LOG("vmaEndDefragmentationPass"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return context->DefragmentPassEnd(*pPassInfo); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkBuffer buffer) +{ + VMA_ASSERT(allocator && allocation && buffer); + + VMA_DEBUG_LOG("vmaBindBufferMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindBufferMemory(allocation, 0, buffer, VMA_NULL); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindBufferMemory2( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize allocationLocalOffset, + VkBuffer buffer, + const void* pNext) +{ + VMA_ASSERT(allocator && allocation && buffer); + + VMA_DEBUG_LOG("vmaBindBufferMemory2"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindBufferMemory(allocation, allocationLocalOffset, buffer, pNext); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory( + VmaAllocator allocator, + VmaAllocation allocation, + VkImage image) +{ + VMA_ASSERT(allocator && allocation && image); + + VMA_DEBUG_LOG("vmaBindImageMemory"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindImageMemory(allocation, 0, image, VMA_NULL); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaBindImageMemory2( + VmaAllocator allocator, + VmaAllocation allocation, + VkDeviceSize allocationLocalOffset, + VkImage image, + const void* pNext) +{ + VMA_ASSERT(allocator && allocation && image); + + VMA_DEBUG_LOG("vmaBindImageMemory2"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + return allocator->BindImageMemory(allocation, allocationLocalOffset, image, pNext); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBuffer( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkBuffer* pBuffer, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pBufferCreateInfo && pAllocationCreateInfo && pBuffer && pAllocation); + + if(pBufferCreateInfo->size == 0) + { + return VK_ERROR_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_LOG("vmaCreateBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *pBuffer = VK_NULL_HANDLE; + *pAllocation = VK_NULL_HANDLE; + + // 1. Create VkBuffer. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateBuffer)( + allocator->m_hDevice, + pBufferCreateInfo, + allocator->GetAllocationCallbacks(), + pBuffer); + if(res >= 0) + { + // 2. vkGetBufferMemoryRequirements. + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetBufferMemoryRequirements(*pBuffer, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + // 3. Allocate memory using allocator. + res = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + *pBuffer, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + VmaBufferImageUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5), // dedicatedBufferImageUsage + *pAllocationCreateInfo, + VMA_SUBALLOCATION_TYPE_BUFFER, + 1, // allocationCount + pAllocation); + + if(res >= 0) + { + // 3. Bind buffer with memory. + if((pAllocationCreateInfo->flags & VMA_ALLOCATION_CREATE_DONT_BIND_BIT) == 0) + { + res = allocator->BindBufferMemory(*pAllocation, 0, *pBuffer, VMA_NULL); + } + if(res >= 0) + { + // All steps succeeded. + #if VMA_STATS_STRING_ENABLED + (*pAllocation)->InitBufferUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5); + #endif + if(pAllocationInfo != VMA_NULL) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return VK_SUCCESS; + } + allocator->FreeMemory( + 1, // allocationCount + pAllocation); + *pAllocation = VK_NULL_HANDLE; + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + return res; +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateBufferWithAlignment( + VmaAllocator allocator, + const VkBufferCreateInfo* pBufferCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkDeviceSize minAlignment, + VkBuffer* pBuffer, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pBufferCreateInfo && pAllocationCreateInfo && VmaIsPow2(minAlignment) && pBuffer && pAllocation); + + 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_LOG("vmaCreateBufferWithAlignment"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *pBuffer = VK_NULL_HANDLE; + *pAllocation = VK_NULL_HANDLE; + + // 1. Create VkBuffer. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateBuffer)( + allocator->m_hDevice, + pBufferCreateInfo, + allocator->GetAllocationCallbacks(), + pBuffer); + if(res >= 0) + { + // 2. vkGetBufferMemoryRequirements. + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetBufferMemoryRequirements(*pBuffer, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + // 2a. Include minAlignment + vkMemReq.alignment = VMA_MAX(vkMemReq.alignment, minAlignment); + + // 3. Allocate memory using allocator. + res = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + *pBuffer, // dedicatedBuffer + VK_NULL_HANDLE, // dedicatedImage + VmaBufferImageUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5), // dedicatedBufferImageUsage + *pAllocationCreateInfo, + VMA_SUBALLOCATION_TYPE_BUFFER, + 1, // allocationCount + pAllocation); + + if(res >= 0) + { + // 3. Bind buffer with memory. + if((pAllocationCreateInfo->flags & VMA_ALLOCATION_CREATE_DONT_BIND_BIT) == 0) + { + res = allocator->BindBufferMemory(*pAllocation, 0, *pBuffer, VMA_NULL); + } + if(res >= 0) + { + // All steps succeeded. + #if VMA_STATS_STRING_ENABLED + (*pAllocation)->InitBufferUsage(*pBufferCreateInfo, allocator->m_UseKhrMaintenance5); + #endif + if(pAllocationInfo != VMA_NULL) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return VK_SUCCESS; + } + allocator->FreeMemory( + 1, // allocationCount + pAllocation); + *pAllocation = VK_NULL_HANDLE; + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, *pBuffer, allocator->GetAllocationCallbacks()); + *pBuffer = VK_NULL_HANDLE; + return res; + } + return res; +} + +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) +{ + return vmaCreateAliasingBuffer2(allocator, allocation, 0, pBufferCreateInfo, pBuffer); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingBuffer2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + const VkBufferCreateInfo* VMA_NOT_NULL pBufferCreateInfo, + VkBuffer VMA_NULLABLE_NON_DISPATCHABLE* VMA_NOT_NULL pBuffer) +{ + VMA_ASSERT(allocator && pBufferCreateInfo && pBuffer && allocation); + VMA_ASSERT(allocationLocalOffset + pBufferCreateInfo->size <= allocation->GetSize()); + + VMA_DEBUG_LOG("vmaCreateAliasingBuffer2"); + + *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, allocationLocalOffset, *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, + VmaAllocation allocation) +{ + VMA_ASSERT(allocator); + + if(buffer == VK_NULL_HANDLE && allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyBuffer"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + if(buffer != VK_NULL_HANDLE) + { + (*allocator->GetVulkanFunctions().vkDestroyBuffer)(allocator->m_hDevice, buffer, allocator->GetAllocationCallbacks()); + } + + if(allocation != VK_NULL_HANDLE) + { + allocator->FreeMemory( + 1, // allocationCount + &allocation); + } +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateImage( + VmaAllocator allocator, + const VkImageCreateInfo* pImageCreateInfo, + const VmaAllocationCreateInfo* pAllocationCreateInfo, + VkImage* pImage, + VmaAllocation* pAllocation, + VmaAllocationInfo* pAllocationInfo) +{ + VMA_ASSERT(allocator && pImageCreateInfo && pAllocationCreateInfo && pImage && pAllocation); + + if(pImageCreateInfo->extent.width == 0 || + pImageCreateInfo->extent.height == 0 || + pImageCreateInfo->extent.depth == 0 || + pImageCreateInfo->mipLevels == 0 || + pImageCreateInfo->arrayLayers == 0) + { + return VK_ERROR_INITIALIZATION_FAILED; + } + + VMA_DEBUG_LOG("vmaCreateImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + *pImage = VK_NULL_HANDLE; + *pAllocation = VK_NULL_HANDLE; + + // 1. Create VkImage. + VkResult res = (*allocator->GetVulkanFunctions().vkCreateImage)( + allocator->m_hDevice, + pImageCreateInfo, + allocator->GetAllocationCallbacks(), + pImage); + if(res == VK_SUCCESS) + { + VmaSuballocationType suballocType = pImageCreateInfo->tiling == VK_IMAGE_TILING_OPTIMAL ? + VMA_SUBALLOCATION_TYPE_IMAGE_OPTIMAL : + VMA_SUBALLOCATION_TYPE_IMAGE_LINEAR; + + // 2. Allocate memory using allocator. + VkMemoryRequirements vkMemReq = {}; + bool requiresDedicatedAllocation = false; + bool prefersDedicatedAllocation = false; + allocator->GetImageMemoryRequirements(*pImage, vkMemReq, + requiresDedicatedAllocation, prefersDedicatedAllocation); + + res = allocator->AllocateMemory( + vkMemReq, + requiresDedicatedAllocation, + prefersDedicatedAllocation, + VK_NULL_HANDLE, // dedicatedBuffer + *pImage, // dedicatedImage + VmaBufferImageUsage(*pImageCreateInfo), // dedicatedBufferImageUsage + *pAllocationCreateInfo, + suballocType, + 1, // allocationCount + pAllocation); + + if(res == VK_SUCCESS) + { + // 3. Bind image with memory. + if((pAllocationCreateInfo->flags & VMA_ALLOCATION_CREATE_DONT_BIND_BIT) == 0) + { + res = allocator->BindImageMemory(*pAllocation, 0, *pImage, VMA_NULL); + } + if(res == VK_SUCCESS) + { + // All steps succeeded. + #if VMA_STATS_STRING_ENABLED + (*pAllocation)->InitImageUsage(*pImageCreateInfo); + #endif + if(pAllocationInfo != VMA_NULL) + { + allocator->GetAllocationInfo(*pAllocation, pAllocationInfo); + } + + return VK_SUCCESS; + } + allocator->FreeMemory( + 1, // allocationCount + pAllocation); + *pAllocation = VK_NULL_HANDLE; + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, *pImage, allocator->GetAllocationCallbacks()); + *pImage = VK_NULL_HANDLE; + return res; + } + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, *pImage, allocator->GetAllocationCallbacks()); + *pImage = VK_NULL_HANDLE; + return res; + } + return res; +} + +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) +{ + return vmaCreateAliasingImage2(allocator, allocation, 0, pImageCreateInfo, pImage); +} + +VMA_CALL_PRE VkResult VMA_CALL_POST vmaCreateAliasingImage2( + VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, + VkDeviceSize allocationLocalOffset, + 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("vmaCreateImage2"); + + 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, allocationLocalOffset, *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 VMA_NOT_NULL allocator, + VkImage VMA_NULLABLE_NON_DISPATCHABLE image, + VmaAllocation VMA_NULLABLE allocation) +{ + VMA_ASSERT(allocator); + + if(image == VK_NULL_HANDLE && allocation == VK_NULL_HANDLE) + { + return; + } + + VMA_DEBUG_LOG("vmaDestroyImage"); + + VMA_DEBUG_GLOBAL_MUTEX_LOCK + + if(image != VK_NULL_HANDLE) + { + (*allocator->GetVulkanFunctions().vkDestroyImage)(allocator->m_hDevice, image, allocator->GetAllocationCallbacks()); + } + if(allocation != VK_NULL_HANDLE) + { + allocator->FreeMemory( + 1, // allocationCount + &allocation); + } +} + +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); + } +} +#if VMA_EXTERNAL_MEMORY_WIN32 +VMA_CALL_PRE VkResult VMA_CALL_POST vmaGetMemoryWin32Handle(VmaAllocator VMA_NOT_NULL allocator, + VmaAllocation VMA_NOT_NULL allocation, HANDLE hTargetProcess, HANDLE* VMA_NOT_NULL pHandle) +{ + VMA_ASSERT(allocator && allocation && pHandle); + VMA_DEBUG_GLOBAL_MUTEX_LOCK; + return allocation->GetWin32Handle(allocator, hTargetProcess, pHandle); +} +#endif // VMA_EXTERNAL_MEMORY_WIN32 +#endif // VMA_STATS_STRING_ENABLED +#endif // _VMA_PUBLIC_INTERFACE +#endif // VMA_IMPLEMENTATION + +/** +\page quick_start Quick start + +\section quick_start_project_setup Project setup + +Vulkan Memory Allocator comes in form of a "stb-style" single header file. +While you can pull the entire repository e.g. as Git module, there is also Cmake script provided, +you don't need to build it as a separate library project. +You can add file "vk_mem_alloc.h" directly to your project and submit it to code repository next to your other source files. + +"Single header" doesn't mean that everything is contained in C/C++ declarations, +like it tends to be in case of inline functions or C++ templates. +It means that implementation is bundled with interface in a single file and needs to be extracted using preprocessor macro. +If you don't do it properly, it will result in linker errors. + +To do it properly: + +-# Include "vk_mem_alloc.h" file in each CPP file where you want to use the library. + This includes declarations of all members of the library. +-# In exactly one CPP file define following macro before this include. + It enables also internal definitions. + +\code +#define VMA_IMPLEMENTATION +#include "vk_mem_alloc.h" +\endcode + +It may be a good idea to create dedicated CPP file just for this purpose, e.g. "VmaUsage.cpp". + +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. +It may be a good idea to create a dedicate header file for this purpose, e.g. "VmaUsage.h", +that will be included in other source files instead of VMA header directly. + +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 are used and required. Features of C++20 are used optionally when available. +Some headers of standard C and C++ library are used, but STL containers, RTTI, or C++ exceptions are not used. + + +\section quick_start_initialization Initialization + +VMA offers library interface in a style similar to Vulkan, with object handles like #VmaAllocation, +structures describing parameters of objects to be created like #VmaAllocationCreateInfo, +and errors codes returned from functions using `VkResult` type. + +The first and the main object that needs to be created is #VmaAllocator. +It represents the initialization of the entire library. +Only one such object should be created per `VkDevice`. +You should create it at program startup, after `VkDevice` was created, and before any device memory allocator needs to be made. +It must be destroyed before `VkDevice` is destroyed. + +At program startup: + +-# Initialize Vulkan to have `VkInstance`, `VkPhysicalDevice`, `VkDevice` object. +-# Fill VmaAllocatorCreateInfo structure and call vmaCreateAllocator() to create #VmaAllocator object. + +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. +Otherwise, VMA would use only features of Vulkan 1.0 core with no extensions. +See below for details. + +\subsection quick_start_initialization_selecting_vulkan_version Selecting Vulkan version + +VMA supports Vulkan version down to 1.0, for backward compatibility. +If you want to use higher version, you need to inform the library about it. +This is a two-step process. + +Step 1: Compile time. By default, VMA compiles with code supporting the highest +Vulkan version found in the included `` that is also supported by the library. +If this is OK, you don't need to do anything. +However, if you want to compile VMA as if only some lower Vulkan version was available, +define macro `VMA_VULKAN_VERSION` before every `#include "vk_mem_alloc.h"`. +It should have decimal numeric value in form of ABBBCCC, where A = major, BBB = minor, CCC = patch Vulkan version. +For example, to compile against Vulkan 1.2: + +\code +#define VMA_VULKAN_VERSION 1002000 // Vulkan 1.2 +#include "vk_mem_alloc.h" +\endcode + +Step 2: Runtime. Even when compiled with higher Vulkan version available, +VMA can use only features of a lower version, which is configurable during creation of the #VmaAllocator object. +By default, only Vulkan 1.0 is used. +To initialize the allocator with support for higher Vulkan version, you need to set member +VmaAllocatorCreateInfo::vulkanApiVersion to an appropriate value, e.g. using constants like `VK_API_VERSION_1_2`. +See code sample below. + +\subsection quick_start_initialization_importing_vulkan_functions Importing Vulkan functions + +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. + +\subsection quick_start_initialization_enabling_extensions Enabling extensions + +VMA can automatically use following Vulkan extensions. +If you found them available on the selected physical device and you enabled them +while creating `VkInstance` / `VkDevice` object, inform VMA about their availability +by setting appropriate flags in VmaAllocatorCreateInfo::flags. + +Vulkan extension | VMA flag +------------------------------|----------------------------------------------------- +VK_KHR_dedicated_allocation | #VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT +VK_KHR_bind_memory2 | #VMA_ALLOCATOR_CREATE_KHR_BIND_MEMORY2_BIT +VK_KHR_maintenance4 | #VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE4_BIT +VK_KHR_maintenance5 | #VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT +VK_EXT_memory_budget | #VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT +VK_KHR_buffer_device_address | #VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT +VK_EXT_memory_priority | #VMA_ALLOCATOR_CREATE_EXT_MEMORY_PRIORITY_BIT +VK_AMD_device_coherent_memory | #VMA_ALLOCATOR_CREATE_AMD_DEVICE_COHERENT_MEMORY_BIT +VK_KHR_external_memory_win32 | #VMA_ALLOCATOR_CREATE_KHR_EXTERNAL_MEMORY_WIN32_BIT + +Example with fetching pointers to Vulkan functions dynamically: + +\code +#define VMA_STATIC_VULKAN_FUNCTIONS 0 +#define VMA_DYNAMIC_VULKAN_FUNCTIONS 1 +#include "vk_mem_alloc.h" + +... + +VmaVulkanFunctions vulkanFunctions = {}; +vulkanFunctions.vkGetInstanceProcAddr = &vkGetInstanceProcAddr; +vulkanFunctions.vkGetDeviceProcAddr = &vkGetDeviceProcAddr; + +VmaAllocatorCreateInfo allocatorCreateInfo = {}; +allocatorCreateInfo.flags = VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT; +allocatorCreateInfo.vulkanApiVersion = VK_API_VERSION_1_2; +allocatorCreateInfo.physicalDevice = physicalDevice; +allocatorCreateInfo.device = device; +allocatorCreateInfo.instance = instance; +allocatorCreateInfo.pVulkanFunctions = &vulkanFunctions; + +VmaAllocator allocator; +vmaCreateAllocator(&allocatorCreateInfo, &allocator); + +// Entire program... + +// At the end, don't forget to: +vmaDestroyAllocator(allocator); +\endcode + + +\subsection quick_start_initialization_other_config Other configuration options + +There are additional configuration options available through preprocessor macros that you can define +before including VMA header and through parameters passed in #VmaAllocatorCreateInfo. +They include a possibility to use your own callbacks for host memory allocations (`VkAllocationCallbacks`), +callbacks for device memory allocations (instead of `vkAllocateMemory`, `vkFreeMemory`), +or your custom `VMA_ASSERT` macro, among others. +For more information, see: @ref configuration. + + +\section quick_start_resource_allocation Resource allocation + +When you want to create a buffer or image: + +-# Fill `VkBufferCreateInfo` / `VkImageCreateInfo` structure. +-# Fill VmaAllocationCreateInfo structure. +-# Call vmaCreateBuffer() / vmaCreateImage() to get `VkBuffer`/`VkImage` with memory + already allocated and bound to it, plus #VmaAllocation objects that represents its underlying memory. + +\code +VkBufferCreateInfo bufferInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufferInfo.size = 65536; +bufferInfo.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.usage = VMA_MEMORY_USAGE_AUTO; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +Don't forget to destroy your buffer and allocation objects when no longer needed: + +\code +vmaDestroyBuffer(allocator, buffer, allocation); +\endcode + +If you need to map the buffer, you must set flag +#VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or #VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT +in VmaAllocationCreateInfo::flags. +There are many additional parameters that can control the choice of memory type to be used for the allocation +and other features. +For more information, see documentation chapters: @ref choosing_memory_type, @ref memory_mapping. + + +\page choosing_memory_type Choosing memory type + +Physical devices in Vulkan support various combinations of memory heaps and +types. Help with choosing correct and optimal memory type for your specific +resource is one of the key features of this library. You can use it by filling +appropriate members of VmaAllocationCreateInfo structure, as described below. +You can also combine multiple methods. + +-# If you just want to find memory type index that meets your requirements, you + can use function: 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. + vmaAllocateMemoryPages() function is also provided for creating multiple allocations at once, + which may be useful for sparse binding. +-# If you already have a buffer or an image created, you want to allocate memory + for it and then you will bind it yourself, you can use function + vmaAllocateMemoryForBuffer(), vmaAllocateMemoryForImage(). + For binding you should use functions: vmaBindBufferMemory(), vmaBindImageMemory() + or their extended versions: vmaBindBufferMemory2(), vmaBindImageMemory2(). +-# 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! + +When using 3. or 4., the library internally queries Vulkan for memory types +supported for that buffer or image (function `vkGetBufferMemoryRequirements()`) +and uses only one of these types. + +If no memory type can be found that meets all the requirements, these functions +return `VK_ERROR_FEATURE_NOT_PRESENT`. + +You can leave VmaAllocationCreateInfo structure completely filled with zeros. +It means no requirements are specified for memory type. +It is valid, although not very useful. + +\section choosing_memory_type_usage Usage + +The easiest way to specify memory requirements is to fill member +VmaAllocationCreateInfo::usage using one of the values of enum #VmaMemoryUsage. +It defines high level, common usage types. +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 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 }; +bufferInfo.size = 65536; +bufferInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocInfo = {}; +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 described previously can be created like this. +It will likely end 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. +See also: @ref memory_mapping. + +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 described 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 deprecated. + +\section choosing_memory_type_required_preferred_flags Required and preferred flags + +You can specify more detailed requirements by filling members +VmaAllocationCreateInfo::requiredFlags and VmaAllocationCreateInfo::preferredFlags +with a combination of bits from enum `VkMemoryPropertyFlags`. For example, +if you want to create a buffer that will be persistently mapped on host (so it +must be `HOST_VISIBLE`) and preferably will also be `HOST_COHERENT` and `HOST_CACHED`, +use following code: + +\code +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.requiredFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT; +allocInfo.preferredFlags = VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; +allocInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT | VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +A memory type is chosen that has all the required flags and as many preferred +flags set as possible. + +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 + +If you inspected memory types available on the physical device and you have +a preference for memory types that you want to use, you can fill member +VmaAllocationCreateInfo::memoryTypeBits. It is a bit mask, where each bit set +means that a memory type with that index is allowed to be used for the +allocation. Special value 0, just like `UINT32_MAX`, means there are no +restrictions to memory type index. + +Please note that this member is NOT just a memory type index. +Still you can use it to choose just one, specific memory type. +For example, if you already determined that your buffer should be created in +memory type 2, use following code: + +\code +uint32_t memoryTypeIndex = 2; + +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.memoryTypeBits = 1u << memoryTypeIndex; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufferInfo, &allocInfo, &buffer, &allocation, nullptr); +\endcode + +You can also use this parameter to exclude some memory types. +If you inspect memory heaps and types available on the current physical device and +you determine that for some reason you don't want to use a specific memory type for the allocation, +you can enable automatic memory type selection but exclude certain memory type or types +by setting all bits of `memoryTypeBits` to 1 except the ones you choose. + +\code +// ... +uint32_t excludedMemoryTypeIndex = 2; +VmaAllocationCreateInfo allocInfo = {}; +allocInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocInfo.memoryTypeBits = ~(1u << excludedMemoryTypeIndex); +// ... +\endcode + + +\section choosing_memory_type_custom_memory_pools Custom memory pools + +If you allocate from custom memory pool, all the ways of specifying memory +requirements described above are not applicable and the aforementioned members +of VmaAllocationCreateInfo structure are ignored. Memory type is selected +explicitly when creating the pool and then used to make all the allocations from +that pool. For further details, see \ref custom_memory_pools. + +\section choosing_memory_type_dedicated_allocations Dedicated allocations + +Memory for allocations is reserved out of larger block of `VkDeviceMemory` +allocated from Vulkan internally. That is the main feature of this whole library. +You can still request a separate memory block to be created for an allocation, +just like you would do in a trivial solution without using any allocator. +In that case, a buffer or image is always bound to that memory at offset 0. +This is called a "dedicated allocation". +You can explicitly request it by using flag #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +The library can also internally decide to use dedicated allocation in some cases, e.g.: + +- When the size of the allocation is large. +- When [VK_KHR_dedicated_allocation](@ref vk_khr_dedicated_allocation) extension is enabled + and it reports that dedicated allocation is required or recommended for the resource. +- When allocation of next big memory block fails due to not enough device memory, + but allocation with the exact requested size succeeds. + + +\page memory_mapping Memory mapping + +To "map memory" in Vulkan means to obtain a CPU pointer to `VkDeviceMemory`, +to be able to read from it or write to it in CPU code. +Mapping is possible only of memory allocated from a memory type that has +`VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT` flag. +Functions `vkMapMemory()`, `vkUnmapMemory()` are designed for this purpose. +You can use them directly with memory allocated by this library, +but it is not recommended because of following issue: +Mapping the same `VkDeviceMemory` block multiple times is illegal - only one mapping at a time is allowed. +This includes mapping disjoint regions. Mapping is not reference-counted internally by Vulkan. +It is also not thread-safe. +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 these flags can still be used for consistency. + +\section memory_mapping_copy_functions Copy functions + +The easiest way to copy data from a host pointer to an allocation is to use convenience function vmaCopyMemoryToAllocation(). +It automatically maps the Vulkan memory temporarily (if not already mapped), performs `memcpy`, +and calls `vkFlushMappedMemoryRanges` (if required - if memory type is not `HOST_COHERENT`). + +It is also the safest one, because using `memcpy` avoids a risk of accidentally introducing memory reads +(e.g. by doing `pMappedVectors[i] += v`), which may be very slow on memory types that are not `HOST_CACHED`. + +\code +struct ConstantBuffer +{ + ... +}; +ConstantBuffer constantBufferData = ... + +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = sizeof(ConstantBuffer); +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT; + +VkBuffer buf; +VmaAllocation alloc; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, nullptr); + +vmaCopyMemoryToAllocation(allocator, &constantBufferData, alloc, 0, sizeof(ConstantBuffer)); +\endcode + +Copy in the other direction - from an allocation to a host pointer can be performed the same way using function vmaCopyAllocationToMemory(). + +\section memory_mapping_mapping_functions Mapping functions + +The library provides following functions for mapping of a specific allocation: vmaMapMemory(), vmaUnmapMemory(). +They are safer and more convenient to use than standard Vulkan functions. +You can map an allocation multiple times simultaneously - mapping is reference-counted internally. +You can also map different allocations simultaneously regardless of whether they use the same `VkDeviceMemory` block. +The way it is implemented is that the library always maps entire memory block, not just region of the allocation. +For further details, see description of vmaMapMemory() function. +Example: + +\code +// Having these objects initialized: +struct ConstantBuffer +{ + ... +}; +ConstantBuffer constantBufferData = ... + +VmaAllocator allocator = ... +VkBuffer constantBuffer = ... +VmaAllocation constantBufferAllocation = ... + +// You can map and fill your buffer using following code: + +void* mappedData; +vmaMapMemory(allocator, constantBufferAllocation, &mappedData); +memcpy(mappedData, &constantBufferData, sizeof(constantBufferData)); +vmaUnmapMemory(allocator, constantBufferAllocation); +\endcode + +When mapping, you may see a warning from Vulkan validation layer similar to this one: + +Mapping an image with layout VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL can result in undefined behavior if this memory is used by the device. Only GENERAL or PREINITIALIZED should be used. + +It happens because the library maps entire `VkDeviceMemory` block, where different +types of images and buffers may end up together, especially on GPUs with unified memory like Intel. +You can safely ignore it if you are sure you access only memory of the intended +object that you wanted to map. + + +\section memory_mapping_persistently_mapped_memory Persistently mapped memory + +Keeping your memory persistently mapped is generally OK in Vulkan. +You don't need to unmap it before using its data on the GPU. +The library provides a special feature designed for that: +Allocations made with #VMA_ALLOCATION_CREATE_MAPPED_BIT flag set in +VmaAllocationCreateInfo::flags stay mapped all the time, +so you can just access CPU pointer to it any time +without a need to call any "map" or "unmap" function. +Example: + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = sizeof(ConstantBuffer); +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +// Buffer is already mapped. You can access its memory. +memcpy(allocInfo.pMappedData, &constantBufferData, sizeof(constantBufferData)); +\endcode + +\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 + +Memory in Vulkan doesn't need to be unmapped before using it on GPU, +but unless a memory types has `VK_MEMORY_PROPERTY_HOST_COHERENT_BIT` flag set, +you need to manually **invalidate** cache before reading of mapped pointer +and **flush** cache after writing to mapped pointer. +Map/unmap operations don't do that automatically. +Vulkan provides following functions for this purpose `vkFlushMappedMemoryRanges()`, +`vkInvalidateMappedMemoryRanges()`, but this library provides more convenient +functions that refer to given allocation object: vmaFlushAllocation(), +vmaInvalidateAllocation(), +or multiple objects at once: vmaFlushAllocations(), vmaInvalidateAllocations(). + +Regions of memory specified for flush/invalidate must be aligned to +`VkPhysicalDeviceLimits::nonCoherentAtomSize`. This is automatically ensured by the library. +In any memory type that is `HOST_VISIBLE` but not `HOST_COHERENT`, all allocations +within blocks are aligned to this value, so their offsets are always multiply of +`nonCoherentAtomSize` and two different allocations never share same "line" of this size. + +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 PC you may not need to bother. + + +\page staying_within_budget Staying within budget + +When developing a graphics-intensive game or program, it is important to avoid allocating +more GPU memory than it is physically available. When the memory is over-committed, +various bad things can happen, depending on the specific GPU, graphics driver, and +operating system: + +- It may just work without any problems. +- The application may slow down because some memory blocks are moved to system RAM + and the GPU has to access them through PCI Express bus. +- A new allocation may take very long time to complete, even few seconds, and possibly + freeze entire system. +- The new allocation may fail with `VK_ERROR_OUT_OF_DEVICE_MEMORY`. +- It may even result in GPU crash (TDR), observed as `VK_ERROR_DEVICE_LOST` + returned somewhere later. + +\section staying_within_budget_querying_for_budget Querying for budget + +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 +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 +about the budget from Vulkan device. VMA is able to use this extension automatically. +When not enabled, the allocator behaves same way, but then it estimates current usage +and available budget based on its internal information and Vulkan memory heap sizes, +which may be less precise. In order to use this extension: + +1. Make sure extensions VK_EXT_memory_budget and VK_KHR_get_physical_device_properties2 + required by it are available and enable them. Please note that the first is a device + extension and the second is instance extension! +2. Use flag #VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT when creating #VmaAllocator object. +3. Make sure to call vmaSetCurrentFrameIndex() every frame. Budget is queried from + Vulkan inside of it to avoid overhead of querying it with every allocation. + +\section staying_within_budget_controlling_memory_usage Controlling memory usage + +There are many ways in which you can try to stay within the budget. + +First, when making new allocation requires allocating a new memory block, the library +tries not to exceed the budget automatically. If a block with default recommended size +(e.g. 256 MB) would go over budget, a smaller block is allocated, possibly even +dedicated memory for just this resource. + +If the size of the requested resource plus current memory usage is more than the +budget, by default the library still tries to create it, leaving it to the Vulkan +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. +VMA then tries to make the allocation from the next eligible Vulkan memory type. +If 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. + +\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. + + +\page resource_aliasing Resource aliasing (overlap) + +New explicit graphics APIs (Vulkan and Direct3D 12), thanks to manual memory +management, give an opportunity to alias (overlap) multiple resources in the +same region of memory - a feature not available in the old APIs (Direct3D 11, OpenGL). +It can be useful to save video memory, but it must be used with caution. + +For example, if you know the flow of your whole render frame in advance, you +are going to use some intermediate textures or buffers only during a small range of render passes, +and you know these ranges don't overlap in time, you can bind these resources to +the same place in memory, even if they have completely different parameters (width, height, format etc.). + +![Resource aliasing (overlap)](../gfx/Aliasing.png) + +Such scenario is possible using VMA, but you need to create your images manually. +Then you need to calculate parameters of an allocation to be made using formula: + +- allocation size = max(size of each image) +- allocation alignment = max(alignment of each image) +- allocation memoryTypeBits = bitwise AND(memoryTypeBits of each image) + +Following example shows two different images bound to the same place in memory, +allocated to fit largest of them. + +\code +// A 512x512 texture to be sampled. +VkImageCreateInfo img1CreateInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO }; +img1CreateInfo.imageType = VK_IMAGE_TYPE_2D; +img1CreateInfo.extent.width = 512; +img1CreateInfo.extent.height = 512; +img1CreateInfo.extent.depth = 1; +img1CreateInfo.mipLevels = 10; +img1CreateInfo.arrayLayers = 1; +img1CreateInfo.format = VK_FORMAT_R8G8B8A8_SRGB; +img1CreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL; +img1CreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; +img1CreateInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT; +img1CreateInfo.samples = VK_SAMPLE_COUNT_1_BIT; + +// A full screen texture to be used as color attachment. +VkImageCreateInfo img2CreateInfo = { VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO }; +img2CreateInfo.imageType = VK_IMAGE_TYPE_2D; +img2CreateInfo.extent.width = 1920; +img2CreateInfo.extent.height = 1080; +img2CreateInfo.extent.depth = 1; +img2CreateInfo.mipLevels = 1; +img2CreateInfo.arrayLayers = 1; +img2CreateInfo.format = VK_FORMAT_R8G8B8A8_UNORM; +img2CreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL; +img2CreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; +img2CreateInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; +img2CreateInfo.samples = VK_SAMPLE_COUNT_1_BIT; + +VkImage img1; +res = vkCreateImage(device, &img1CreateInfo, nullptr, &img1); +VkImage img2; +res = vkCreateImage(device, &img2CreateInfo, nullptr, &img2); + +VkMemoryRequirements img1MemReq; +vkGetImageMemoryRequirements(device, img1, &img1MemReq); +VkMemoryRequirements img2MemReq; +vkGetImageMemoryRequirements(device, img2, &img2MemReq); + +VkMemoryRequirements finalMemReq = {}; +finalMemReq.size = std::max(img1MemReq.size, img2MemReq.size); +finalMemReq.alignment = std::max(img1MemReq.alignment, img2MemReq.alignment); +finalMemReq.memoryTypeBits = img1MemReq.memoryTypeBits & img2MemReq.memoryTypeBits; +// Validate if(finalMemReq.memoryTypeBits != 0) + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + +VmaAllocation alloc; +res = vmaAllocateMemory(allocator, &finalMemReq, &allocCreateInfo, &alloc, nullptr); + +res = vmaBindImageMemory(allocator, alloc, img1); +res = vmaBindImageMemory(allocator, alloc, img2); + +// You can use img1, img2 here, but not at the same time! + +vmaFreeMemory(allocator, alloc); +vkDestroyImage(allocator, img2, nullptr); +vkDestroyImage(allocator, img1, nullptr); +\endcode + +VMA also provides convenience functions that create a buffer or image and bind it to memory +represented by an existing #VmaAllocation: +vmaCreateAliasingBuffer(), vmaCreateAliasingBuffer2(), +vmaCreateAliasingImage(), vmaCreateAliasingImage2(). +Versions with "2" offer additional parameter `allocationLocalOffset`. + +Remember that using resources that alias in memory requires proper synchronization. +You need to issue a memory barrier to make sure commands that use `img1` and `img2` +don't overlap on GPU timeline. +You also need to treat a resource after aliasing as uninitialized - containing garbage data. +For example, if you use `img1` and then want to use `img2`, you need to issue +an image memory barrier for `img2` with `oldLayout` = `VK_IMAGE_LAYOUT_UNDEFINED`. + +Additional considerations: + +- Vulkan also allows to interpret contents of memory between aliasing resources consistently in some cases. +See chapter 11.8. "Memory Aliasing" of Vulkan specification or `VK_IMAGE_CREATE_ALIAS_BIT` flag. +- 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 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. +Some GPUs may expose multiple memory types suitable e.g. only for buffers or +images with `COLOR_ATTACHMENT` usage, so the sets of memory types supported by your +resources may be disjoint. Aliasing them is not possible in that case. + + +\page custom_memory_pools Custom memory pools + +A memory pool contains a number of `VkDeviceMemory` blocks. +The library automatically creates and manages default pool for each memory type available on the device. +Default memory pool automatically grows in size. +Size of allocated blocks is also variable and managed automatically. +You are using default pools whenever you leave VmaAllocationCreateInfo::pool = null. + +You can create custom pool and allocate memory out of it. +It can be useful if you want to: + +- Keep certain kind of allocations separate from others. +- Enforce particular, fixed size of Vulkan memory blocks. +- Limit maximum amount of Vulkan memory allocated for that pool. +- Reserve minimum or fixed amount of Vulkan memory always preallocated for that pool. +- 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: + +-# Fill VmaPoolCreateInfo structure. +-# Call vmaCreatePool() to obtain #VmaPool handle. +-# When making an allocation, set VmaAllocationCreateInfo::pool to this handle. + You don't need to specify any other parameters of this structure, like `usage`. + +Example: + +\code +// 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 = memTypeIndex; +poolCreateInfo.blockSize = 128ull * 1024 * 1024; +poolCreateInfo.maxBlockCount = 2; + +VmaPool pool; +res = vmaCreatePool(allocator, &poolCreateInfo, &pool); +// Check res... + +// Allocate a buffer out of it. +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 1024; +bufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.pool = pool; + +VkBuffer buf; +VmaAllocation alloc; +res = vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, nullptr); +// Check res... +\endcode + +You have to free all allocations made from this pool before destroying it. + +\code +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. + + +\section custom_memory_pools_MemTypeIndex Choosing memory type index + +When creating a pool, you must explicitly specify memory type index. +To find the one suitable for your buffers or images, you can use helper functions +vmaFindMemoryTypeIndexForBufferInfo(), vmaFindMemoryTypeIndexForImageInfo(). +You need to provide structures with example parameters of buffers or images +that you are going to create in that pool. + +\code +VkBufferCreateInfo exampleBufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +exampleBufCreateInfo.size = 1024; // Doesn't matter +exampleBufCreateInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; + +uint32_t memTypeIndex; +vmaFindMemoryTypeIndexForBufferInfo(allocator, &exampleBufCreateInfo, &allocCreateInfo, &memTypeIndex); + +VmaPoolCreateInfo poolCreateInfo = {}; +poolCreateInfo.memoryTypeIndex = memTypeIndex; +// ... +\endcode + +When creating buffers/images allocated in that pool, provide following parameters: + +- `VkBufferCreateInfo`: Prefer to pass same parameters as above. + Otherwise you risk creating resources in a memory type that is not suitable for them, which may result in undefined behavior. + Using different `VK_BUFFER_USAGE_` flags may work, but you shouldn't create images in a pool intended for buffers + or the other way around. +- VmaAllocationCreateInfo: You don't need to pass same parameters. Fill only `pool` member. + Other members are ignored anyway. + + +\section custom_memory_pools_when_not_use When not to use custom pools + +Custom pools are commonly overused by VMA users. +While it may feel natural to keep some logical groups of resources separate in memory, +in most cases it does more harm than good. +Using custom pool shouldn't be your first choice. +Instead, please make all allocations from default pools first and only use custom pools +if you can prove and measure that it is beneficial in some way, +e.g. it results in lower memory usage, better performance, etc. + +Using custom pools has disadvantages: + +- Each pool has its own collection of `VkDeviceMemory` blocks. + Some of them may be partially or even completely empty. + Spreading allocations across multiple pools increases the amount of wasted (allocated but unbound) memory. +- You must manually choose specific memory type to be used by a custom pool (set as VmaPoolCreateInfo::memoryTypeIndex). + When using default pools, best memory type for each of your allocations can be selected automatically + using a carefully design algorithm that works across all kinds of GPUs. +- If an allocation from a custom pool at specific memory type fails, entire allocation operation returns failure. + When using default pools, VMA tries another compatible memory type. +- If you set VmaPoolCreateInfo::blockSize != 0, each memory block has the same size, + while default pools start from small blocks and only allocate next blocks larger and larger + up to the preferred block size. + +Many of the common concerns can be addressed in a different way than using custom pools: + +- If you want to keep your allocations of certain size (small versus large) or certain lifetime (transient versus long lived) + separate, you likely don't need to. + VMA uses a high quality allocation algorithm that manages memory well in various cases. + Please measure and check if using custom pools provides a benefit. +- If you want to keep your images and buffers separate, you don't need to. + VMA respects `bufferImageGranularity` limit automatically. +- If you want to keep your mapped and not mapped allocations separate, you don't need to. + VMA respects `nonCoherentAtomSize` limit automatically. + It also maps only those `VkDeviceMemory` blocks that need to map any allocation. + It even tries to keep mappable and non-mappable allocations in separate blocks to minimize the amount of mapped memory. +- If you want to choose a custom size for the default memory block, you can set it globally instead + using VmaAllocatorCreateInfo::preferredLargeHeapBlockSize. +- If you want to select specific memory type for your allocation, + you can set VmaAllocationCreateInfo::memoryTypeBits to `(1u << myMemoryTypeIndex)` instead. +- If you need to create a buffer with certain minimum alignment, you can still do it + using default pools with dedicated function vmaCreateBufferWithAlignment(). + + +\section linear_algorithm Linear allocation algorithm + +Each Vulkan memory block managed by this library has accompanying metadata that +keeps track of used and unused regions. By default, the metadata structure and +algorithm tries to find best place for new allocations among free regions to +optimize memory usage. This way you can allocate and free objects in any order. + +![Default allocation algorithm](../gfx/Linear_allocator_1_algo_default.png) + +Sometimes there is a need to use simpler, linear allocation algorithm. You can +create custom pool that uses such algorithm by adding flag +#VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT to VmaPoolCreateInfo::flags while creating +#VmaPool object. Then an alternative metadata management is used. It always +creates new allocations after last one and doesn't reuse free regions after +allocations freed in the middle. It results in better allocation performance and +less memory consumed by metadata. + +![Linear allocation algorithm](../gfx/Linear_allocator_2_algo_linear.png) + +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 + +In a pool that uses linear algorithm, you still need to free all the allocations +individually, e.g. by using vmaFreeMemory() or vmaDestroyBuffer(). You can free +them in any order. New allocations are always made after last one - free space +in the middle is not reused. However, when you release all the allocation and +the pool becomes empty, allocation starts from the beginning again. This way you +can use linear algorithm to speed up creation of allocations that you are going +to release all at once. + +![Free-at-once](../gfx/Linear_allocator_3_free_at_once.png) + +This mode is also available for pools created with VmaPoolCreateInfo::maxBlockCount +value that allows multiple memory blocks. + +\subsection linear_algorithm_stack Stack + +When you free an allocation that was created last, its space can be reused. +Thanks to this, if you always release allocations in the order opposite to their +creation (LIFO - Last In First Out), you can achieve behavior of a stack. + +![Stack](../gfx/Linear_allocator_4_stack.png) + +This mode is also available for pools created with VmaPoolCreateInfo::maxBlockCount +value that allows multiple memory blocks. + +\subsection linear_algorithm_double_stack Double stack + +The space reserved by a custom pool with linear algorithm may be used by two +stacks: + +- First, default one, growing up from offset 0. +- Second, "upper" one, growing down from the end towards lower offsets. + +To make allocation from the upper stack, add flag #VMA_ALLOCATION_CREATE_UPPER_ADDRESS_BIT +to VmaAllocationCreateInfo::flags. + +![Double stack](../gfx/Linear_allocator_7_double_stack.png) + +Double stack is available only in pools with one memory block - +VmaPoolCreateInfo::maxBlockCount must be 1. Otherwise behavior is undefined. + +When the two stacks' ends meet so there is not enough space between them for a +new allocation, such allocation fails with usual +`VK_ERROR_OUT_OF_DEVICE_MEMORY` error. + +\subsection linear_algorithm_ring_buffer Ring buffer + +When you free some allocations from the beginning and there is not enough free space +for a new one at the end of a pool, allocator's "cursor" wraps around to the +beginning and starts allocation there. Thanks to this, if you always release +allocations in the same order as you created them (FIFO - First In First Out), +you can achieve behavior of a ring buffer / queue. + +![Ring buffer](../gfx/Linear_allocator_5_ring_buffer.png) + +Ring buffer is available only in pools with one memory block - +VmaPoolCreateInfo::maxBlockCount must be 1. Otherwise behavior is undefined. + +\note \ref defragmentation is not supported in custom pools created with #VMA_POOL_CREATE_LINEAR_ALGORITHM_BIT. + + +\page defragmentation Defragmentation + +Interleaved allocations and deallocations of many objects of varying size can +cause fragmentation over time, which can lead to a situation where the library is unable +to find a continuous range of free memory for a new allocation despite there is +enough free space, just scattered across many small free ranges between existing +allocations. + +To mitigate this problem, you can use defragmentation feature. +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 +VmaDefragmentationInfo defragInfo = {}; +defragInfo.pool = myPool; +defragInfo.flags = VMA_DEFRAGMENTATION_FLAG_ALGORITHM_FAST_BIT; + +VmaDefragmentationContext defragCtx; +VkResult res = vmaBeginDefragmentation(allocator, &defragInfo, &defragCtx); +// Check res... + +for(;;) +{ + 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) + { + // Inspect pass.pMoves[i].srcAllocation, identify what buffer/image it represents. + VmaAllocationInfo allocInfo; + vmaGetAllocationInfo(allocator, pass.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, pass.pMoves[i].dstTmpAllocation, newImg); + // Check res... + + // Issue a vkCmdCopyBuffer/vkCmdCopyImage to copy its content to the new place. + vkCmdCopyImage(cmdBuf, resData->img, ..., newImg, ...); + } + + // Make sure the copy commands finished executing. + vkWaitForFences(...); + + // Destroy old buffers/images bound with pass.pMoves[i].srcAllocation. + for(uint32_t i = 0; i < pass.moveCount; ++i) + { + // ... + vkDestroyImage(device, resData->img, 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 + +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: + +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. + +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. + +Inside a pass, for each allocation that should be moved: + +- 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 several functions that return information about its internal state, +especially the amount of memory allocated from Vulkan. + +\section statistics_numeric_statistics Numeric statistics + +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 within budget +(see also \ref staying_within_budget). +Example: + +\code +uint32_t heapIndex = ... + +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 + +You can dump internal state of the allocator to a string in JSON format using function vmaBuildStatsString(). +The result is guaranteed to be correct JSON. +It uses ANSI encoding. +Any strings provided by user (see [Allocation names](@ref allocation_names)) +are copied as-is and properly escaped for JSON, so if they use UTF-8, ISO-8859-2 or any other encoding, +this JSON string can be treated as using this encoding. +It must be freed using function vmaFreeStatsString(). + +The format of this JSON string is not part of official documentation of the library, +but it will not change in backward-incompatible way without increasing library major version number +and appropriate mention in changelog. + +The JSON string contains all the data that can be obtained using 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. + + +\page allocation_annotation Allocation names and user data + +\section allocation_user_data Allocation user data + +You can annotate allocations with your own information, e.g. for debugging purposes. +To do that, fill VmaAllocationCreateInfo::pUserData field when creating +an allocation. It 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 is useful to identify appropriate data structures in your engine given #VmaAllocation, +e.g. when doing \ref defragmentation. + +\code +VkBufferCreateInfo bufCreateInfo = ... + +MyBufferMetadata* pMetadata = CreateBufferMetadata(); + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.pUserData = pMetadata; + +VkBuffer buffer; +VmaAllocation allocation; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buffer, &allocation, nullptr); +\endcode + +The pointer may be later retrieved as VmaAllocationInfo::pUserData: + +\code +VmaAllocationInfo allocInfo; +vmaGetAllocationInfo(allocator, allocation, &allocInfo); +MyBufferMetadata* pMetadata = (MyBufferMetadata*)allocInfo.pUserData; +\endcode + +It can also be changed using function vmaSetAllocationUserData(). + +Values of (non-zero) allocations' `pUserData` are printed in JSON report created by +vmaBuildStatsString() in hexadecimal form. + +\section allocation_names Allocation names + +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 +std::string imageName = "Texture: "; +imageName += fileName; +vmaSetAllocationName(allocator, allocation, imageName.c_str()); +\endcode + +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 +VmaVirtualBlockCreateInfo blockCreateInfo = {}; +blockCreateInfo.size = 1048576; // 1 MB + +VmaVirtualBlock block; +VkResult res = vmaCreateVirtualBlock(&blockCreateInfo, &block); +\endcode + +\section virtual_allocator_making_virtual_allocations Making virtual allocations + +#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 within 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 + +If you suspect a bug with memory usage, like usage of uninitialized memory or +memory being overwritten out of bounds of an allocation, +you can use debug features of this library to verify this. + +\section debugging_memory_usage_initialization Memory initialization + +If you experience a bug with incorrect and nondeterministic data in your program and you suspect uninitialized memory to be used, +you can enable automatic memory initialization to verify this. +To do it, define macro `VMA_DEBUG_INITIALIZE_ALLOCATIONS` to 1. + +\code +#define VMA_DEBUG_INITIALIZE_ALLOCATIONS 1 +#include "vk_mem_alloc.h" +\endcode + +It makes memory of 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` and with allocations that can be mapped. +It works also with dedicated allocations. + +\section debugging_memory_usage_margins Margins + +By default, allocations are laid out in memory blocks next to each other if possible +(considering required alignment, `bufferImageGranularity`, and `nonCoherentAtomSize`). + +![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 after every allocation. + +\code +#define VMA_DEBUG_MARGIN 16 +#include "vk_mem_alloc.h" +\endcode + +![Allocations with margin](../gfx/Margins_2.png) + +If your bug goes away after enabling margins, it means it may be caused by memory +being overwritten outside of allocation boundaries. It is not 100% certain though. +Change in application behavior may also be caused by different order and distribution +of allocations across memory blocks after margins are applied. + +Margins work with all types of memory. + +Margin is applied only to allocations made out of memory blocks and not to dedicated +allocations, which have their own memory block of specific size. +It is thus not applied to allocations made using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT flag +or those automatically decided to put into dedicated allocations, e.g. due to its +large size or recommended by VK_KHR_dedicated_allocation extension. + +Margins 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 +of contents of the margins. + +\code +#define VMA_DEBUG_MARGIN 16 +#define VMA_DEBUG_DETECT_CORRUPTION 1 +#include "vk_mem_alloc.h" +\endcode + +When this feature is enabled, number of bytes specified as `VMA_DEBUG_MARGIN` +(it must be multiply of 4) after every allocation is filled with a magic number. +This idea is also know as "canary". +Memory is automatically mapped and unmapped if necessary. + +This number is validated automatically when the allocation is destroyed. +If it is not equal to the expected value, `VMA_ASSERT()` is executed. +It clearly means that either CPU or GPU overwritten the memory outside of boundaries of the allocation, +which indicates a serious bug. + +You can also explicitly request checking margins of all allocations in all memory blocks +that belong to specified memory types by using function vmaCheckCorruption(), +or in memory blocks that belong to specified custom pool, by using function +vmaCheckPoolCorruption(). + +Margin validation (corruption detection) works only for memory types that are +`HOST_VISIBLE` and `HOST_COHERENT`. + + +\section debugging_memory_usage_leak_detection Leak detection features + +At allocation and allocator destruction time VMA checks for unfreed and unmapped blocks using +`VMA_ASSERT_LEAK()`. This macro defaults to an assertion, triggering a typically fatal error in Debug +builds, and doing nothing in Release builds. You can provide your own definition of `VMA_ASSERT_LEAK()` +to change this behavior. + +At memory block destruction time VMA lists out all unfreed allocations using the `VMA_LEAK_LOG_FORMAT()` +macro, which defaults to `VMA_DEBUG_LOG_FORMAT`, which in turn defaults to a no-op. +If you're having trouble with leaks - for example, the aforementioned assertion triggers, but you don't +quite know \em why -, overriding this macro to print out the the leaking blocks, combined with assigning +individual names to allocations using vmaSetAllocationName(), can greatly aid in fixing them. + +\page other_api_interop Interop with other graphics APIs + +VMA provides some features that help with interoperability with other graphics APIs, e.g. OpenGL. + +\section opengl_interop_exporting_memory Exporting memory + +If you want to attach `VkExportMemoryAllocateInfoKHR` or other structure to `pNext` chain of memory allocations made by the library: + +You can create \ref custom_memory_pools for such allocations. +Define and fill in your `VkExportMemoryAllocateInfoKHR` structure and attach it to VmaPoolCreateInfo::pMemoryAllocateNext +while creating the custom pool. +Please note that the structure must remain alive and unchanged for the whole lifetime of the #VmaPool, +not only while creating it, as no copy of the structure is made, +but its original pointer is used for each allocation instead. + +If you want to export all memory allocated by VMA from certain memory types, +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. +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`. + + +\section opengl_interop_custom_alignment Custom alignment + +Buffers or images exported to a different API like OpenGL may require a different alignment, +higher than the one used by the library automatically, queried from functions like `vkGetBufferMemoryRequirements`. +To impose such alignment: + +You can create \ref custom_memory_pools for such allocations. +Set VmaPoolCreateInfo::minAllocationAlignment member to the minimum alignment required for each allocation +to be made out of this pool. +The alignment actually used will be the maximum of this member and the alignment returned for the specific buffer or image +from a function like `vkGetBufferMemoryRequirements`, which is called by VMA automatically. + +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`. + +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. +You can ensure that an allocation is created as dedicated by using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +Contrary to Direct3D 12, Vulkan doesn't have a concept of alignment of the entire memory block passed on its allocation. + +\section opengl_interop_extended_allocation_information Extended allocation information + +If you want to rely on VMA to allocate your buffers and images inside larger memory blocks, +but you need to know the size of the entire block and whether the allocation was made +with its own dedicated memory, use function vmaGetAllocationInfo2() to retrieve +extended allocation information in structure #VmaAllocationInfo2. + + + +\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_gpu_only GPU-only resource + +When: +Any resources that you frequently write and read on GPU, +e.g. images used as color attachments (aka "render targets"), depth-stencil attachments, +images/buffers used as storage image/buffer (aka "Unordered Access View (UAV)"). + +What to do: +Let the library select the optimal memory type, which will likely have `VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT`. + +\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. + +\section usage_patterns_staging_copy_upload Staging copy for upload + +When: +A "staging" buffer than you want to map and fill from CPU code, then use as a source of transfer +to some GPU resource. + +What to do: +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`. + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 65536; +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +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; +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: +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: +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`. + +\code +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = 65536; +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; +allocCreateInfo.flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT | + VMA_ALLOCATION_CREATE_MAPPED_BIT; + +VkBuffer buf; +VmaAllocation alloc; +VmaAllocationInfo allocInfo; +vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); + +... + +const float* downloadedData = (const float*)allocInfo.pMappedData; +\endcode + + +\section usage_patterns_advanced_data_uploading Advanced data uploading + +For resources that you frequently write on CPU via mapped pointer and +frequently read on GPU e.g. as a uniform buffer (also called "dynamic"), multiple options are possible: + +-# 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 performance 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. + +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. + +\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; +VkResult result = vmaCreateBuffer(allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, &allocInfo); +// Check result... + +VkMemoryPropertyFlags memPropFlags; +vmaGetAllocationMemoryProperties(allocator, alloc, &memPropFlags); + +if(memPropFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) +{ + // Allocation ended up in a mappable memory and is already mapped - write to it directly. + + // [Executed in runtime]: + memcpy(allocInfo.pMappedData, myData, myDataSize); + result = vmaFlushAllocation(allocator, alloc, 0, VK_WHOLE_SIZE); + // Check result... + + VkBufferMemoryBarrier bufMemBarrier = { VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER }; + bufMemBarrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT; + bufMemBarrier.dstAccessMask = VK_ACCESS_UNIFORM_READ_BIT; + bufMemBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier.buffer = buf; + bufMemBarrier.offset = 0; + bufMemBarrier.size = VK_WHOLE_SIZE; + + vkCmdPipelineBarrier(cmdBuf, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT, + 0, 0, nullptr, 1, &bufMemBarrier, 0, nullptr); +} +else +{ + // Allocation ended up in a non-mappable memory - a transfer using a staging buffer is required. + 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; + result = vmaCreateBuffer(allocator, &stagingBufCreateInfo, &stagingAllocCreateInfo, + &stagingBuf, &stagingAlloc, &stagingAllocInfo); + // Check result... + + // [Executed in runtime]: + memcpy(stagingAllocInfo.pMappedData, myData, myDataSize); + result = vmaFlushAllocation(allocator, stagingAlloc, 0, VK_WHOLE_SIZE); + // Check result... + + VkBufferMemoryBarrier bufMemBarrier = { VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER }; + bufMemBarrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT; + bufMemBarrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT; + bufMemBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier.buffer = stagingBuf; + bufMemBarrier.offset = 0; + bufMemBarrier.size = VK_WHOLE_SIZE; + + vkCmdPipelineBarrier(cmdBuf, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, 0, nullptr, 1, &bufMemBarrier, 0, nullptr); + + VkBufferCopy bufCopy = { + 0, // srcOffset + 0, // dstOffset, + myDataSize, // size + }; + + vkCmdCopyBuffer(cmdBuf, stagingBuf, buf, 1, &bufCopy); + + VkBufferMemoryBarrier bufMemBarrier2 = { VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER }; + bufMemBarrier2.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; + bufMemBarrier2.dstAccessMask = VK_ACCESS_UNIFORM_READ_BIT; // We created a uniform buffer + bufMemBarrier2.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier2.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; + bufMemBarrier2.buffer = buf; + bufMemBarrier2.offset = 0; + bufMemBarrier2.size = VK_WHOLE_SIZE; + + vkCmdPipelineBarrier(cmdBuf, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT, + 0, 0, nullptr, 1, &bufMemBarrier2, 0, nullptr); +} +\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 + +Please check "CONFIGURATION SECTION" in the code to find macros that you can define +before each include of this file or change directly in this file to provide +your own implementation of basic facilities like assert, `min()` and `max()` functions, +mutex, atomic etc. + +For example, define `VMA_ASSERT(expr)` before including the library to provide +custom implementation of the assertion, compatible with your project. +By default it is defined to standard C `assert(expr)` in `_DEBUG` configuration +and empty otherwise. + +Similarly, you can define `VMA_LEAK_LOG_FORMAT` macro to enable printing of leaked (unfreed) allocations, +including their names and other parameters. Example: + +\code +#define VMA_LEAK_LOG_FORMAT(format, ...) do { \ + printf((format), __VA_ARGS__); \ + printf("\n"); \ + } while(false) +\endcode + +\section config_Vulkan_functions Pointers to Vulkan functions + +There are multiple ways to import pointers to Vulkan functions in the library. +In the simplest case you don't need to do anything. +If the compilation or linking of your program or the initialization of the #VmaAllocator +doesn't work for you, you can try to reconfigure it. + +First, the allocator tries to fetch pointers to Vulkan functions linked statically, +like this: + +\code +m_VulkanFunctions.vkAllocateMemory = (PFN_vkAllocateMemory)vkAllocateMemory; +\endcode + +If you want to disable this feature, set configuration macro: `#define VMA_STATIC_VULKAN_FUNCTIONS 0`. + +Second, you can provide the pointers yourself by setting member VmaAllocatorCreateInfo::pVulkanFunctions. +You can fetch them e.g. using functions `vkGetInstanceProcAddr` and `vkGetDeviceProcAddr` or +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 +Vulkan version and enabled extensions) are checked with `VMA_ASSERT` if they are not null. + + +\section custom_memory_allocator Custom host memory allocator + +If you use custom allocator for CPU memory rather than default operator `new` +and `delete` from C++, you can make this library using your allocator as well +by filling optional member VmaAllocatorCreateInfo::pAllocationCallbacks. These +functions will be passed to Vulkan, as well as used by the library itself to +make any CPU-side allocations. + +\section allocation_callbacks Device memory allocation callbacks + +The library makes calls to `vkAllocateMemory()` and `vkFreeMemory()` internally. +You can setup callbacks to be informed about these calls, e.g. for the purpose +of gathering some statistics. To do it, fill optional member +VmaAllocatorCreateInfo::pDeviceMemoryCallbacks. + +\section heap_memory_limit Device heap memory limit + +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_ +memory blocks from GPU VRAM to system RAM (which degrades performance). This +behavior is implementation-dependent - it depends on GPU vendor and graphics +driver. + +On AMD cards it can be controlled while creating Vulkan device object by using +VK_AMD_memory_overallocation_behavior extension, if available. + +Alternatively, if you want to test how your program behaves with limited amount of Vulkan device +memory available without switching your graphics card to one that really has +smaller VRAM, you can use a feature of this library intended for this purpose. +To do it, fill optional member VmaAllocatorCreateInfo::pHeapSizeLimit. + + + +\page vk_khr_dedicated_allocation VK_KHR_dedicated_allocation + +VK_KHR_dedicated_allocation is a Vulkan extension which can be used to improve +performance on some GPUs. It augments Vulkan API with possibility to query +driver whether it prefers particular buffer or image to have its own, dedicated +allocation (separate `VkDeviceMemory` block) for better efficiency - to be able +to do some internal optimizations. The extension is supported by this library. +It will be used automatically when enabled. + +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()`). +If yes, enable them (fill `VkDeviceCreateInfo::ppEnabledExtensionNames`). + +- VK_KHR_get_memory_requirements2 +- VK_KHR_dedicated_allocation + +If you enabled these extensions: + +2 . Use #VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT flag when creating +your #VmaAllocator to inform the library that you enabled required extensions +and you want the library to use them. + +\code +allocatorInfo.flags |= VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT; + +vmaCreateAllocator(&allocatorInfo, &allocator); +\endcode + +That is all. The extension will be automatically used whenever you create a +buffer using vmaCreateBuffer() or image using vmaCreateImage(). + +When using the extension together with Vulkan Validation Layer, you will receive +warnings like this: + +_vkBindBufferMemory(): Binding memory to buffer 0x33 but vkGetBufferMemoryRequirements() has not been called on that buffer._ + +It is OK, you should just ignore it. It happens because you use function +`vkGetBufferMemoryRequirements2KHR()` instead of standard +`vkGetBufferMemoryRequirements()`, while the validation layer seems to be +unaware of it. + +To learn more about this extension, see: + +- [VK_KHR_dedicated_allocation in Vulkan specification](https://www.khronos.org/registry/vulkan/specs/1.2-extensions/html/chap50.html#VK_KHR_dedicated_allocation) +- [VK_KHR_dedicated_allocation unofficial manual](http://asawicki.info/articles/VK_KHR_dedicated_allocation.php5) + + + +\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 parameters 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 +additional memory types with `VK_MEMORY_PROPERTY_DEVICE_COHERENT_BIT_AMD` and +`VK_MEMORY_PROPERTY_DEVICE_UNCACHED_BIT_AMD` flag. It is useful mostly for +allocation of buffers intended for writing "breadcrumb markers" in between passes +or draw calls, which in turn are useful for debugging GPU crash/hang/TDR cases. + +When the extension is available but has not been enabled, Vulkan physical device +still exposes those memory types, but their usage is forbidden. VMA automatically +takes care of that - it returns `VK_ERROR_FEATURE_NOT_PRESENT` when an attempt +to allocate memory of such type is made. + +If you want to use this extension in connection with VMA, follow these steps: + +\section vk_amd_device_coherent_memory_initialization Initialization + +1) Call `vkEnumerateDeviceExtensionProperties` for the physical device. +Check if the extension is supported - if returned array of `VkExtensionProperties` contains "VK_AMD_device_coherent_memory". + +2) Call `vkGetPhysicalDeviceFeatures2` for the physical device instead of old `vkGetPhysicalDeviceFeatures`. +Attach additional structure `VkPhysicalDeviceCoherentMemoryFeaturesAMD` to `VkPhysicalDeviceFeatures2::pNext` to be returned. +Check if the device feature is really supported - check if `VkPhysicalDeviceCoherentMemoryFeaturesAMD::deviceCoherentMemory` is true. + +3) While creating device with `vkCreateDevice`, enable this extension - add "VK_AMD_device_coherent_memory" +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 `VkPhysicalDeviceCoherentMemoryFeaturesAMD` to +`VkPhysicalDeviceFeatures2::pNext` and set its member `deviceCoherentMemory` to `VK_TRUE`. + +5) While creating #VmaAllocator with vmaCreateAllocator() inform VMA that you +have enabled this extension and feature - add #VMA_ALLOCATOR_CREATE_AMD_DEVICE_COHERENT_MEMORY_BIT +to VmaAllocatorCreateInfo::flags. + +\section vk_amd_device_coherent_memory_usage Usage + +After following steps described above, you can create VMA allocations and custom pools +out of the special `DEVICE_COHERENT` and `DEVICE_UNCACHED` memory types on eligible +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_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 + from this specific index by setting VmaAllocationCreateInfo::memoryTypeBits `= 1u << index`. + +\section vk_amd_device_coherent_memory_more_information More information + +To learn more about this extension, see [VK_AMD_device_coherent_memory in Vulkan specification](https://www.khronos.org/registry/vulkan/specs/1.2-extensions/man/html/VK_AMD_device_coherent_memory.html) + +Example use of this extension can be found in the code of the sample and test suite +accompanying this library. + + +\page vk_khr_external_memory_win32 VK_KHR_external_memory_win32 + +On Windows, the VK_KHR_external_memory_win32 device extension allows exporting a Win32 `HANDLE` +of a `VkDeviceMemory` block, to be able to reference the memory on other Vulkan logical devices or instances, +in multiple processes, and/or in multiple APIs. +VMA offers support for it. + +\section vk_khr_external_memory_win32_initialization Initialization + +1) Make sure the extension is defined in the code by including following header before including VMA: + +\code +#include +\endcode + +2) Check if "VK_KHR_external_memory_win32" is available among device extensions. +Enable it when creating the `VkDevice` object. + +3) Enable the usage of this extension in VMA by setting flag #VMA_ALLOCATOR_CREATE_KHR_EXTERNAL_MEMORY_WIN32_BIT +when calling vmaCreateAllocator(). + +4) Make sure that VMA has access to the `vkGetMemoryWin32HandleKHR` function by either enabling `VMA_DYNAMIC_VULKAN_FUNCTIONS` macro +or setting VmaVulkanFunctions::vkGetMemoryWin32HandleKHR explicitly. +For more information, see \ref quick_start_initialization_importing_vulkan_functions. + +\section vk_khr_external_memory_win32_preparations Preparations + +You can find example usage among tests, in file "Tests.cpp", function `TestWin32Handles()`. + +To use the extenion, buffers need to be created with `VkExternalMemoryBufferCreateInfoKHR` attached to their `pNext` chain, +and memory allocations need to be made with `VkExportMemoryAllocateInfoKHR` attached to their `pNext` chain. +To make use of them, you need to use \ref custom_memory_pools. Example: + +\code +// Define an example buffer and allocation parameters. +VkExternalMemoryBufferCreateInfoKHR externalMemBufCreateInfo = { + VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO_KHR, + nullptr, + VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT +}; +VkBufferCreateInfo exampleBufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +exampleBufCreateInfo.size = 0x10000; // Doesn't matter here. +exampleBufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; +exampleBufCreateInfo.pNext = &externalMemBufCreateInfo; + +VmaAllocationCreateInfo exampleAllocCreateInfo = {}; +exampleAllocCreateInfo.usage = VMA_MEMORY_USAGE_AUTO; + +// Find memory type index to use for the custom pool. +uint32_t memTypeIndex; +VkResult res = vmaFindMemoryTypeIndexForBufferInfo(g_Allocator, + &exampleBufCreateInfo, &exampleAllocCreateInfo, &memTypeIndex); +// Check res... + +// Create a custom pool. +constexpr static VkExportMemoryAllocateInfoKHR exportMemAllocInfo = { + VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO_KHR, + nullptr, + VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT +}; +VmaPoolCreateInfo poolCreateInfo = {}; +poolCreateInfo.memoryTypeIndex = memTypeIndex; +poolCreateInfo.pMemoryAllocateNext = (void*)&exportMemAllocInfo; + +VmaPool pool; +res = vmaCreatePool(g_Allocator, &poolCreateInfo, &pool); +// Check res... + +// YOUR OTHER CODE COMES HERE.... + +// At the end, don't forget to destroy it! +vmaDestroyPool(g_Allocator, pool); +\endcode + +Note that the structure passed as VmaPoolCreateInfo::pMemoryAllocateNext must remain alive and unchanged +for the whole lifetime of the custom pool, because it will be used when the pool allocates a new device memory block. +No copy is made internally. This is why variable `exportMemAllocInfo` is defined as `static`. + +\section vk_khr_external_memory_win32_memory_allocation Memory allocation + +Finally, you can create a buffer with an allocation out of the custom pool. +The buffer should use same flags as the sample buffer used to find the memory type. +It should also specify `VkExternalMemoryBufferCreateInfoKHR` in its `pNext` chain. + +\code +VkExternalMemoryBufferCreateInfoKHR externalMemBufCreateInfo = { + VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO_KHR, + nullptr, + VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT +}; +VkBufferCreateInfo bufCreateInfo = { VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO }; +bufCreateInfo.size = // Your desired buffer size. +bufCreateInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT; +bufCreateInfo.pNext = &externalMemBufCreateInfo; + +VmaAllocationCreateInfo allocCreateInfo = {}; +allocCreateInfo.pool = pool; // It is enough to set this one member. + +VkBuffer buf; +VmaAllocation alloc; +res = vmaCreateBuffer(g_Allocator, &bufCreateInfo, &allocCreateInfo, &buf, &alloc, nullptr); +// Check res... + +// YOUR OTHER CODE COMES HERE.... + +// At the end, don't forget to destroy it! +vmaDestroyBuffer(g_Allocator, buf, alloc); +\endcode + +If you need each allocation to have its own device memory block and start at offset 0, you can still do +by using #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT flag. It works also with custom pools. + +\section vk_khr_external_memory_win32_exporting_win32_handle Exporting Win32 handle + +After the allocation is created, you can acquire a Win32 `HANDLE` to the `VkDeviceMemory` block it belongs to. +VMA function vmaGetMemoryWin32Handle() is a replacement of the Vulkan function `vkGetMemoryWin32HandleKHR`. + +\code +HANDLE handle; +res = vmaGetMemoryWin32Handle(g_Allocator, alloc, nullptr, &handle); +// Check res... + +// YOUR OTHER CODE COMES HERE.... + +// At the end, you must close the handle. +CloseHandle(handle); +\endcode + +Documentation of the VK_KHR_external_memory_win32 extension states that: + +> If handleType is defined as an NT handle, vkGetMemoryWin32HandleKHR must be called no more than once for each valid unique combination of memory and handleType. + +This is ensured automatically inside VMA. +The library fetches the handle on first use, remembers it internally, and closes it when the memory block or dedicated allocation is destroyed. +Every time you call vmaGetMemoryWin32Handle(), VMA calls `DuplicateHandle` and returns a new handle that you need to close. + +For further information, please check documentation of the vmaGetMemoryWin32Handle() function. + + +\page enabling_buffer_device_address Enabling buffer device address + +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 has been promoted to core Vulkan 1.2. + +If you want to use this feature in connection with VMA, follow these steps: + +\section enabling_buffer_device_address_initialization Initialization + +1) (For Vulkan version < 1.2) Call `vkEnumerateDeviceExtensionProperties` for the physical device. +Check if the extension is supported - if returned array of `VkExtensionProperties` contains +"VK_KHR_buffer_device_address". + +2) Call `vkGetPhysicalDeviceFeatures2` for the physical device instead of old `vkGetPhysicalDeviceFeatures`. +Attach additional structure `VkPhysicalDeviceBufferDeviceAddressFeatures*` to `VkPhysicalDeviceFeatures2::pNext` to be returned. +Check if the device feature is really supported - check if `VkPhysicalDeviceBufferDeviceAddressFeatures::bufferDeviceAddress` is true. + +3) (For Vulkan version < 1.2) While creating device with `vkCreateDevice`, enable this extension - add +"VK_KHR_buffer_device_address" 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 `VkPhysicalDeviceBufferDeviceAddressFeatures*` to +`VkPhysicalDeviceFeatures2::pNext` and set its member `bufferDeviceAddress` to `VK_TRUE`. + +5) While creating #VmaAllocator with vmaCreateAllocator() inform VMA that you +have enabled this feature - add #VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT +to VmaAllocatorCreateInfo::flags. + +\section enabling_buffer_device_address_usage Usage + +After following steps described above, you can create buffers with `VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT*` using VMA. +The library automatically adds `VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT*` to +allocated memory blocks wherever it might be needed. + +Please note that the library supports only `VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT*`. +The second part of this functionality related to "capture and replay" is not supported, +as it is intended for usage in debugging tools like RenderDoc, not in everyday Vulkan usage. + +\section enabling_buffer_device_address_more_information More information + +To learn more about this extension, see [VK_KHR_buffer_device_address in Vulkan specification](https://www.khronos.org/registry/vulkan/specs/1.2-extensions/html/chap46.html#VK_KHR_buffer_device_address) + +Example use of this extension can be found in the code of the sample and test suite +accompanying this library. + +\page general_considerations General considerations + +\section general_considerations_thread_safety Thread safety + +- The library has no global state, so separate #VmaAllocator objects can be used + independently. + There should be no need to create multiple such objects though - one per `VkDevice` is enough. +- By default, all calls to functions that take #VmaAllocator as first parameter + are safe to call from multiple threads simultaneously because they are + synchronized internally when needed. + 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. +- Access to a #VmaAllocation object must be externally synchronized. For example, + you must not call vmaGetAllocationInfo() and vmaMapMemory() from different + threads at the same time if you pass the same #VmaAllocation object to these + functions. +- #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 compatibility 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 + +When using this library, you can meet following types of warnings issued by +Vulkan validation layer. They don't necessarily indicate a bug, so you may need +to just ignore them. + +- *vkBindBufferMemory(): Binding memory to buffer 0xeb8e4 but vkGetBufferMemoryRequirements() has not been called on that buffer.* + - It happens when VK_KHR_dedicated_allocation extension is enabled. + `vkGetBufferMemoryRequirements2KHR` function is used instead, while validation layer seems to be unaware of it. +- *Mapping an image with layout VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL can result in undefined behavior if this memory is used by the device. Only GENERAL or PREINITIALIZED should be used.* + - It happens when you map a buffer or image, because the library maps entire + `VkDeviceMemory` block, where different types of images and buffers may end + up together, especially on GPUs with unified memory like Intel. +- *Non-linear image 0xebc91 is aliased with linear buffer 0xeb8e4 which may indicate a bug.* + - It may happen when you use [defragmentation](@ref defragmentation). + +\section general_considerations_allocation_algorithm Allocation algorithm + +The library uses following algorithm for allocation, in order: + +-# Try to find free range of memory in existing blocks. +-# If failed, try to create a new block of `VkDeviceMemory`, with preferred block size. +-# If failed, try to create such block with size / 2, size / 4, size / 8. +-# If failed, try to allocate separate `VkDeviceMemory` for this allocation, + just like when you use #VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT. +-# If failed, choose other memory type that meets the requirements specified in + VmaAllocationCreateInfo and go to point 1. +-# If failed, return `VK_ERROR_OUT_OF_DEVICE_MEMORY`. + +\section general_considerations_features_not_supported Features not supported + +Features deliberately excluded from the scope of this library: + +-# **Data transfer.** Uploading (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/external/vma/vma.cpp b/external/vma/vma.cpp new file mode 100644 index 0000000..7d0e814 --- /dev/null +++ b/external/vma/vma.cpp @@ -0,0 +1,14 @@ +#define VMA_IMPLEMENTATION +#define VMA_STATIC_VULKAN_FUNCTIONS 0 +#define VMA_DYNAMIC_VULKAN_FUNCTIONS 1 +#define VMA_VULKAN_VERSION 1003000 + +#ifdef __linux__ +#define VK_USE_PLATFORM_XCB_KHR +#elif _WIN32 +#define VK_USE_PLATFORM_WIN32_KHR +#else +#error not yet implemented +#endif + +#include "vk_mem_alloc.h" diff --git a/external/xxhash/LICENSE b/external/xxhash/LICENSE new file mode 100644 index 0000000..e4c5da7 --- /dev/null +++ b/external/xxhash/LICENSE @@ -0,0 +1,26 @@ +xxHash Library +Copyright (c) 2012-2021 Yann Collet +All rights reserved. + +BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + +Redistribution and use in source and binary forms, with or without modification, +are permitted provided that the following conditions are met: + +* Redistributions of source code must retain the above copyright notice, this + list of conditions and the following disclaimer. + +* Redistributions in binary form must reproduce the above copyright notice, this + list of conditions and the following disclaimer in the documentation and/or + other materials provided with the distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR +ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON +ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/external/xxhash/README.md b/external/xxhash/README.md new file mode 100644 index 0000000..1ff987b --- /dev/null +++ b/external/xxhash/README.md @@ -0,0 +1,274 @@ + +xxHash - Extremely fast hash algorithm +====================================== + +xxHash is an Extremely fast Hash algorithm, processing at RAM speed limits. +Code is highly portable, and produces hashes identical across all platforms (little / big endian). +The library includes the following algorithms : +- XXH32 : generates 32-bit hashes, using 32-bit arithmetic +- XXH64 : generates 64-bit hashes, using 64-bit arithmetic +- XXH3 (since `v0.8.0`): generates 64 or 128-bit hashes, using vectorized arithmetic. + The 128-bit variant is called XXH128. + +All variants successfully complete the [SMHasher](https://code.google.com/p/smhasher/wiki/SMHasher) test suite +which evaluates the quality of hash functions (collision, dispersion and randomness). +Additional tests, which evaluate more thoroughly speed and collision properties of 64-bit hashes, [are also provided](https://github.com/Cyan4973/xxHash/tree/dev/tests). + +|Branch |Status | +|------------|---------| +|release | [![Build Status](https://github.com/Cyan4973/xxHash/actions/workflows/ci.yml/badge.svg?branch=release)](https://github.com/Cyan4973/xxHash/actions?query=branch%3Arelease+) | +|dev | [![Build Status](https://github.com/Cyan4973/xxHash/actions/workflows/ci.yml/badge.svg?branch=dev)](https://github.com/Cyan4973/xxHash/actions?query=branch%3Adev+) | + + +Benchmarks +------------------------- + +The benchmarked reference system uses an Intel i7-9700K cpu, and runs Ubuntu x64 20.04. +The [open source benchmark program] is compiled with `clang` v10.0 using `-O3` flag. + +| Hash Name | Width | Bandwidth (GB/s) | Small Data Velocity | Quality | Comment | +| --------- | ----- | ---------------- | ----- | --- | --- | +| __XXH3__ (SSE2) | 64 | 31.5 GB/s | 133.1 | 10 +| __XXH128__ (SSE2) | 128 | 29.6 GB/s | 118.1 | 10 +| _RAM sequential read_ | N/A | 28.0 GB/s | N/A | N/A | _for reference_ +| City64 | 64 | 22.0 GB/s | 76.6 | 10 +| T1ha2 | 64 | 22.0 GB/s | 99.0 | 9 | Slightly worse [collisions] +| City128 | 128 | 21.7 GB/s | 57.7 | 10 +| __XXH64__ | 64 | 19.4 GB/s | 71.0 | 10 +| SpookyHash | 64 | 19.3 GB/s | 53.2 | 10 +| Mum | 64 | 18.0 GB/s | 67.0 | 9 | Slightly worse [collisions] +| __XXH32__ | 32 | 9.7 GB/s | 71.9 | 10 +| City32 | 32 | 9.1 GB/s | 66.0 | 10 +| Murmur3 | 32 | 3.9 GB/s | 56.1 | 10 +| SipHash | 64 | 3.0 GB/s | 43.2 | 10 +| FNV64 | 64 | 1.2 GB/s | 62.7 | 5 | Poor avalanche properties +| Blake2 | 256 | 1.1 GB/s | 5.1 | 10 | Cryptographic +| SHA1 | 160 | 0.8 GB/s | 5.6 | 10 | Cryptographic but broken +| MD5 | 128 | 0.6 GB/s | 7.8 | 10 | Cryptographic but broken + +[open source benchmark program]: https://github.com/Cyan4973/xxHash/tree/release/tests/bench +[collisions]: https://github.com/Cyan4973/xxHash/wiki/Collision-ratio-comparison#collision-study + +note 1: Small data velocity is a _rough_ evaluation of algorithm's efficiency on small data. For more detailed analysis, please refer to next paragraph. + +note 2: some algorithms feature _faster than RAM_ speed. In which case, they can only reach their full speed potential when input is already in CPU cache (L3 or better). Otherwise, they max out on RAM speed limit. + +### Small data + +Performance on large data is only one part of the picture. +Hashing is also very useful in constructions like hash tables and bloom filters. +In these use cases, it's frequent to hash a lot of small data (starting at a few bytes). +Algorithm's performance can be very different for such scenarios, since parts of the algorithm, +such as initialization or finalization, become fixed cost. +The impact of branch mis-prediction also becomes much more present. + +XXH3 has been designed for excellent performance on both long and small inputs, +which can be observed in the following graph: + +![XXH3, latency, random size](https://user-images.githubusercontent.com/750081/61976089-aedeab00-af9f-11e9-9239-e5375d6c080f.png) + +For a more detailed analysis, please visit the wiki : +https://github.com/Cyan4973/xxHash/wiki/Performance-comparison#benchmarks-concentrating-on-small-data- + +Quality +------------------------- + +Speed is not the only property that matters. +Produced hash values must respect excellent dispersion and randomness properties, +so that any sub-section of it can be used to maximally spread out a table or index, +as well as reduce the amount of collisions to the minimal theoretical level, following the [birthday paradox]. + +`xxHash` has been tested with Austin Appleby's excellent SMHasher test suite, +and passes all tests, ensuring reasonable quality levels. +It also passes extended tests from [newer forks of SMHasher], featuring additional scenarios and conditions. + +Finally, xxHash provides its own [massive collision tester](https://github.com/Cyan4973/xxHash/tree/dev/tests/collisions), +able to generate and compare billions of hashes to test the limits of 64-bit hash algorithms. +On this front too, xxHash features good results, in line with the [birthday paradox]. +A more detailed analysis is documented [in the wiki](https://github.com/Cyan4973/xxHash/wiki/Collision-ratio-comparison). + +[birthday paradox]: https://en.wikipedia.org/wiki/Birthday_problem +[newer forks of SMHasher]: https://github.com/rurban/smhasher + + +### Build modifiers + +The following macros can be set at compilation time to modify `libxxhash`'s behavior. They are generally disabled by default. + +- `XXH_INLINE_ALL`: Make all functions `inline`, implementation is directly included within `xxhash.h`. + Inlining functions is beneficial for speed, notably for small keys. + It's _extremely effective_ when key's length is expressed as _a compile time constant_, + with performance improvements observed in the +200% range . + See [this article](https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html) for details. +- `XXH_PRIVATE_API`: same outcome as `XXH_INLINE_ALL`. Still available for legacy support. + The name underlines that `XXH_*` symbol names will not be exported. +- `XXH_STATIC_LINKING_ONLY`: gives access to internal state declaration, required for static allocation. + Incompatible with dynamic linking, due to risks of ABI changes. +- `XXH_NAMESPACE`: Prefixes all symbols with the value of `XXH_NAMESPACE`. + This macro can only use compilable character set. + Useful to evade symbol naming collisions, + in case of multiple inclusions of xxHash's source code. + Client applications still use the regular function names, + as symbols are automatically translated through `xxhash.h`. +- `XXH_FORCE_ALIGN_CHECK`: Use a faster direct read path when input is aligned. + This option can result in dramatic performance improvement on architectures unable to load memory from unaligned addresses + when input to hash happens to be aligned on 32 or 64-bit boundaries. + It is (slightly) detrimental on platform with good unaligned memory access performance (same instruction for both aligned and unaligned accesses). + This option is automatically disabled on `x86`, `x64` and `aarch64`, and enabled on all other platforms. +- `XXH_FORCE_MEMORY_ACCESS`: The default method `0` uses a portable `memcpy()` notation. + Method `1` uses a gcc-specific `packed` attribute, which can provide better performance for some targets. + Method `2` forces unaligned reads, which is not standard compliant, but might sometimes be the only way to extract better read performance. + Method `3` uses a byteshift operation, which is best for old compilers which don't inline `memcpy()` or big-endian systems without a byteswap instruction. +- `XXH_CPU_LITTLE_ENDIAN`: By default, endianness is determined by a runtime test resolved at compile time. + If, for some reason, the compiler cannot simplify the runtime test, it can cost performance. + It's possible to skip auto-detection and simply state that the architecture is little-endian by setting this macro to 1. + Setting it to 0 states big-endian. +- `XXH_ENABLE_AUTOVECTORIZE`: Auto-vectorization may be triggered for XXH32 and XXH64, depending on cpu vector capabilities and compiler version. + Note: auto-vectorization tends to be triggered more easily with recent versions of `clang`. + For XXH32, SSE4.1 or equivalent (NEON) is enough, while XXH64 requires AVX512. + Unfortunately, auto-vectorization is generally detrimental to XXH performance. + For this reason, the xxhash source code tries to prevent auto-vectorization by default. + That being said, systems evolve, and this conclusion is not forthcoming. + For example, it has been reported that recent Zen4 cpus are more likely to improve performance with vectorization. + Therefore, should you prefer or want to test vectorized code, you can enable this flag: + it will remove the no-vectorization protection code, thus making it more likely for XXH32 and XXH64 to be auto-vectorized. +- `XXH32_ENDJMP`: Switch multi-branch finalization stage of XXH32 by a single jump. + This is generally undesirable for performance, especially when hashing inputs of random sizes. + But depending on exact architecture and compiler, a jump might provide slightly better performance on small inputs. Disabled by default. +- `XXH_IMPORT`: MSVC specific: should only be defined for dynamic linking, as it prevents linkage errors. +- `XXH_NO_STDLIB`: Disable invocation of `` functions, notably `malloc()` and `free()`. + `libxxhash`'s `XXH*_createState()` will always fail and return `NULL`. + But one-shot hashing (like `XXH32()`) or streaming using statically allocated states + still work as expected. + This build flag is useful for embedded environments without dynamic allocation. +- `XXH_memcpy`, `XXH_memset`, `XXH_memcmp` : redirect `memcpy()`, `memset()` and `memcmp()` to some user-selected symbol at compile time. + Redirecting all 3 removes the need to include `` standard library. +- `XXH_NO_EXTERNC_GUARD`: When `xxhash.h` is compiled in C++ mode, removes the `extern "C" { .. }` block guard. +- `XXH_DEBUGLEVEL` : When set to any value >= 1, enables `assert()` statements. + This (slightly) slows down execution, but may help finding bugs during debugging sessions. + +#### Binary size control +- `XXH_NO_XXH3` : removes symbols related to `XXH3` (both 64 & 128 bits) from generated binary. + `XXH3` is by far the largest contributor to `libxxhash` size, + so it's useful to reduce binary size for applications which do not employ `XXH3`. +- `XXH_NO_LONG_LONG`: removes compilation of algorithms relying on 64-bit `long long` types + which include `XXH3` and `XXH64`. + Only `XXH32` will be compiled. + Useful for targets (architectures and compilers) without 64-bit support. +- `XXH_NO_STREAM`: Disables the streaming API, limiting the library to single shot variants only. +- `XXH_NO_INLINE_HINTS`: By default, xxHash uses `__attribute__((always_inline))` and `__forceinline` to improve performance at the cost of code size. + Defining this macro to 1 will mark all internal functions as `static`, allowing the compiler to decide whether to inline a function or not. + This is very useful when optimizing for smallest binary size, + and is automatically defined when compiling with `-O0`, `-Os`, `-Oz`, or `-fno-inline` on GCC and Clang. + It may also be required to successfully compile using `-Og`, depending on compiler version. +- `XXH_SIZE_OPT`: `0`: default, optimize for speed + `1`: default for `-Os` and `-Oz`: disables some speed hacks for size optimization + `2`: makes code as small as possible, performance may cry + +#### Build modifiers specific for XXH3 +- `XXH_VECTOR` : manually select a vector instruction set (default: auto-selected at compilation time). Available instruction sets are `XXH_SCALAR`, `XXH_SSE2`, `XXH_AVX2`, `XXH_AVX512`, `XXH_NEON` and `XXH_VSX`. Compiler may require additional flags to ensure proper support (for example, `gcc` on x86_64 requires `-mavx2` for `AVX2`, or `-mavx512f` for `AVX512`). +- `XXH_PREFETCH_DIST` : select prefetching distance. For close-to-metal adaptation to specific hardware platforms. XXH3 only. +- `XXH_NO_PREFETCH` : disable prefetching. Some platforms or situations may perform better without prefetching. XXH3 only. + +#### Makefile variables +When compiling the Command Line Interface `xxhsum` using `make`, the following environment variables can also be set : +- `DISPATCH=1` : use `xxh_x86dispatch.c`, select at runtime between `scalar`, `sse2`, `avx2` or `avx512` instruction set. This option is only valid for `x86`/`x64` systems. It is enabled by default when target `x86`/`x64` is detected. It can be forcefully turned off using `DISPATCH=0`. +- `LIBXXH_DISPATCH=1` : same idea, implemented a runtime vector extension detector, but within `libxxhash`. This parameter is disabled by default. When enabled (only valid for `x86`/`x64` systems), new symbols published in `xxh_x86dispatch.h` become accessible. At the time of this writing, it's required to include `xxh_x86dispatch.h` in order to access the symbols with runtime vector extension detection. +- `XXH_1ST_SPEED_TARGET` : select an initial speed target, expressed in MB/s, for the first speed test in benchmark mode. Benchmark will adjust the target at subsequent iterations, but the first test is made "blindly" by targeting this speed. Currently conservatively set to 10 MB/s, to support very slow (emulated) platforms. +- `NODE_JS=1` : When compiling `xxhsum` for Node.js with Emscripten, this links the `NODERAWFS` library for unrestricted filesystem access and patches `isatty` to make the command line utility correctly detect the terminal. This does make the binary specific to Node.js. + +### Building xxHash - Using vcpkg + +You can download and install xxHash using the [vcpkg](https://github.com/Microsoft/vcpkg) dependency manager: + + git clone https://github.com/Microsoft/vcpkg.git + cd vcpkg + ./bootstrap-vcpkg.sh + ./vcpkg integrate install + ./vcpkg install xxhash + +The xxHash port in vcpkg is kept up to date by Microsoft team members and community contributors. If the version is out of date, please [create an issue or pull request](https://github.com/Microsoft/vcpkg) on the vcpkg repository. + +### Example + +The simplest example calls xxhash 64-bit variant as a one-shot function +generating a hash value from a single buffer, and invoked from a C/C++ program: + +```C +#include "xxhash.h" + + (...) + XXH64_hash_t hash = XXH64(buffer, size, seed); +} +``` + +Streaming variant is more involved, but makes it possible to provide data incrementally: + +```C +#include "stdlib.h" /* abort() */ +#include "xxhash.h" + + +XXH64_hash_t calcul_hash_streaming(FileHandler fh) +{ + /* create a hash state */ + XXH64_state_t* const state = XXH64_createState(); + if (state==NULL) abort(); + + size_t const bufferSize = SOME_SIZE; + void* const buffer = malloc(bufferSize); + if (buffer==NULL) abort(); + + /* Initialize state with selected seed */ + XXH64_hash_t const seed = 0; /* or any other value */ + if (XXH64_reset(state, seed) == XXH_ERROR) abort(); + + /* Feed the state with input data, any size, any number of times */ + (...) + while ( /* some data left */ ) { + size_t const length = get_more_data(buffer, bufferSize, fh); + if (XXH64_update(state, buffer, length) == XXH_ERROR) abort(); + (...) + } + (...) + + /* Produce the final hash value */ + XXH64_hash_t const hash = XXH64_digest(state); + + /* State could be re-used; but in this example, it is simply freed */ + free(buffer); + XXH64_freeState(state); + + return hash; +} +``` + + +### License + +The library files `xxhash.c` and `xxhash.h` are BSD licensed. +The utility `xxhsum` is GPL licensed. + + +### Other programming languages + +Beyond the C reference version, +xxHash is also available from many different programming languages, +thanks to great contributors. +They are [listed here](http://www.xxhash.com/#other-languages). + + +### Packaging status + +Many distributions bundle a package manager +which allows easy xxhash installation as both a `libxxhash` library +and `xxhsum` command line interface. + +[![Packaging status](https://repology.org/badge/vertical-allrepos/xxhash.svg)](https://repology.org/project/xxhash/versions) + + +### Special Thanks + +- Takayuki Matsuoka, aka @t-mat, for creating `xxhsum -c` and great support during early xxh releases +- Mathias Westerdahl, aka @JCash, for introducing the first version of `XXH64` +- Devin Hussey, aka @easyaspi314, for incredible low-level optimizations on `XXH3` and `XXH128` diff --git a/external/xxhash/SECURITY.md b/external/xxhash/SECURITY.md new file mode 100644 index 0000000..2a8b4c8 --- /dev/null +++ b/external/xxhash/SECURITY.md @@ -0,0 +1,13 @@ +# Security Policy + +## Supported Versions + +Security updates are applied only to the latest release. + +## Reporting a Vulnerability + +If you have discovered a security vulnerability in this project, please report it privately. **Do not disclose it as a public issue.** This gives us time to work with you to fix the issue before public exposure, reducing the chance that the exploit will be used before a patch is released. + +Please disclose it at [security advisory](https://github.com/Cyan4973/xxHash/security/advisories/new). + +This project is maintained by a team of volunteers on a reasonable-effort basis. As such, please give us at least 90 days to work on a fix before public exposure. diff --git a/external/xxhash/xxh3.h b/external/xxhash/xxh3.h new file mode 100644 index 0000000..7e3ce68 --- /dev/null +++ b/external/xxhash/xxh3.h @@ -0,0 +1,55 @@ +/* + * xxHash - Extremely Fast Hash algorithm + * Development source file for `xxh3` + * Copyright (C) 2019-2021 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + +/* + * Note: This file used to host the source code of XXH3_* variants. + * during the development period. + * The source code is now properly integrated within xxhash.h. + * + * xxh3.h is no longer useful, + * but it is still provided for compatibility with source code + * which used to include it directly. + * + * Programs are now highly discouraged to include xxh3.h. + * Include `xxhash.h` instead, which is the officially supported interface. + * + * In the future, xxh3.h will start to generate warnings, then errors, + * then it will be removed from source package and from include directory. + */ + +/* Simulate the same impact as including the old xxh3.h source file */ + +#define XXH_INLINE_ALL +#include "xxhash.h" diff --git a/external/xxhash/xxh_x86dispatch.c b/external/xxhash/xxh_x86dispatch.c new file mode 100644 index 0000000..0c15820 --- /dev/null +++ b/external/xxhash/xxh_x86dispatch.c @@ -0,0 +1,821 @@ +/* + * xxHash - Extremely Fast Hash algorithm + * Copyright (C) 2020-2021 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + + +/*! + * @file xxh_x86dispatch.c + * + * Automatic dispatcher code for the @ref XXH3_family on x86-based targets. + * + * Optional add-on. + * + * **Compile this file with the default flags for your target.** + * Note that compiling with flags like `-mavx*`, `-march=native`, or `/arch:AVX*` + * will make the resulting binary incompatible with cpus not supporting the requested instruction set. + * + * @defgroup dispatch x86 Dispatcher + * @{ + */ + +#if defined (__cplusplus) +extern "C" { +#endif + +#if !(defined(__x86_64__) || defined(__i386__) || defined(_M_IX86) || defined(_M_X64)) +# error "Dispatching is currently only supported on x86 and x86_64." +#endif + +/*! @cond Doxygen ignores this part */ +#ifndef XXH_HAS_INCLUDE +# ifdef __has_include +/* + * Not defined as XXH_HAS_INCLUDE(x) (function-like) because + * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion) + */ +# define XXH_HAS_INCLUDE __has_include +# else +# define XXH_HAS_INCLUDE(x) 0 +# endif +#endif +/*! @endcond */ + +/*! + * @def XXH_DISPATCH_SCALAR + * @brief Enables/dispatching the scalar code path. + * + * If this is defined to 0, SSE2 support is assumed. This reduces code size + * when the scalar path is not needed. + * + * This is automatically defined to 0 when... + * - SSE2 support is enabled in the compiler + * - Targeting x86_64 + * - Targeting Android x86 + * - Targeting macOS + */ +#ifndef XXH_DISPATCH_SCALAR +# if defined(__SSE2__) || (defined(_M_IX86_FP) && _M_IX86_FP >= 2) /* SSE2 on by default */ \ + || defined(__x86_64__) || defined(_M_X64) /* x86_64 */ \ + || defined(__ANDROID__) || defined(__APPLE__) /* Android or macOS */ +# define XXH_DISPATCH_SCALAR 0 /* disable */ +# else +# define XXH_DISPATCH_SCALAR 1 +# endif +#endif +/*! + * @def XXH_DISPATCH_AVX2 + * @brief Enables/disables dispatching for AVX2. + * + * This is automatically detected if it is not defined. + * - GCC 4.7 and later are known to support AVX2, but >4.9 is required for + * to get the AVX2 intrinsics and typedefs without -mavx -mavx2. + * - Visual Studio 2013 Update 2 and later are known to support AVX2. + * - The GCC/Clang internal header `` is detected. While this is + * not allowed to be included directly, it still appears in the builtin + * include path and is detectable with `__has_include`. + * + * @see XXH_AVX2 + */ +#ifndef XXH_DISPATCH_AVX2 +# if (defined(__GNUC__) && (__GNUC__ > 4)) /* GCC 5.0+ */ \ + || (defined(_MSC_VER) && _MSC_VER >= 1900) /* VS 2015+ */ \ + || (defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 180030501) /* VS 2013 Update 2 */ \ + || XXH_HAS_INCLUDE() /* GCC/Clang internal header */ +# define XXH_DISPATCH_AVX2 1 /* enable dispatch towards AVX2 */ +# else +# define XXH_DISPATCH_AVX2 0 +# endif +#endif /* XXH_DISPATCH_AVX2 */ + +/*! + * @def XXH_DISPATCH_AVX512 + * @brief Enables/disables dispatching for AVX512. + * + * Automatically detected if one of the following conditions is met: + * - GCC 4.9 and later are known to support AVX512. + * - Visual Studio 2017 and later are known to support AVX2. + * - The GCC/Clang internal header `` is detected. While this + * is not allowed to be included directly, it still appears in the builtin + * include path and is detectable with `__has_include`. + * + * @see XXH_AVX512 + */ +#ifndef XXH_DISPATCH_AVX512 +# if (defined(__GNUC__) \ + && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9))) /* GCC 4.9+ */ \ + || (defined(_MSC_VER) && _MSC_VER >= 1910) /* VS 2017+ */ \ + || XXH_HAS_INCLUDE() /* GCC/Clang internal header */ +# define XXH_DISPATCH_AVX512 1 /* enable dispatch towards AVX512 */ +# else +# define XXH_DISPATCH_AVX512 0 +# endif +#endif /* XXH_DISPATCH_AVX512 */ + +/*! + * @def XXH_TARGET_SSE2 + * @brief Allows a function to be compiled with SSE2 intrinsics. + * + * Uses `__attribute__((__target__("sse2")))` on GCC to allow SSE2 to be used + * even with `-mno-sse2`. + * + * @def XXH_TARGET_AVX2 + * @brief Like @ref XXH_TARGET_SSE2, but for AVX2. + * + * @def XXH_TARGET_AVX512 + * @brief Like @ref XXH_TARGET_SSE2, but for AVX512. + * + */ +#if defined(__GNUC__) +# include /* SSE2 */ +# if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512 +# include /* AVX2, AVX512F */ +# endif +# define XXH_TARGET_SSE2 __attribute__((__target__("sse2"))) +# define XXH_TARGET_AVX2 __attribute__((__target__("avx2"))) +# define XXH_TARGET_AVX512 __attribute__((__target__("avx512f"))) +#elif defined(__clang__) && defined(_MSC_VER) /* clang-cl.exe */ +# include /* SSE2 */ +# if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512 +# include /* AVX2, AVX512F */ +# include +# include +# include +# include +# endif +# define XXH_TARGET_SSE2 __attribute__((__target__("sse2"))) +# define XXH_TARGET_AVX2 __attribute__((__target__("avx2"))) +# define XXH_TARGET_AVX512 __attribute__((__target__("avx512f"))) +#elif defined(_MSC_VER) +# include +# define XXH_TARGET_SSE2 +# define XXH_TARGET_AVX2 +# define XXH_TARGET_AVX512 +#else +# error "Dispatching is currently not supported for your compiler." +#endif + +/*! @cond Doxygen ignores this part */ +#ifdef XXH_DISPATCH_DEBUG +/* debug logging */ +# include +# define XXH_debugPrint(str) { fprintf(stderr, "DEBUG: xxHash dispatch: %s \n", str); fflush(NULL); } +#else +# define XXH_debugPrint(str) ((void)0) +# undef NDEBUG /* avoid redefinition */ +# define NDEBUG +#endif +/*! @endcond */ +#include + +#ifndef XXH_DOXYGEN +#define XXH_INLINE_ALL +#define XXH_X86DISPATCH +#include "xxhash.h" +#endif + +/*! @cond Doxygen ignores this part */ +#ifndef XXH_HAS_ATTRIBUTE +# ifdef __has_attribute +# define XXH_HAS_ATTRIBUTE(...) __has_attribute(__VA_ARGS__) +# else +# define XXH_HAS_ATTRIBUTE(...) 0 +# endif +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +#if XXH_HAS_ATTRIBUTE(constructor) +# define XXH_CONSTRUCTOR __attribute__((constructor)) +# define XXH_DISPATCH_MAYBE_NULL 0 +#else +# define XXH_CONSTRUCTOR +# define XXH_DISPATCH_MAYBE_NULL 1 +#endif +/*! @endcond */ + + +/*! @cond Doxygen ignores this part */ +/* + * Support both AT&T and Intel dialects + * + * GCC doesn't convert AT&T syntax to Intel syntax, and will error out if + * compiled with -masm=intel. Instead, it supports dialect switching with + * curly braces: { AT&T syntax | Intel syntax } + * + * Clang's integrated assembler automatically converts AT&T syntax to Intel if + * needed, making the dialect switching useless (it isn't even supported). + * + * Note: Comments are written in the inline assembly itself. + */ +#ifdef __clang__ +# define XXH_I_ATT(intel, att) att "\n\t" +#else +# define XXH_I_ATT(intel, att) "{" att "|" intel "}\n\t" +#endif +/*! @endcond */ + +/*! + * @private + * @brief Runs CPUID. + * + * @param eax , ecx The parameters to pass to CPUID, %eax and %ecx respectively. + * @param abcd The array to store the result in, `{ eax, ebx, ecx, edx }` + */ +static void XXH_cpuid(xxh_u32 eax, xxh_u32 ecx, xxh_u32* abcd) +{ +#if defined(_MSC_VER) + __cpuidex((int*)abcd, eax, ecx); +#else + xxh_u32 ebx, edx; +# if defined(__i386__) && defined(__PIC__) + __asm__( + "# Call CPUID\n\t" + "#\n\t" + "# On 32-bit x86 with PIC enabled, we are not allowed to overwrite\n\t" + "# EBX, so we use EDI instead.\n\t" + XXH_I_ATT("mov edi, ebx", "movl %%ebx, %%edi") + XXH_I_ATT("cpuid", "cpuid" ) + XXH_I_ATT("xchg edi, ebx", "xchgl %%ebx, %%edi") + : "=D" (ebx), +# else + __asm__( + "# Call CPUID\n\t" + XXH_I_ATT("cpuid", "cpuid") + : "=b" (ebx), +# endif + "+a" (eax), "+c" (ecx), "=d" (edx)); + abcd[0] = eax; + abcd[1] = ebx; + abcd[2] = ecx; + abcd[3] = edx; +#endif +} + +/* + * Modified version of Intel's guide + * https://software.intel.com/en-us/articles/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family + */ + +#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512 +/*! + * @private + * @brief Runs `XGETBV`. + * + * While the CPU may support AVX2, the operating system might not properly save + * the full YMM/ZMM registers. + * + * xgetbv is used for detecting this: Any compliant operating system will define + * a set of flags in the xcr0 register indicating how it saves the AVX registers. + * + * You can manually disable this flag on Windows by running, as admin: + * + * bcdedit.exe /set xsavedisable 1 + * + * and rebooting. Run the same command with 0 to re-enable it. + */ +static xxh_u64 XXH_xgetbv(void) +{ +#if defined(_MSC_VER) + return _xgetbv(0); /* min VS2010 SP1 compiler is required */ +#else + xxh_u32 xcr0_lo, xcr0_hi; + __asm__( + "# Call XGETBV\n\t" + "#\n\t" + "# Older assemblers (e.g. macOS's ancient GAS version) don't support\n\t" + "# the XGETBV opcode, so we encode it by hand instead.\n\t" + "# See for details.\n\t" + ".byte 0x0f, 0x01, 0xd0\n\t" + : "=a" (xcr0_lo), "=d" (xcr0_hi) : "c" (0)); + return xcr0_lo | ((xxh_u64)xcr0_hi << 32); +#endif +} +#endif + +/*! @cond Doxygen ignores this part */ +#define XXH_SSE2_CPUID_MASK (1 << 26) +#define XXH_OSXSAVE_CPUID_MASK ((1 << 26) | (1 << 27)) +#define XXH_AVX2_CPUID_MASK (1 << 5) +#define XXH_AVX2_XGETBV_MASK ((1 << 2) | (1 << 1)) +#define XXH_AVX512F_CPUID_MASK (1 << 16) +#define XXH_AVX512F_XGETBV_MASK ((7 << 5) | (1 << 2) | (1 << 1)) +/*! @endcond */ + +/*! + * @private + * @brief Returns the best XXH3 implementation. + * + * Runs various CPUID/XGETBV tests to try and determine the best implementation. + * + * @return The best @ref XXH_VECTOR implementation. + * @see XXH_VECTOR_TYPES + */ +int XXH_featureTest(void) +{ + xxh_u32 abcd[4]; + xxh_u32 max_leaves; + int best = XXH_SCALAR; +#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512 + xxh_u64 xgetbv_val; +#endif +#if defined(__GNUC__) && defined(__i386__) + xxh_u32 cpuid_supported; + __asm__( + "# For the sake of ruthless backwards compatibility, check if CPUID\n\t" + "# is supported in the EFLAGS on i386.\n\t" + "# This is not necessary on x86_64 - CPUID is mandatory.\n\t" + "# The ID flag (bit 21) in the EFLAGS register indicates support\n\t" + "# for the CPUID instruction. If a software procedure can set and\n\t" + "# clear this flag, the processor executing the procedure supports\n\t" + "# the CPUID instruction.\n\t" + "# \n\t" + "#\n\t" + "# Routine is from .\n\t" + + "# Save EFLAGS\n\t" + XXH_I_ATT("pushfd", "pushfl" ) + "# Store EFLAGS\n\t" + XXH_I_ATT("pushfd", "pushfl" ) + "# Invert the ID bit in stored EFLAGS\n\t" + XXH_I_ATT("xor dword ptr[esp], 0x200000", "xorl $0x200000, (%%esp)") + "# Load stored EFLAGS (with ID bit inverted)\n\t" + XXH_I_ATT("popfd", "popfl" ) + "# Store EFLAGS again (ID bit may or not be inverted)\n\t" + XXH_I_ATT("pushfd", "pushfl" ) + "# eax = modified EFLAGS (ID bit may or may not be inverted)\n\t" + XXH_I_ATT("pop eax", "popl %%eax" ) + "# eax = whichever bits were changed\n\t" + XXH_I_ATT("xor eax, dword ptr[esp]", "xorl (%%esp), %%eax" ) + "# Restore original EFLAGS\n\t" + XXH_I_ATT("popfd", "popfl" ) + "# eax = zero if ID bit can't be changed, else non-zero\n\t" + XXH_I_ATT("and eax, 0x200000", "andl $0x200000, %%eax" ) + : "=a" (cpuid_supported) :: "cc"); + + if (XXH_unlikely(!cpuid_supported)) { + XXH_debugPrint("CPUID support is not detected!"); + return best; + } + +#endif + /* Check how many CPUID pages we have */ + XXH_cpuid(0, 0, abcd); + max_leaves = abcd[0]; + + /* Shouldn't happen on hardware, but happens on some QEMU configs. */ + if (XXH_unlikely(max_leaves == 0)) { + XXH_debugPrint("Max CPUID leaves == 0!"); + return best; + } + + /* Check for SSE2, OSXSAVE and xgetbv */ + XXH_cpuid(1, 0, abcd); + + /* + * Test for SSE2. The check is redundant on x86_64, but it doesn't hurt. + */ + if (XXH_unlikely((abcd[3] & XXH_SSE2_CPUID_MASK) != XXH_SSE2_CPUID_MASK)) + return best; + + XXH_debugPrint("SSE2 support detected."); + + best = XXH_SSE2; +#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512 + /* Make sure we have enough leaves */ + if (XXH_unlikely(max_leaves < 7)) + return best; + + /* Test for OSXSAVE and XGETBV */ + if ((abcd[2] & XXH_OSXSAVE_CPUID_MASK) != XXH_OSXSAVE_CPUID_MASK) + return best; + + /* CPUID check for AVX features */ + XXH_cpuid(7, 0, abcd); + + xgetbv_val = XXH_xgetbv(); +#if XXH_DISPATCH_AVX2 + /* Validate that AVX2 is supported by the CPU */ + if ((abcd[1] & XXH_AVX2_CPUID_MASK) != XXH_AVX2_CPUID_MASK) + return best; + + /* Validate that the OS supports YMM registers */ + if ((xgetbv_val & XXH_AVX2_XGETBV_MASK) != XXH_AVX2_XGETBV_MASK) { + XXH_debugPrint("AVX2 supported by the CPU, but not the OS."); + return best; + } + + /* AVX2 supported */ + XXH_debugPrint("AVX2 support detected."); + best = XXH_AVX2; +#endif +#if XXH_DISPATCH_AVX512 + /* Check if AVX512F is supported by the CPU */ + if ((abcd[1] & XXH_AVX512F_CPUID_MASK) != XXH_AVX512F_CPUID_MASK) { + XXH_debugPrint("AVX512F not supported by CPU"); + return best; + } + + /* Validate that the OS supports ZMM registers */ + if ((xgetbv_val & XXH_AVX512F_XGETBV_MASK) != XXH_AVX512F_XGETBV_MASK) { + XXH_debugPrint("AVX512F supported by the CPU, but not the OS."); + return best; + } + + /* AVX512F supported */ + XXH_debugPrint("AVX512F support detected."); + best = XXH_AVX512; +#endif +#endif + return best; +} + + +/* === Vector implementations === */ + +/*! @cond PRIVATE */ +/*! + * @private + * @brief Defines the various dispatch functions. + * + * TODO: Consolidate? + * + * @param suffix The suffix for the functions, e.g. sse2 or scalar + * @param target XXH_TARGET_* or empty. + */ + +#define XXH_DEFINE_DISPATCH_FUNCS(suffix, target) \ + \ +/* === XXH3, default variants === */ \ + \ +XXH_NO_INLINE target XXH64_hash_t \ +XXHL64_default_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, \ + size_t len) \ +{ \ + return XXH3_hashLong_64b_internal( \ + input, len, XXH3_kSecret, sizeof(XXH3_kSecret), \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix \ + ); \ +} \ + \ +/* === XXH3, Seeded variants === */ \ + \ +XXH_NO_INLINE target XXH64_hash_t \ +XXHL64_seed_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, size_t len, \ + XXH64_hash_t seed) \ +{ \ + return XXH3_hashLong_64b_withSeed_internal( \ + input, len, seed, XXH3_accumulate_##suffix, \ + XXH3_scrambleAcc_##suffix, XXH3_initCustomSecret_##suffix \ + ); \ +} \ + \ +/* === XXH3, Secret variants === */ \ + \ +XXH_NO_INLINE target XXH64_hash_t \ +XXHL64_secret_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, \ + size_t len, XXH_NOESCAPE const void* secret, \ + size_t secretLen) \ +{ \ + return XXH3_hashLong_64b_internal( \ + input, len, secret, secretLen, \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix \ + ); \ +} \ + \ +/* === XXH3 update variants === */ \ + \ +XXH_NO_INLINE target XXH_errorcode \ +XXH3_update_##suffix(XXH_NOESCAPE XXH3_state_t* state, \ + XXH_NOESCAPE const void* input, size_t len) \ +{ \ + return XXH3_update(state, (const xxh_u8*)input, len, \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix); \ +} \ + \ +/* === XXH128 default variants === */ \ + \ +XXH_NO_INLINE target XXH128_hash_t \ +XXHL128_default_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, \ + size_t len) \ +{ \ + return XXH3_hashLong_128b_internal( \ + input, len, XXH3_kSecret, sizeof(XXH3_kSecret), \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix \ + ); \ +} \ + \ +/* === XXH128 Secret variants === */ \ + \ +XXH_NO_INLINE target XXH128_hash_t \ +XXHL128_secret_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, \ + size_t len, \ + XXH_NOESCAPE const void* XXH_RESTRICT secret, \ + size_t secretLen) \ +{ \ + return XXH3_hashLong_128b_internal( \ + input, len, (const xxh_u8*)secret, secretLen, \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix); \ +} \ + \ +/* === XXH128 Seeded variants === */ \ + \ +XXH_NO_INLINE target XXH128_hash_t \ +XXHL128_seed_##suffix(XXH_NOESCAPE const void* XXH_RESTRICT input, size_t len,\ + XXH64_hash_t seed) \ +{ \ + return XXH3_hashLong_128b_withSeed_internal(input, len, seed, \ + XXH3_accumulate_##suffix, XXH3_scrambleAcc_##suffix, \ + XXH3_initCustomSecret_##suffix); \ +} + +/*! @endcond */ +/* End XXH_DEFINE_DISPATCH_FUNCS */ + +/*! @cond Doxygen ignores this part */ +#if XXH_DISPATCH_SCALAR +XXH_DEFINE_DISPATCH_FUNCS(scalar, /* nothing */) +#endif +XXH_DEFINE_DISPATCH_FUNCS(sse2, XXH_TARGET_SSE2) +#if XXH_DISPATCH_AVX2 +XXH_DEFINE_DISPATCH_FUNCS(avx2, XXH_TARGET_AVX2) +#endif +#if XXH_DISPATCH_AVX512 +XXH_DEFINE_DISPATCH_FUNCS(avx512, XXH_TARGET_AVX512) +#endif +#undef XXH_DEFINE_DISPATCH_FUNCS +/*! @endcond */ + +/* ==== Dispatchers ==== */ + +/*! @cond Doxygen ignores this part */ +typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_default)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t); + +typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_withSeed)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t, XXH64_hash_t); + +typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_withSecret)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t, XXH_NOESCAPE const void* XXH_RESTRICT, size_t); + +typedef XXH_errorcode (*XXH3_dispatchx86_update)(XXH_NOESCAPE XXH3_state_t*, XXH_NOESCAPE const void*, size_t); + +typedef struct { + XXH3_dispatchx86_hashLong64_default hashLong64_default; + XXH3_dispatchx86_hashLong64_withSeed hashLong64_seed; + XXH3_dispatchx86_hashLong64_withSecret hashLong64_secret; + XXH3_dispatchx86_update update; +} XXH_dispatchFunctions_s; + +#define XXH_NB_DISPATCHES 4 +/*! @endcond */ + +/*! + * @private + * @brief Table of dispatchers for @ref XXH3_64bits(). + * + * @pre The indices must match @ref XXH_VECTOR_TYPE. + */ +static const XXH_dispatchFunctions_s XXH_kDispatch[XXH_NB_DISPATCHES] = { +#if XXH_DISPATCH_SCALAR + /* Scalar */ { XXHL64_default_scalar, XXHL64_seed_scalar, XXHL64_secret_scalar, XXH3_update_scalar }, +#else + /* Scalar */ { NULL, NULL, NULL, NULL }, +#endif + /* SSE2 */ { XXHL64_default_sse2, XXHL64_seed_sse2, XXHL64_secret_sse2, XXH3_update_sse2 }, +#if XXH_DISPATCH_AVX2 + /* AVX2 */ { XXHL64_default_avx2, XXHL64_seed_avx2, XXHL64_secret_avx2, XXH3_update_avx2 }, +#else + /* AVX2 */ { NULL, NULL, NULL, NULL }, +#endif +#if XXH_DISPATCH_AVX512 + /* AVX512 */ { XXHL64_default_avx512, XXHL64_seed_avx512, XXHL64_secret_avx512, XXH3_update_avx512 } +#else + /* AVX512 */ { NULL, NULL, NULL, NULL } +#endif +}; +/*! + * @private + * @brief The selected dispatch table for @ref XXH3_64bits(). + */ +static XXH_dispatchFunctions_s XXH_g_dispatch = { NULL, NULL, NULL, NULL }; + + +/*! @cond Doxygen ignores this part */ +typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_default)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t); + +typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_withSeed)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t, XXH64_hash_t); + +typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_withSecret)(XXH_NOESCAPE const void* XXH_RESTRICT, size_t, XXH_NOESCAPE const void* XXH_RESTRICT, size_t); + +typedef struct { + XXH3_dispatchx86_hashLong128_default hashLong128_default; + XXH3_dispatchx86_hashLong128_withSeed hashLong128_seed; + XXH3_dispatchx86_hashLong128_withSecret hashLong128_secret; + XXH3_dispatchx86_update update; +} XXH_dispatch128Functions_s; +/*! @endcond */ + + +/*! + * @private + * @brief Table of dispatchers for @ref XXH3_128bits(). + * + * @pre The indices must match @ref XXH_VECTOR_TYPE. + */ +static const XXH_dispatch128Functions_s XXH_kDispatch128[XXH_NB_DISPATCHES] = { +#if XXH_DISPATCH_SCALAR + /* Scalar */ { XXHL128_default_scalar, XXHL128_seed_scalar, XXHL128_secret_scalar, XXH3_update_scalar }, +#else + /* Scalar */ { NULL, NULL, NULL, NULL }, +#endif + /* SSE2 */ { XXHL128_default_sse2, XXHL128_seed_sse2, XXHL128_secret_sse2, XXH3_update_sse2 }, +#if XXH_DISPATCH_AVX2 + /* AVX2 */ { XXHL128_default_avx2, XXHL128_seed_avx2, XXHL128_secret_avx2, XXH3_update_avx2 }, +#else + /* AVX2 */ { NULL, NULL, NULL, NULL }, +#endif +#if XXH_DISPATCH_AVX512 + /* AVX512 */ { XXHL128_default_avx512, XXHL128_seed_avx512, XXHL128_secret_avx512, XXH3_update_avx512 } +#else + /* AVX512 */ { NULL, NULL, NULL, NULL } +#endif +}; + +/*! + * @private + * @brief The selected dispatch table for @ref XXH3_64bits(). + */ +static XXH_dispatch128Functions_s XXH_g_dispatch128 = { NULL, NULL, NULL, NULL }; + +/*! + * @private + * @brief Runs a CPUID check and sets the correct dispatch tables. + */ +static XXH_CONSTRUCTOR void XXH_setDispatch(void) +{ + int vecID = XXH_featureTest(); + XXH_STATIC_ASSERT(XXH_AVX512 == XXH_NB_DISPATCHES-1); + assert(XXH_SCALAR <= vecID && vecID <= XXH_AVX512); +#if !XXH_DISPATCH_SCALAR + assert(vecID != XXH_SCALAR); +#endif +#if !XXH_DISPATCH_AVX512 + assert(vecID != XXH_AVX512); +#endif +#if !XXH_DISPATCH_AVX2 + assert(vecID != XXH_AVX2); +#endif + XXH_g_dispatch = XXH_kDispatch[vecID]; + XXH_g_dispatch128 = XXH_kDispatch128[vecID]; +} + + +/* ==== XXH3 public functions ==== */ +/*! @cond Doxygen ignores this part */ + +static XXH64_hash_t +XXH3_hashLong_64b_defaultSecret_selection(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch.hashLong64_default == NULL) + XXH_setDispatch(); + return XXH_g_dispatch.hashLong64_default(input, len); +} + +XXH64_hash_t XXH3_64bits_dispatch(XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_defaultSecret_selection); +} + +static XXH64_hash_t +XXH3_hashLong_64b_withSeed_selection(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch.hashLong64_seed == NULL) + XXH_setDispatch(); + return XXH_g_dispatch.hashLong64_seed(input, len, seed64); +} + +XXH64_hash_t XXH3_64bits_withSeed_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed_selection); +} + +static XXH64_hash_t +XXH3_hashLong_64b_withSecret_selection(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch.hashLong64_secret == NULL) + XXH_setDispatch(); + return XXH_g_dispatch.hashLong64_secret(input, len, secret, secretLen); +} + +XXH64_hash_t XXH3_64bits_withSecret_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretLen) +{ + return XXH3_64bits_internal(input, len, 0, secret, secretLen, XXH3_hashLong_64b_withSecret_selection); +} + +XXH_errorcode +XXH3_64bits_update_dispatch(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch.update == NULL) + XXH_setDispatch(); + + return XXH_g_dispatch.update(state, (const xxh_u8*)input, len); +} + +/*! @endcond */ + + +/* ==== XXH128 public functions ==== */ +/*! @cond Doxygen ignores this part */ + +static XXH128_hash_t +XXH3_hashLong_128b_defaultSecret_selection(const void* input, size_t len, + XXH64_hash_t seed64, const void* secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch128.hashLong128_default == NULL) + XXH_setDispatch(); + return XXH_g_dispatch128.hashLong128_default(input, len); +} + +XXH128_hash_t XXH3_128bits_dispatch(XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_128bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_128b_defaultSecret_selection); +} + +static XXH128_hash_t +XXH3_hashLong_128b_withSeed_selection(const void* input, size_t len, + XXH64_hash_t seed64, const void* secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch128.hashLong128_seed == NULL) + XXH_setDispatch(); + return XXH_g_dispatch128.hashLong128_seed(input, len, seed64); +} + +XXH128_hash_t XXH3_128bits_withSeed_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_128b_withSeed_selection); +} + +static XXH128_hash_t +XXH3_hashLong_128b_withSecret_selection(const void* input, size_t len, + XXH64_hash_t seed64, const void* secret, size_t secretLen) +{ + (void)seed64; + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch128.hashLong128_secret == NULL) + XXH_setDispatch(); + return XXH_g_dispatch128.hashLong128_secret(input, len, secret, secretLen); +} + +XXH128_hash_t XXH3_128bits_withSecret_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretLen) +{ + return XXH3_128bits_internal(input, len, 0, secret, secretLen, XXH3_hashLong_128b_withSecret_selection); +} + +XXH_errorcode +XXH3_128bits_update_dispatch(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + if (XXH_DISPATCH_MAYBE_NULL && XXH_g_dispatch128.update == NULL) + XXH_setDispatch(); + return XXH_g_dispatch128.update(state, (const xxh_u8*)input, len); +} + +/*! @endcond */ + +#if defined (__cplusplus) +} +#endif +/*! @} */ diff --git a/external/xxhash/xxh_x86dispatch.h b/external/xxhash/xxh_x86dispatch.h new file mode 100644 index 0000000..7085221 --- /dev/null +++ b/external/xxhash/xxh_x86dispatch.h @@ -0,0 +1,93 @@ +/* + * xxHash - XXH3 Dispatcher for x86-based targets + * Copyright (C) 2020-2024 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + +#ifndef XXH_X86DISPATCH_H_13563687684 +#define XXH_X86DISPATCH_H_13563687684 + +#include "xxhash.h" /* XXH64_hash_t, XXH3_state_t */ + +#if defined (__cplusplus) +extern "C" { +#endif + +/*! + * @brief Returns the best XXH3 implementation for x86 + * + * @return The best @ref XXH_VECTOR implementation. + * @see XXH_VECTOR_TYPES + */ +XXH_PUBLIC_API int XXH_featureTest(void); + +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_dispatch(XXH_NOESCAPE const void* input, size_t len); +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed); +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretLen); +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update_dispatch(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len); + +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_dispatch(XXH_NOESCAPE const void* input, size_t len); +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed); +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret_dispatch(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretLen); +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update_dispatch(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len); + +#if defined (__cplusplus) +} +#endif + + +/* automatic replacement of XXH3 functions. + * can be disabled by setting XXH_DISPATCH_DISABLE_REPLACE */ +#ifndef XXH_DISPATCH_DISABLE_REPLACE + +# undef XXH3_64bits +# define XXH3_64bits XXH3_64bits_dispatch +# undef XXH3_64bits_withSeed +# define XXH3_64bits_withSeed XXH3_64bits_withSeed_dispatch +# undef XXH3_64bits_withSecret +# define XXH3_64bits_withSecret XXH3_64bits_withSecret_dispatch +# undef XXH3_64bits_update +# define XXH3_64bits_update XXH3_64bits_update_dispatch + +# undef XXH128 +# define XXH128 XXH3_128bits_withSeed_dispatch +# undef XXH3_128bits +# define XXH3_128bits XXH3_128bits_dispatch +# undef XXH3_128bits_withSeed +# define XXH3_128bits_withSeed XXH3_128bits_withSeed_dispatch +# undef XXH3_128bits_withSecret +# define XXH3_128bits_withSecret XXH3_128bits_withSecret_dispatch +# undef XXH3_128bits_update +# define XXH3_128bits_update XXH3_128bits_update_dispatch + +#endif /* XXH_DISPATCH_DISABLE_REPLACE */ + +#endif /* XXH_X86DISPATCH_H_13563687684 */ diff --git a/external/xxhash/xxhash.c b/external/xxhash/xxhash.c new file mode 100644 index 0000000..e60cc37 --- /dev/null +++ b/external/xxhash/xxhash.c @@ -0,0 +1,42 @@ +/* + * xxHash - Extremely Fast Hash algorithm + * Copyright (C) 2012-2023 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + +/* + * xxhash.c instantiates functions defined in xxhash.h + */ + +#define XXH_STATIC_LINKING_ONLY /* access advanced declarations */ +#define XXH_IMPLEMENTATION /* access definitions */ + +#include "xxhash.h" diff --git a/external/xxhash/xxhash.h b/external/xxhash/xxhash.h new file mode 100644 index 0000000..c4cb3ba --- /dev/null +++ b/external/xxhash/xxhash.h @@ -0,0 +1,7343 @@ +/* + * xxHash - Extremely Fast Hash algorithm + * Header File + * Copyright (C) 2012-2023 Yann Collet + * + * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php) + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following disclaimer + * in the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * You can contact the author at: + * - xxHash homepage: https://www.xxhash.com + * - xxHash source repository: https://github.com/Cyan4973/xxHash + */ + +/*! + * @mainpage xxHash + * + * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed + * limits. + * + * It is proposed in four flavors, in three families: + * 1. @ref XXH32_family + * - Classic 32-bit hash function. Simple, compact, and runs on almost all + * 32-bit and 64-bit systems. + * 2. @ref XXH64_family + * - Classic 64-bit adaptation of XXH32. Just as simple, and runs well on most + * 64-bit systems (but _not_ 32-bit systems). + * 3. @ref XXH3_family + * - Modern 64-bit and 128-bit hash function family which features improved + * strength and performance across the board, especially on smaller data. + * It benefits greatly from SIMD and 64-bit without requiring it. + * + * Benchmarks + * --- + * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04. + * The open source benchmark program is compiled with clang v10.0 using -O3 flag. + * + * | Hash Name | ISA ext | Width | Large Data Speed | Small Data Velocity | + * | -------------------- | ------- | ----: | ---------------: | ------------------: | + * | XXH3_64bits() | @b AVX2 | 64 | 59.4 GB/s | 133.1 | + * | MeowHash | AES-NI | 128 | 58.2 GB/s | 52.5 | + * | XXH3_128bits() | @b AVX2 | 128 | 57.9 GB/s | 118.1 | + * | CLHash | PCLMUL | 64 | 37.1 GB/s | 58.1 | + * | XXH3_64bits() | @b SSE2 | 64 | 31.5 GB/s | 133.1 | + * | XXH3_128bits() | @b SSE2 | 128 | 29.6 GB/s | 118.1 | + * | RAM sequential read | | N/A | 28.0 GB/s | N/A | + * | ahash | AES-NI | 64 | 22.5 GB/s | 107.2 | + * | City64 | | 64 | 22.0 GB/s | 76.6 | + * | T1ha2 | | 64 | 22.0 GB/s | 99.0 | + * | City128 | | 128 | 21.7 GB/s | 57.7 | + * | FarmHash | AES-NI | 64 | 21.3 GB/s | 71.9 | + * | XXH64() | | 64 | 19.4 GB/s | 71.0 | + * | SpookyHash | | 64 | 19.3 GB/s | 53.2 | + * | Mum | | 64 | 18.0 GB/s | 67.0 | + * | CRC32C | SSE4.2 | 32 | 13.0 GB/s | 57.9 | + * | XXH32() | | 32 | 9.7 GB/s | 71.9 | + * | City32 | | 32 | 9.1 GB/s | 66.0 | + * | Blake3* | @b AVX2 | 256 | 4.4 GB/s | 8.1 | + * | Murmur3 | | 32 | 3.9 GB/s | 56.1 | + * | SipHash* | | 64 | 3.0 GB/s | 43.2 | + * | Blake3* | @b SSE2 | 256 | 2.4 GB/s | 8.1 | + * | HighwayHash | | 64 | 1.4 GB/s | 6.0 | + * | FNV64 | | 64 | 1.2 GB/s | 62.7 | + * | Blake2* | | 256 | 1.1 GB/s | 5.1 | + * | SHA1* | | 160 | 0.8 GB/s | 5.6 | + * | MD5* | | 128 | 0.6 GB/s | 7.8 | + * @note + * - Hashes which require a specific ISA extension are noted. SSE2 is also noted, + * even though it is mandatory on x64. + * - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic + * by modern standards. + * - Small data velocity is a rough average of algorithm's efficiency for small + * data. For more accurate information, see the wiki. + * - More benchmarks and strength tests are found on the wiki: + * https://github.com/Cyan4973/xxHash/wiki + * + * Usage + * ------ + * All xxHash variants use a similar API. Changing the algorithm is a trivial + * substitution. + * + * @pre + * For functions which take an input and length parameter, the following + * requirements are assumed: + * - The range from [`input`, `input + length`) is valid, readable memory. + * - The only exception is if the `length` is `0`, `input` may be `NULL`. + * - For C++, the objects must have the *TriviallyCopyable* property, as the + * functions access bytes directly as if it was an array of `unsigned char`. + * + * @anchor single_shot_example + * **Single Shot** + * + * These functions are stateless functions which hash a contiguous block of memory, + * immediately returning the result. They are the easiest and usually the fastest + * option. + * + * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits() + * + * @code{.c} + * #include + * #include "xxhash.h" + * + * // Example for a function which hashes a null terminated string with XXH32(). + * XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed) + * { + * // NULL pointers are only valid if the length is zero + * size_t length = (string == NULL) ? 0 : strlen(string); + * return XXH32(string, length, seed); + * } + * @endcode + * + * + * @anchor streaming_example + * **Streaming** + * + * These groups of functions allow incremental hashing of unknown size, even + * more than what would fit in a size_t. + * + * XXH32_reset(), XXH64_reset(), XXH3_64bits_reset(), XXH3_128bits_reset() + * + * @code{.c} + * #include + * #include + * #include "xxhash.h" + * // Example for a function which hashes a FILE incrementally with XXH3_64bits(). + * XXH64_hash_t hashFile(FILE* f) + * { + * // Allocate a state struct. Do not just use malloc() or new. + * XXH3_state_t* state = XXH3_createState(); + * assert(state != NULL && "Out of memory!"); + * // Reset the state to start a new hashing session. + * XXH3_64bits_reset(state); + * char buffer[4096]; + * size_t count; + * // Read the file in chunks + * while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) { + * // Run update() as many times as necessary to process the data + * XXH3_64bits_update(state, buffer, count); + * } + * // Retrieve the finalized hash. This will not change the state. + * XXH64_hash_t result = XXH3_64bits_digest(state); + * // Free the state. Do not use free(). + * XXH3_freeState(state); + * return result; + * } + * @endcode + * + * Streaming functions generate the xxHash value from an incremental input. + * This method is slower than single-call functions, due to state management. + * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized. + * + * An XXH state must first be allocated using `XXH*_createState()`. + * + * Start a new hash by initializing the state with a seed using `XXH*_reset()`. + * + * Then, feed the hash state by calling `XXH*_update()` as many times as necessary. + * + * The function returns an error code, with 0 meaning OK, and any other value + * meaning there is an error. + * + * Finally, a hash value can be produced anytime, by using `XXH*_digest()`. + * This function returns the nn-bits hash as an int or long long. + * + * It's still possible to continue inserting input into the hash state after a + * digest, and generate new hash values later on by invoking `XXH*_digest()`. + * + * When done, release the state using `XXH*_freeState()`. + * + * + * @anchor canonical_representation_example + * **Canonical Representation** + * + * The default return values from XXH functions are unsigned 32, 64 and 128 bit + * integers. + * This the simplest and fastest format for further post-processing. + * + * However, this leaves open the question of what is the order on the byte level, + * since little and big endian conventions will store the same number differently. + * + * The canonical representation settles this issue by mandating big-endian + * convention, the same convention as human-readable numbers (large digits first). + * + * When writing hash values to storage, sending them over a network, or printing + * them, it's highly recommended to use the canonical representation to ensure + * portability across a wider range of systems, present and future. + * + * The following functions allow transformation of hash values to and from + * canonical format. + * + * XXH32_canonicalFromHash(), XXH32_hashFromCanonical(), + * XXH64_canonicalFromHash(), XXH64_hashFromCanonical(), + * XXH128_canonicalFromHash(), XXH128_hashFromCanonical(), + * + * @code{.c} + * #include + * #include "xxhash.h" + * + * // Example for a function which prints XXH32_hash_t in human readable format + * void printXxh32(XXH32_hash_t hash) + * { + * XXH32_canonical_t cano; + * XXH32_canonicalFromHash(&cano, hash); + * size_t i; + * for(i = 0; i < sizeof(cano.digest); ++i) { + * printf("%02x", cano.digest[i]); + * } + * printf("\n"); + * } + * + * // Example for a function which converts XXH32_canonical_t to XXH32_hash_t + * XXH32_hash_t convertCanonicalToXxh32(XXH32_canonical_t cano) + * { + * XXH32_hash_t hash = XXH32_hashFromCanonical(&cano); + * return hash; + * } + * @endcode + * + * + * @file xxhash.h + * xxHash prototypes and implementation + */ + +#if defined(__cplusplus) && !defined(XXH_NO_EXTERNC_GUARD) +extern "C" { +#endif + +/* **************************** + * INLINE mode + ******************************/ +/*! + * @defgroup public Public API + * Contains details on the public xxHash functions. + * @{ + */ +#ifdef XXH_DOXYGEN +/*! + * @brief Gives access to internal state declaration, required for static allocation. + * + * Incompatible with dynamic linking, due to risks of ABI changes. + * + * Usage: + * @code{.c} + * #define XXH_STATIC_LINKING_ONLY + * #include "xxhash.h" + * @endcode + */ +# define XXH_STATIC_LINKING_ONLY +/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */ + +/*! + * @brief Gives access to internal definitions. + * + * Usage: + * @code{.c} + * #define XXH_STATIC_LINKING_ONLY + * #define XXH_IMPLEMENTATION + * #include "xxhash.h" + * @endcode + */ +# define XXH_IMPLEMENTATION +/* Do not undef XXH_IMPLEMENTATION for Doxygen */ + +/*! + * @brief Exposes the implementation and marks all functions as `inline`. + * + * Use these build macros to inline xxhash into the target unit. + * Inlining improves performance on small inputs, especially when the length is + * expressed as a compile-time constant: + * + * https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html + * + * It also keeps xxHash symbols private to the unit, so they are not exported. + * + * Usage: + * @code{.c} + * #define XXH_INLINE_ALL + * #include "xxhash.h" + * @endcode + * Do not compile and link xxhash.o as a separate object, as it is not useful. + */ +# define XXH_INLINE_ALL +# undef XXH_INLINE_ALL +/*! + * @brief Exposes the implementation without marking functions as inline. + */ +# define XXH_PRIVATE_API +# undef XXH_PRIVATE_API +/*! + * @brief Emulate a namespace by transparently prefixing all symbols. + * + * If you want to include _and expose_ xxHash functions from within your own + * library, but also want to avoid symbol collisions with other libraries which + * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix + * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE + * (therefore, avoid empty or numeric values). + * + * Note that no change is required within the calling program as long as it + * includes `xxhash.h`: Regular symbol names will be automatically translated + * by this header. + */ +# define XXH_NAMESPACE /* YOUR NAME HERE */ +# undef XXH_NAMESPACE +#endif + +#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \ + && !defined(XXH_INLINE_ALL_31684351384) + /* this section should be traversed only once */ +# define XXH_INLINE_ALL_31684351384 + /* give access to the advanced API, required to compile implementations */ +# undef XXH_STATIC_LINKING_ONLY /* avoid macro redef */ +# define XXH_STATIC_LINKING_ONLY + /* make all functions private */ +# undef XXH_PUBLIC_API +# if defined(__GNUC__) +# define XXH_PUBLIC_API static __inline __attribute__((__unused__)) +# elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) +# define XXH_PUBLIC_API static inline +# elif defined(_MSC_VER) +# define XXH_PUBLIC_API static __inline +# else + /* note: this version may generate warnings for unused static functions */ +# define XXH_PUBLIC_API static +# endif + + /* + * This part deals with the special case where a unit wants to inline xxHash, + * but "xxhash.h" has previously been included without XXH_INLINE_ALL, + * such as part of some previously included *.h header file. + * Without further action, the new include would just be ignored, + * and functions would effectively _not_ be inlined (silent failure). + * The following macros solve this situation by prefixing all inlined names, + * avoiding naming collision with previous inclusions. + */ + /* Before that, we unconditionally #undef all symbols, + * in case they were already defined with XXH_NAMESPACE. + * They will then be redefined for XXH_INLINE_ALL + */ +# undef XXH_versionNumber + /* XXH32 */ +# undef XXH32 +# undef XXH32_createState +# undef XXH32_freeState +# undef XXH32_reset +# undef XXH32_update +# undef XXH32_digest +# undef XXH32_copyState +# undef XXH32_canonicalFromHash +# undef XXH32_hashFromCanonical + /* XXH64 */ +# undef XXH64 +# undef XXH64_createState +# undef XXH64_freeState +# undef XXH64_reset +# undef XXH64_update +# undef XXH64_digest +# undef XXH64_copyState +# undef XXH64_canonicalFromHash +# undef XXH64_hashFromCanonical + /* XXH3_64bits */ +# undef XXH3_64bits +# undef XXH3_64bits_withSecret +# undef XXH3_64bits_withSeed +# undef XXH3_64bits_withSecretandSeed +# undef XXH3_createState +# undef XXH3_freeState +# undef XXH3_copyState +# undef XXH3_64bits_reset +# undef XXH3_64bits_reset_withSeed +# undef XXH3_64bits_reset_withSecret +# undef XXH3_64bits_update +# undef XXH3_64bits_digest +# undef XXH3_generateSecret + /* XXH3_128bits */ +# undef XXH128 +# undef XXH3_128bits +# undef XXH3_128bits_withSeed +# undef XXH3_128bits_withSecret +# undef XXH3_128bits_reset +# undef XXH3_128bits_reset_withSeed +# undef XXH3_128bits_reset_withSecret +# undef XXH3_128bits_reset_withSecretandSeed +# undef XXH3_128bits_update +# undef XXH3_128bits_digest +# undef XXH128_isEqual +# undef XXH128_cmp +# undef XXH128_canonicalFromHash +# undef XXH128_hashFromCanonical + /* Finally, free the namespace itself */ +# undef XXH_NAMESPACE + + /* employ the namespace for XXH_INLINE_ALL */ +# define XXH_NAMESPACE XXH_INLINE_ + /* + * Some identifiers (enums, type names) are not symbols, + * but they must nonetheless be renamed to avoid redeclaration. + * Alternative solution: do not redeclare them. + * However, this requires some #ifdefs, and has a more dispersed impact. + * Meanwhile, renaming can be achieved in a single place. + */ +# define XXH_IPREF(Id) XXH_NAMESPACE ## Id +# define XXH_OK XXH_IPREF(XXH_OK) +# define XXH_ERROR XXH_IPREF(XXH_ERROR) +# define XXH_errorcode XXH_IPREF(XXH_errorcode) +# define XXH32_canonical_t XXH_IPREF(XXH32_canonical_t) +# define XXH64_canonical_t XXH_IPREF(XXH64_canonical_t) +# define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t) +# define XXH32_state_s XXH_IPREF(XXH32_state_s) +# define XXH32_state_t XXH_IPREF(XXH32_state_t) +# define XXH64_state_s XXH_IPREF(XXH64_state_s) +# define XXH64_state_t XXH_IPREF(XXH64_state_t) +# define XXH3_state_s XXH_IPREF(XXH3_state_s) +# define XXH3_state_t XXH_IPREF(XXH3_state_t) +# define XXH128_hash_t XXH_IPREF(XXH128_hash_t) + /* Ensure the header is parsed again, even if it was previously included */ +# undef XXHASH_H_5627135585666179 +# undef XXHASH_H_STATIC_13879238742 +#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */ + +/* **************************************************************** + * Stable API + *****************************************************************/ +#ifndef XXHASH_H_5627135585666179 +#define XXHASH_H_5627135585666179 1 + +/*! @brief Marks a global symbol. */ +#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API) +# if defined(_WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT)) +# ifdef XXH_EXPORT +# define XXH_PUBLIC_API __declspec(dllexport) +# elif XXH_IMPORT +# define XXH_PUBLIC_API __declspec(dllimport) +# endif +# else +# define XXH_PUBLIC_API /* do nothing */ +# endif +#endif + +#ifdef XXH_NAMESPACE +# define XXH_CAT(A,B) A##B +# define XXH_NAME2(A,B) XXH_CAT(A,B) +# define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber) +/* XXH32 */ +# define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32) +# define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState) +# define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState) +# define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset) +# define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update) +# define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest) +# define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState) +# define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash) +# define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical) +/* XXH64 */ +# define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64) +# define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState) +# define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState) +# define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset) +# define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update) +# define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest) +# define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState) +# define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash) +# define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical) +/* XXH3_64bits */ +# define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits) +# define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret) +# define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed) +# define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed) +# define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState) +# define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState) +# define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState) +# define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset) +# define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed) +# define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret) +# define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed) +# define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update) +# define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest) +# define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret) +# define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed) +/* XXH3_128bits */ +# define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128) +# define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits) +# define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed) +# define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret) +# define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed) +# define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset) +# define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed) +# define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret) +# define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed) +# define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update) +# define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest) +# define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual) +# define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp) +# define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash) +# define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical) +#endif + + +/* ************************************* +* Compiler specifics +***************************************/ + +/* specific declaration modes for Windows */ +#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API) +# if defined(_WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT)) +# ifdef XXH_EXPORT +# define XXH_PUBLIC_API __declspec(dllexport) +# elif XXH_IMPORT +# define XXH_PUBLIC_API __declspec(dllimport) +# endif +# else +# define XXH_PUBLIC_API /* do nothing */ +# endif +#endif + +#if defined (__GNUC__) +# define XXH_CONSTF __attribute__((__const__)) +# define XXH_PUREF __attribute__((__pure__)) +# define XXH_MALLOCF __attribute__((__malloc__)) +#else +# define XXH_CONSTF /* disable */ +# define XXH_PUREF +# define XXH_MALLOCF +#endif + +/* ************************************* +* Version +***************************************/ +#define XXH_VERSION_MAJOR 0 +#define XXH_VERSION_MINOR 8 +#define XXH_VERSION_RELEASE 3 +/*! @brief Version number, encoded as two digits each */ +#define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE) + +/*! + * @brief Obtains the xxHash version. + * + * This is mostly useful when xxHash is compiled as a shared library, + * since the returned value comes from the library, as opposed to header file. + * + * @return @ref XXH_VERSION_NUMBER of the invoked library. + */ +XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void); + + +/* **************************** +* Common basic types +******************************/ +#include /* size_t */ +/*! + * @brief Exit code for the streaming API. + */ +typedef enum { + XXH_OK = 0, /*!< OK */ + XXH_ERROR /*!< Error */ +} XXH_errorcode; + + +/*-********************************************************************** +* 32-bit hash +************************************************************************/ +#if defined(XXH_DOXYGEN) /* Don't show include */ +/*! + * @brief An unsigned 32-bit integer. + * + * Not necessarily defined to `uint32_t` but functionally equivalent. + */ +typedef uint32_t XXH32_hash_t; + +#elif !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint32_t XXH32_hash_t; + +#else +# include +# if UINT_MAX == 0xFFFFFFFFUL + typedef unsigned int XXH32_hash_t; +# elif ULONG_MAX == 0xFFFFFFFFUL + typedef unsigned long XXH32_hash_t; +# else +# error "unsupported platform: need a 32-bit type" +# endif +#endif + +/*! + * @} + * + * @defgroup XXH32_family XXH32 family + * @ingroup public + * Contains functions used in the classic 32-bit xxHash algorithm. + * + * @note + * XXH32 is useful for older platforms, with no or poor 64-bit performance. + * Note that the @ref XXH3_family provides competitive speed for both 32-bit + * and 64-bit systems, and offers true 64/128 bit hash results. + * + * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families + * @see @ref XXH32_impl for implementation details + * @{ + */ + +/*! + * @brief Calculates the 32-bit hash of @p input using xxHash32. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 32-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 32-bit xxHash32 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed); + +#ifndef XXH_NO_STREAM +/*! + * @typedef struct XXH32_state_s XXH32_state_t + * @brief The opaque state struct for the XXH32 streaming API. + * + * @see XXH32_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH32_state_s XXH32_state_t; + +/*! + * @brief Allocates an @ref XXH32_state_t. + * + * @return An allocated pointer of @ref XXH32_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH32_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void); +/*! + * @brief Frees an @ref XXH32_state_t. + * + * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState(). + * + * @return @ref XXH_OK. + * + * @note @p statePtr must be allocated with XXH32_createState(). + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr); +/*! + * @brief Copies one @ref XXH32_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state); + +/*! + * @brief Resets an @ref XXH32_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 32-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note This function resets and seeds a state. Call it before @ref XXH32_update(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH32_reset (XXH32_state_t* statePtr, XXH32_hash_t seed); + +/*! + * @brief Consumes a block of @p input to an @ref XXH32_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length); + +/*! + * @brief Returns the calculated hash value from an @ref XXH32_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated 32-bit xxHash32 value from that state. + * + * @note + * Calling XXH32_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! + * @brief Canonical (big endian) representation of @ref XXH32_hash_t. + */ +typedef struct { + unsigned char digest[4]; /*!< Hash bytes, big endian */ +} XXH32_canonical_t; + +/*! + * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t. + * + * @param dst The @ref XXH32_canonical_t pointer to be stored to. + * @param hash The @ref XXH32_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash); + +/*! + * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t. + * + * @param src The @ref XXH32_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src); + + +/*! @cond Doxygen ignores this part */ +#ifdef __has_attribute +# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x) +#else +# define XXH_HAS_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * C23 __STDC_VERSION__ number hasn't been specified yet. For now + * leave as `201711L` (C17 + 1). + * TODO: Update to correct value when its been specified. + */ +#define XXH_C23_VN 201711L +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* C-language Attributes are added in C23. */ +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute) +# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x) +#else +# define XXH_HAS_C_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +#if defined(__cplusplus) && defined(__has_cpp_attribute) +# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x) +#else +# define XXH_HAS_CPP_ATTRIBUTE(x) 0 +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute + * introduced in CPP17 and C23. + * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough + * C23 : https://en.cppreference.com/w/c/language/attributes/fallthrough + */ +#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough) +# define XXH_FALLTHROUGH [[fallthrough]] +#elif XXH_HAS_ATTRIBUTE(__fallthrough__) +# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__)) +#else +# define XXH_FALLTHROUGH /* fallthrough */ +#endif +/*! @endcond */ + +/*! @cond Doxygen ignores this part */ +/* + * Define XXH_NOESCAPE for annotated pointers in public API. + * https://clang.llvm.org/docs/AttributeReference.html#noescape + * As of writing this, only supported by clang. + */ +#if XXH_HAS_ATTRIBUTE(noescape) +# define XXH_NOESCAPE __attribute__((__noescape__)) +#else +# define XXH_NOESCAPE +#endif +/*! @endcond */ + + +/*! + * @} + * @ingroup public + * @{ + */ + +#ifndef XXH_NO_LONG_LONG +/*-********************************************************************** +* 64-bit hash +************************************************************************/ +#if defined(XXH_DOXYGEN) /* don't include */ +/*! + * @brief An unsigned 64-bit integer. + * + * Not necessarily defined to `uint64_t` but functionally equivalent. + */ +typedef uint64_t XXH64_hash_t; +#elif !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint64_t XXH64_hash_t; +#else +# include +# if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL + /* LP64 ABI says uint64_t is unsigned long */ + typedef unsigned long XXH64_hash_t; +# else + /* the following type must have a width of 64-bit */ + typedef unsigned long long XXH64_hash_t; +# endif +#endif + +/*! + * @} + * + * @defgroup XXH64_family XXH64 family + * @ingroup public + * @{ + * Contains functions used in the classic 64-bit xxHash algorithm. + * + * @note + * XXH3 provides competitive speed for both 32-bit and 64-bit systems, + * and offers true 64/128 bit hash results. + * It provides better speed for systems with vector processing capabilities. + */ + +/*! + * @brief Calculates the 64-bit hash of @p input using xxHash64. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 64-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit xxHash64 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed); + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/*! + * @brief The opaque state struct for the XXH64 streaming API. + * + * @see XXH64_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */ + +/*! + * @brief Allocates an @ref XXH64_state_t. + * + * @return An allocated pointer of @ref XXH64_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH64_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void); + +/*! + * @brief Frees an @ref XXH64_state_t. + * + * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState(). + * + * @return @ref XXH_OK. + * + * @note @p statePtr must be allocated with XXH64_createState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr); + +/*! + * @brief Copies one @ref XXH64_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state); + +/*! + * @brief Resets an @ref XXH64_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note This function resets and seeds a state. Call it before @ref XXH64_update(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed); + +/*! + * @brief Consumes a block of @p input to an @ref XXH64_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated hash value from an @ref XXH64_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated 64-bit xxHash64 value from that state. + * + * @note + * Calling XXH64_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ +/******* Canonical representation *******/ + +/*! + * @brief Canonical (big endian) representation of @ref XXH64_hash_t. + */ +typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t; + +/*! + * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t. + * + * @param dst The @ref XXH64_canonical_t pointer to be stored to. + * @param hash The @ref XXH64_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash); + +/*! + * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t. + * + * @param src The @ref XXH64_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src); + +#ifndef XXH_NO_XXH3 + +/*! + * @} + * ************************************************************************ + * @defgroup XXH3_family XXH3 family + * @ingroup public + * @{ + * + * XXH3 is a more recent hash algorithm featuring: + * - Improved speed for both small and large inputs + * - True 64-bit and 128-bit outputs + * - SIMD acceleration + * - Improved 32-bit viability + * + * Speed analysis methodology is explained here: + * + * https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html + * + * Compared to XXH64, expect XXH3 to run approximately + * ~2x faster on large inputs and >3x faster on small ones, + * exact differences vary depending on platform. + * + * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic, + * but does not require it. + * Most 32-bit and 64-bit targets that can run XXH32 smoothly can run XXH3 + * at competitive speeds, even without vector support. Further details are + * explained in the implementation. + * + * XXH3 has a fast scalar implementation, but it also includes accelerated SIMD + * implementations for many common platforms: + * - AVX512 + * - AVX2 + * - SSE2 + * - ARM NEON + * - WebAssembly SIMD128 + * - POWER8 VSX + * - s390x ZVector + * This can be controlled via the @ref XXH_VECTOR macro, but it automatically + * selects the best version according to predefined macros. For the x86 family, an + * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c. + * + * XXH3 implementation is portable: + * it has a generic C90 formulation that can be compiled on any platform, + * all implementations generate exactly the same hash value on all platforms. + * Starting from v0.8.0, it's also labelled "stable", meaning that + * any future version will also generate the same hash value. + * + * XXH3 offers 2 variants, _64bits and _128bits. + * + * When only 64 bits are needed, prefer invoking the _64bits variant, as it + * reduces the amount of mixing, resulting in faster speed on small inputs. + * It's also generally simpler to manipulate a scalar return type than a struct. + * + * The API supports one-shot hashing, streaming mode, and custom secrets. + */ + +/*! + * @ingroup tuning + * @brief Possible values for @ref XXH_VECTOR. + * + * Unless set explicitly, determined automatically. + */ +# define XXH_SCALAR 0 /*!< Portable scalar version */ +# define XXH_SSE2 1 /*!< SSE2 for Pentium 4, Opteron, all x86_64. */ +# define XXH_AVX2 2 /*!< AVX2 for Haswell and Bulldozer */ +# define XXH_AVX512 3 /*!< AVX512 for Skylake and Icelake */ +# define XXH_NEON 4 /*!< NEON for most ARMv7-A, all AArch64, and WASM SIMD128 */ +# define XXH_VSX 5 /*!< VSX and ZVector for POWER8/z13 (64-bit) */ +# define XXH_SVE 6 /*!< SVE for some ARMv8-A and ARMv9-A */ +# define XXH_LSX 7 /*!< LSX (128-bit SIMD) for LoongArch64 */ +# define XXH_LASX 8 /*!< LASX (256-bit SIMD) for LoongArch64 */ + + +/*-********************************************************************** +* XXH3 64-bit variant +************************************************************************/ + +/*! + * @brief Calculates 64-bit unseeded variant of XXH3 hash of @p input. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @note + * This is equivalent to @ref XXH3_64bits_withSeed() with a seed of `0`, however + * it may have slightly better performance due to constant propagation of the + * defaults. + * + * @see + * XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Calculates 64-bit seeded variant of XXH3 hash of @p input. + * + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @note + * seed == 0 produces the same results as @ref XXH3_64bits(). + * + * This variant generates a custom secret on the fly based on default secret + * altered using the @p seed value. + * + * While this operation is decently fast, note that it's not completely free. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed); + +/*! + * The bare minimum size for a custom secret. + * + * @see + * XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(), + * XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret(). + */ +#define XXH3_SECRET_SIZE_MIN 136 + +/*! + * @brief Calculates 64-bit variant of XXH3 with a custom "secret". + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @return The calculated 64-bit XXH3 hash value. + * + * @pre + * The memory between @p data and @p data + @p len must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p data may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * It's possible to provide any blob of bytes as a "secret" to generate the hash. + * This makes it more difficult for an external actor to prepare an intentional collision. + * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN). + * However, the quality of the secret impacts the dispersion of the hash algorithm. + * Therefore, the secret _must_ look like a bunch of random bytes. + * Avoid "trivial" or structured data such as repeated sequences or a text document. + * Whenever in doubt about the "randomness" of the blob of bytes, + * consider employing @ref XXH3_generateSecret() instead (see below). + * It will generate a proper high entropy secret derived from the blob of bytes. + * Another advantage of using XXH3_generateSecret() is that + * it guarantees that all bits within the initial blob of bytes + * will impact every bit of the output. + * This is not necessarily the case when using the blob of bytes directly + * because, when hashing _small_ inputs, only a portion of the secret is employed. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize); + + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/* + * Streaming requires state maintenance. + * This operation costs memory and CPU. + * As a consequence, streaming is slower than one-shot hashing. + * For better performance, prefer one-shot functions whenever applicable. + */ + +/*! + * @brief The opaque state struct for the XXH3 streaming API. + * + * @see XXH3_state_s for details. + * @see @ref streaming_example "Streaming Example" + */ +typedef struct XXH3_state_s XXH3_state_t; +XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void); +XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr); + +/*! + * @brief Copies one @ref XXH3_state_t to another. + * + * @param dst_state The state to copy to. + * @param src_state The state to copy from. + * @pre + * @p dst_state and @p src_state must not be `NULL` and must not overlap. + */ +XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state); + +/*! + * @brief Resets an @ref XXH3_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret with default parameters. + * - Call this function before @ref XXH3_64bits_update(). + * - Digest will be equivalent to `XXH3_64bits()`. + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr); + +/*! + * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret from `seed`. + * - Call this function before @ref XXH3_64bits_update(). + * - Digest will be equivalent to `XXH3_64bits_withSeed()`. + * + * @see @ref streaming_example "Streaming Example" + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed); + +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * `secret` is referenced, it _must outlive_ the hash streaming session. + * + * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN, + * and the quality of produced hash values depends on secret's entropy + * (secret's content should look like a bunch of random bytes). + * When in doubt about the randomness of a candidate `secret`, + * consider employing `XXH3_generateSecret()` instead (see below). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize); + +/*! + * @brief Consumes a block of @p input to an @ref XXH3_state_t. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * @pre + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note Call this to incrementally consume blocks of data. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated XXH3 64-bit hash value from that state. + * + * @note + * Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/* note : canonical representation of XXH3 is the same as XXH64 + * since they both produce XXH64_hash_t values */ + + +/*-********************************************************************** +* XXH3 128-bit variant +************************************************************************/ + +/*! + * @brief The return value from 128-bit hashes. + * + * Stored in little endian order, although the fields themselves are in native + * endianness. + */ +typedef struct { + XXH64_hash_t low64; /*!< `value & 0xFFFFFFFFFFFFFFFF` */ + XXH64_hash_t high64; /*!< `value >> 64` */ +} XXH128_hash_t; + +/*! + * @brief Calculates 128-bit unseeded variant of XXH3 of @p data. + * + * @param data The block of data to be hashed, at least @p length bytes in size. + * @param len The length of @p data, in bytes. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead + * for shorter inputs. + * + * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of `0`, however + * it may have slightly better performance due to constant propagation of the + * defaults. + * + * @see XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len); +/*! @brief Calculates 128-bit seeded variant of XXH3 hash of @p data. + * + * @param data The block of data to be hashed, at least @p length bytes in size. + * @param len The length of @p data, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * @note + * seed == 0 produces the same results as @ref XXH3_64bits(). + * + * This variant generates a custom secret on the fly based on default secret + * altered using the @p seed value. + * + * While this operation is decently fast, note that it's not completely free. + * + * @see XXH3_128bits(), XXH3_128bits_withSecret(): other seeding variants + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed); +/*! + * @brief Calculates 128-bit variant of XXH3 with a custom "secret". + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @return The calculated 128-bit variant of XXH3 value. + * + * It's possible to provide any blob of bytes as a "secret" to generate the hash. + * This makes it more difficult for an external actor to prepare an intentional collision. + * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN). + * However, the quality of the secret impacts the dispersion of the hash algorithm. + * Therefore, the secret _must_ look like a bunch of random bytes. + * Avoid "trivial" or structured data such as repeated sequences or a text document. + * Whenever in doubt about the "randomness" of the blob of bytes, + * consider employing @ref XXH3_generateSecret() instead (see below). + * It will generate a proper high entropy secret derived from the blob of bytes. + * Another advantage of using XXH3_generateSecret() is that + * it guarantees that all bits within the initial blob of bytes + * will impact every bit of the output. + * This is not necessarily the case when using the blob of bytes directly + * because, when hashing _small_ inputs, only a portion of the secret is employed. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize); + +/******* Streaming *******/ +#ifndef XXH_NO_STREAM +/* + * Streaming requires state maintenance. + * This operation costs memory and CPU. + * As a consequence, streaming is slower than one-shot hashing. + * For better performance, prefer one-shot functions whenever applicable. + * + * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits(). + * Use already declared XXH3_createState() and XXH3_freeState(). + * + * All reset and streaming functions have same meaning as their 64-bit counterpart. + */ + +/*! + * @brief Resets an @ref XXH3_state_t to begin a new hash. + * + * @param statePtr The state struct to reset. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret with default parameters. + * - Call it before @ref XXH3_128bits_update(). + * - Digest will be equivalent to `XXH3_128bits()`. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr); + +/*! + * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * - This function resets `statePtr` and generate a secret from `seed`. + * - Call it before @ref XXH3_128bits_update(). + * - Digest will be equivalent to `XXH3_128bits_withSeed()`. + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed); +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr The state struct to reset. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * `secret` is referenced, it _must outlive_ the hash streaming session. + * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN, + * and the quality of produced hash values depends on secret's entropy + * (secret's content should look like a bunch of random bytes). + * When in doubt about the randomness of a candidate `secret`, + * consider employing `XXH3_generateSecret()` instead (see below). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize); + +/*! + * @brief Consumes a block of @p input to an @ref XXH3_state_t. + * + * Call this to incrementally consume blocks of data. + * + * @param statePtr The state struct to update. + * @param input The block of data to be hashed, at least @p length bytes in size. + * @param length The length of @p input, in bytes. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @note + * The memory between @p input and @p input + @p length must be valid, + * readable, contiguous memory. However, if @p length is `0`, @p input may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + */ +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length); + +/*! + * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t. + * + * @param statePtr The state struct to calculate the hash from. + * + * @pre + * @p statePtr must not be `NULL`. + * + * @return The calculated XXH3 128-bit hash value from that state. + * + * @note + * Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update, + * digest, and update again. + * + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr); +#endif /* !XXH_NO_STREAM */ + +/* Following helper functions make it possible to compare XXH128_hast_t values. + * Since XXH128_hash_t is a structure, this capability is not offered by the language. + * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */ + +/*! + * @brief Check equality of two XXH128_hash_t values + * + * @param h1 The 128-bit hash value. + * @param h2 Another 128-bit hash value. + * + * @return `1` if `h1` and `h2` are equal. + * @return `0` if they are not. + */ +XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2); + +/*! + * @brief Compares two @ref XXH128_hash_t + * + * This comparator is compatible with stdlib's `qsort()`/`bsearch()`. + * + * @param h128_1 Left-hand side value + * @param h128_2 Right-hand side value + * + * @return >0 if @p h128_1 > @p h128_2 + * @return =0 if @p h128_1 == @p h128_2 + * @return <0 if @p h128_1 < @p h128_2 + */ +XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2); + + +/******* Canonical representation *******/ +typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t; + + +/*! + * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t. + * + * @param dst The @ref XXH128_canonical_t pointer to be stored to. + * @param hash The @ref XXH128_hash_t to be converted. + * + * @pre + * @p dst must not be `NULL`. + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash); + +/*! + * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t. + * + * @param src The @ref XXH128_canonical_t to convert. + * + * @pre + * @p src must not be `NULL`. + * + * @return The converted hash. + * @see @ref canonical_representation_example "Canonical Representation Example" + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src); + + +#endif /* !XXH_NO_XXH3 */ +#endif /* XXH_NO_LONG_LONG */ + +/*! + * @} + */ +#endif /* XXHASH_H_5627135585666179 */ + + + +#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) +#define XXHASH_H_STATIC_13879238742 +/* **************************************************************************** + * This section contains declarations which are not guaranteed to remain stable. + * They may change in future versions, becoming incompatible with a different + * version of the library. + * These declarations should only be used with static linking. + * Never use them in association with dynamic linking! + ***************************************************************************** */ + +/* + * These definitions are only present to allow static allocation + * of XXH states, on stack or in a struct, for example. + * Never **ever** access their members directly. + */ + +/*! + * @internal + * @brief Structure for XXH32 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is + * an opaque type. This allows fields to safely be changed. + * + * Typedef'd to @ref XXH32_state_t. + * Do not access the members of this struct directly. + * @see XXH64_state_s, XXH3_state_s + */ +struct XXH32_state_s { + XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */ + XXH32_hash_t large_len; /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */ + XXH32_hash_t acc[4]; /*!< Accumulator lanes */ + unsigned char buffer[16]; /*!< Internal buffer for partial reads. */ + XXH32_hash_t bufferedSize; /*!< Amount of data in @ref buffer */ + XXH32_hash_t reserved; /*!< Reserved field. Do not read nor write to it. */ +}; /* typedef'd to XXH32_state_t */ + + +#ifndef XXH_NO_LONG_LONG /* defined when there is no 64-bit support */ + +/*! + * @internal + * @brief Structure for XXH64 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is + * an opaque type. This allows fields to safely be changed. + * + * Typedef'd to @ref XXH64_state_t. + * Do not access the members of this struct directly. + * @see XXH32_state_s, XXH3_state_s + */ +struct XXH64_state_s { + XXH64_hash_t total_len; /*!< Total length hashed. This is always 64-bit. */ + XXH64_hash_t acc[4]; /*!< Accumulator lanes */ + unsigned char buffer[32]; /*!< Internal buffer for partial reads.. */ + XXH32_hash_t bufferedSize; /*!< Amount of data in @ref buffer */ + XXH32_hash_t reserved32; /*!< Reserved field, needed for padding anyways*/ + XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it. */ +}; /* typedef'd to XXH64_state_t */ + +#ifndef XXH_NO_XXH3 + +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */ +# define XXH_ALIGN(n) _Alignas(n) +#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */ +/* In C++ alignas() is a keyword */ +# define XXH_ALIGN(n) alignas(n) +#elif defined(__GNUC__) +# define XXH_ALIGN(n) __attribute__ ((aligned(n))) +#elif defined(_MSC_VER) +# define XXH_ALIGN(n) __declspec(align(n)) +#else +# define XXH_ALIGN(n) /* disabled */ +#endif + +/* Old GCC versions only accept the attribute after the type in structures. */ +#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)) /* C11+ */ \ + && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \ + && defined(__GNUC__) +# define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align) +#else +# define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type +#endif + +/*! + * @internal + * @brief The size of the internal XXH3 buffer. + * + * This is the optimal update size for incremental hashing. + * + * @see XXH3_64b_update(), XXH3_128b_update(). + */ +#define XXH3_INTERNALBUFFER_SIZE 256 + +/*! + * @def XXH3_SECRET_DEFAULT_SIZE + * @brief Default Secret's size + * + * This is the size of internal XXH3_kSecret + * and is needed by XXH3_generateSecret_fromSeed(). + * + * Not to be confused with @ref XXH3_SECRET_SIZE_MIN. + */ +#define XXH3_SECRET_DEFAULT_SIZE 192 + +/*! + * @internal + * @brief Structure for XXH3 streaming API. + * + * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY, + * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. + * Otherwise it is an opaque type. + * Never use this definition in combination with dynamic library. + * This allows fields to safely be changed in the future. + * + * @note ** This structure has a strict alignment requirement of 64 bytes!! ** + * Do not allocate this with `malloc()` or `new`, + * it will not be sufficiently aligned. + * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation. + * + * Typedef'd to @ref XXH3_state_t. + * Do never access the members of this struct directly. + * + * @see XXH3_INITSTATE() for stack initialization. + * @see XXH3_createState(), XXH3_freeState(). + * @see XXH32_state_s, XXH64_state_s + */ +struct XXH3_state_s { + XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]); + /*!< The 8 accumulators. See @ref XXH32_state_s::acc and @ref XXH64_state_s::acc */ + XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]); + /*!< Used to store a custom secret generated from a seed. */ + XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]); + /*!< The internal buffer. @see XXH32_state_s::mem32 */ + XXH32_hash_t bufferedSize; + /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */ + XXH32_hash_t useSeed; + /*!< Reserved field. Needed for padding on 64-bit. */ + size_t nbStripesSoFar; + /*!< Number or stripes processed. */ + XXH64_hash_t totalLen; + /*!< Total length hashed. 64-bit even on 32-bit targets. */ + size_t nbStripesPerBlock; + /*!< Number of stripes per block. */ + size_t secretLimit; + /*!< Size of @ref customSecret or @ref extSecret */ + XXH64_hash_t seed; + /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */ + XXH64_hash_t reserved64; + /*!< Reserved field. */ + const unsigned char* extSecret; + /*!< Reference to an external secret for the _withSecret variants, NULL + * for other variants. */ + /* note: there may be some padding at the end due to alignment on 64 bytes */ +}; /* typedef'd to XXH3_state_t */ + +#undef XXH_ALIGN_MEMBER + +/*! + * @brief Initializes a stack-allocated `XXH3_state_s`. + * + * When the @ref XXH3_state_t structure is merely emplaced on stack, + * it should be initialized with XXH3_INITSTATE() or a memset() + * in case its first reset uses XXH3_NNbits_reset_withSeed(). + * This init can be omitted if the first reset uses default or _withSecret mode. + * This operation isn't necessary when the state is created with XXH3_createState(). + * Note that this doesn't prepare the state for a streaming operation, + * it's still necessary to use XXH3_NNbits_reset*() afterwards. + */ +#define XXH3_INITSTATE(XXH3_state_ptr) \ + do { \ + XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \ + tmp_xxh3_state_ptr->seed = 0; \ + tmp_xxh3_state_ptr->extSecret = NULL; \ + } while(0) + + +/*! + * @brief Calculates the 128-bit hash of @p data using XXH3. + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param seed The 64-bit seed to alter the hash's output predictably. + * + * @pre + * The memory between @p data and @p data + @p len must be valid, + * readable, contiguous memory. However, if @p len is `0`, @p data may be + * `NULL`. In C++, this also must be *TriviallyCopyable*. + * + * @return The calculated 128-bit XXH3 value. + * + * @see @ref single_shot_example "Single Shot Example" for an example. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed); + + +/* === Experimental API === */ +/* Symbols defined below must be considered tied to a specific library version. */ + +/*! + * @brief Derive a high-entropy secret from any user-defined content, named customSeed. + * + * @param secretBuffer A writable buffer for derived high-entropy secret data. + * @param secretSize Size of secretBuffer, in bytes. Must be >= XXH3_SECRET_SIZE_MIN. + * @param customSeed A user-defined content. + * @param customSeedSize Size of customSeed, in bytes. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * The generated secret can be used in combination with `*_withSecret()` functions. + * The `_withSecret()` variants are useful to provide a higher level of protection + * than 64-bit seed, as it becomes much more difficult for an external actor to + * guess how to impact the calculation logic. + * + * The function accepts as input a custom seed of any length and any content, + * and derives from it a high-entropy secret of length @p secretSize into an + * already allocated buffer @p secretBuffer. + * + * The generated secret can then be used with any `*_withSecret()` variant. + * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(), + * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret() + * are part of this list. They all accept a `secret` parameter + * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN) + * _and_ feature very high entropy (consist of random-looking bytes). + * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can + * be employed to ensure proper quality. + * + * @p customSeed can be anything. It can have any size, even small ones, + * and its content can be anything, even "poor entropy" sources such as a bunch + * of zeroes. The resulting `secret` will nonetheless provide all required qualities. + * + * @pre + * - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN + * - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior. + * + * Example code: + * @code{.c} + * #include + * #include + * #include + * #define XXH_STATIC_LINKING_ONLY // expose unstable API + * #include "xxhash.h" + * // Hashes argv[2] using the entropy from argv[1]. + * int main(int argc, char* argv[]) + * { + * char secret[XXH3_SECRET_SIZE_MIN]; + * if (argv != 3) { return 1; } + * XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1])); + * XXH64_hash_t h = XXH3_64bits_withSecret( + * argv[2], strlen(argv[2]), + * secret, sizeof(secret) + * ); + * printf("%016llx\n", (unsigned long long) h); + * } + * @endcode + */ +XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize); + +/*! + * @brief Generate the same secret as the _withSeed() variants. + * + * @param secretBuffer A writable buffer of @ref XXH3_SECRET_DEFAULT_SIZE bytes + * @param seed The 64-bit seed to alter the hash result predictably. + * + * The generated secret can be used in combination with + *`*_withSecret()` and `_withSecretandSeed()` variants. + * + * Example C++ `std::string` hash class: + * @code{.cpp} + * #include + * #define XXH_STATIC_LINKING_ONLY // expose unstable API + * #include "xxhash.h" + * // Slow, seeds each time + * class HashSlow { + * XXH64_hash_t seed; + * public: + * HashSlow(XXH64_hash_t s) : seed{s} {} + * size_t operator()(const std::string& x) const { + * return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)}; + * } + * }; + * // Fast, caches the seeded secret for future uses. + * class HashFast { + * unsigned char secret[XXH3_SECRET_DEFAULT_SIZE]; + * public: + * HashFast(XXH64_hash_t s) { + * XXH3_generateSecret_fromSeed(secret, seed); + * } + * size_t operator()(const std::string& x) const { + * return size_t{ + * XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret)) + * }; + * } + * }; + * @endcode + */ +XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed); + +/*! + * @brief Maximum size of "short" key in bytes. + */ +#define XXH3_MIDSIZE_MAX 240 + +/*! + * @brief Calculates 64/128-bit seeded variant of XXH3 hash of @p data. + * + * @param data The block of data to be hashed, at least @p len bytes in size. + * @param len The length of @p data, in bytes. + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed The 64-bit seed to alter the hash result predictably. + * + * These variants generate hash values using either: + * - @p seed for "short" keys (< @ref XXH3_MIDSIZE_MAX = 240 bytes) + * - @p secret for "large" keys (>= @ref XXH3_MIDSIZE_MAX). + * + * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`. + * `_withSeed()` has to generate the secret on the fly for "large" keys. + * It's fast, but can be perceptible for "not so large" keys (< 1 KB). + * `_withSecret()` has to generate the masks on the fly for "small" keys, + * which requires more instructions than _withSeed() variants. + * Therefore, _withSecretandSeed variant combines the best of both worlds. + * + * When @p secret has been generated by XXH3_generateSecret_fromSeed(), + * this variant produces *exactly* the same results as `_withSeed()` variant, + * hence offering only a pure speed benefit on "large" input, + * by skipping the need to regenerate the secret for every large input. + * + * Another usage scenario is to hash the secret to a 64-bit hash value, + * for example with XXH3_64bits(), which then becomes the seed, + * and then employ both the seed and the secret in _withSecretandSeed(). + * On top of speed, an added benefit is that each bit in the secret + * has a 50% chance to swap each bit in the output, via its impact to the seed. + * + * This is not guaranteed when using the secret directly in "small data" scenarios, + * because only portions of the secret are employed for small data. + */ +XXH_PUBLIC_API XXH_PUREF XXH64_hash_t +XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed); + +/*! + * @brief Calculates 128-bit seeded variant of XXH3 hash of @p data. + * + * @param input The memory segment to be hashed, at least @p len bytes in size. + * @param length The length of @p data, in bytes. + * @param secret The secret used to alter hash result predictably. + * @param secretSize The length of @p secret, in bytes (must be >= XXH3_SECRET_SIZE_MIN) + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(): contract is the same. + */ +XXH_PUBLIC_API XXH_PUREF XXH128_hash_t +XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +#ifndef XXH_NO_STREAM +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(). Contract is identical. + */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +/*! + * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * @param secret The secret data. + * @param secretSize The length of @p secret, in bytes. + * @param seed64 The 64-bit seed to alter the hash result predictably. + * + * @return @ref XXH_OK on success. + * @return @ref XXH_ERROR on failure. + * + * @see XXH3_64bits_withSecretandSeed(). Contract is identical. + * + * Note: there was a bug in an earlier version of this function (<= v0.8.2) + * that would make it generate an incorrect hash value + * when @p seed == 0 and @p length < XXH3_MIDSIZE_MAX + * and @p secret is different from XXH3_generateSecret_fromSeed(). + * As stated in the contract, the correct hash result must be + * the same as XXH3_128bits_withSeed() when @p length <= XXH3_MIDSIZE_MAX. + * Results generated by this older version are wrong, hence not comparable. + */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, + XXH_NOESCAPE const void* secret, size_t secretSize, + XXH64_hash_t seed64); + +#endif /* !XXH_NO_STREAM */ + +#endif /* !XXH_NO_XXH3 */ +#endif /* XXH_NO_LONG_LONG */ +#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) +# define XXH_IMPLEMENTATION +#endif + +#endif /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */ + + +/* ======================================================================== */ +/* ======================================================================== */ +/* ======================================================================== */ + + +/*-********************************************************************** + * xxHash implementation + *-********************************************************************** + * xxHash's implementation used to be hosted inside xxhash.c. + * + * However, inlining requires implementation to be visible to the compiler, + * hence be included alongside the header. + * Previously, implementation was hosted inside xxhash.c, + * which was then #included when inlining was activated. + * This construction created issues with a few build and install systems, + * as it required xxhash.c to be stored in /include directory. + * + * xxHash implementation is now directly integrated within xxhash.h. + * As a consequence, xxhash.c is no longer needed in /include. + * + * xxhash.c is still available and is still useful. + * In a "normal" setup, when xxhash is not inlined, + * xxhash.h only exposes the prototypes and public symbols, + * while xxhash.c can be built into an object file xxhash.o + * which can then be linked into the final binary. + ************************************************************************/ + +#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \ + || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387) +# define XXH_IMPLEM_13a8737387 + +/* ************************************* +* Tuning parameters +***************************************/ + +/*! + * @defgroup tuning Tuning parameters + * @{ + * + * Various macros to control xxHash's behavior. + */ +#ifdef XXH_DOXYGEN +/*! + * @brief Define this to disable 64-bit code. + * + * Useful if only using the @ref XXH32_family and you have a strict C90 compiler. + */ +# define XXH_NO_LONG_LONG +# undef XXH_NO_LONG_LONG /* don't actually */ +/*! + * @brief Controls how unaligned memory is accessed. + * + * By default, access to unaligned memory is controlled by `memcpy()`, which is + * safe and portable. + * + * Unfortunately, on some target/compiler combinations, the generated assembly + * is sub-optimal. + * + * The below switch allow selection of a different access method + * in the search for improved performance. + * + * @par Possible options: + * + * - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy` + * @par + * Use `memcpy()`. Safe and portable. Note that most modern compilers will + * eliminate the function call and treat it as an unaligned access. + * + * - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))` + * @par + * Depends on compiler extensions and is therefore not portable. + * This method is safe _if_ your compiler supports it, + * and *generally* as fast or faster than `memcpy`. + * + * - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast + * @par + * Casts directly and dereferences. This method doesn't depend on the + * compiler, but it violates the C standard as it directly dereferences an + * unaligned pointer. It can generate buggy code on targets which do not + * support unaligned memory accesses, but in some circumstances, it's the + * only known way to get the most performance. + * + * - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift + * @par + * Also portable. This can generate the best code on old compilers which don't + * inline small `memcpy()` calls, and it might also be faster on big-endian + * systems which lack a native byteswap instruction. However, some compilers + * will emit literal byteshifts even if the target supports unaligned access. + * + * + * @warning + * Methods 1 and 2 rely on implementation-defined behavior. Use these with + * care, as what works on one compiler/platform/optimization level may cause + * another to read garbage data or even crash. + * + * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details. + * + * Prefer these methods in priority order (0 > 3 > 1 > 2) + */ +# define XXH_FORCE_MEMORY_ACCESS 0 + +/*! + * @def XXH_SIZE_OPT + * @brief Controls how much xxHash optimizes for size. + * + * xxHash, when compiled, tends to result in a rather large binary size. This + * is mostly due to heavy usage to forced inlining and constant folding of the + * @ref XXH3_family to increase performance. + * + * However, some developers prefer size over speed. This option can + * significantly reduce the size of the generated code. When using the `-Os` + * or `-Oz` options on GCC or Clang, this is defined to 1 by default, + * otherwise it is defined to 0. + * + * Most of these size optimizations can be controlled manually. + * + * This is a number from 0-2. + * - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed + * comes first. + * - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more + * conservative and disables hacks that increase code size. It implies the + * options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0, + * and @ref XXH3_NEON_LANES == 8 if they are not already defined. + * - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible. + * Performance may cry. For example, the single shot functions just use the + * streaming API. + */ +# define XXH_SIZE_OPT 0 + +/*! + * @def XXH_FORCE_ALIGN_CHECK + * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32() + * and XXH64() only). + * + * This is an important performance trick for architectures without decent + * unaligned memory access performance. + * + * It checks for input alignment, and when conditions are met, uses a "fast + * path" employing direct 32-bit/64-bit reads, resulting in _dramatically + * faster_ read speed. + * + * The check costs one initial branch per hash, which is generally negligible, + * but not zero. + * + * Moreover, it's not useful to generate an additional code path if memory + * access uses the same instruction for both aligned and unaligned + * addresses (e.g. x86 and aarch64). + * + * In these cases, the alignment check can be removed by setting this macro to 0. + * Then the code will always use unaligned memory access. + * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips + * which are platforms known to offer good unaligned memory accesses performance. + * + * It is also disabled by default when @ref XXH_SIZE_OPT >= 1. + * + * This option does not affect XXH3 (only XXH32 and XXH64). + */ +# define XXH_FORCE_ALIGN_CHECK 0 + +/*! + * @def XXH_NO_INLINE_HINTS + * @brief When non-zero, sets all functions to `static`. + * + * By default, xxHash tries to force the compiler to inline almost all internal + * functions. + * + * This can usually improve performance due to reduced jumping and improved + * constant folding, but significantly increases the size of the binary which + * might not be favorable. + * + * Additionally, sometimes the forced inlining can be detrimental to performance, + * depending on the architecture. + * + * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the + * compiler full control on whether to inline or not. + * + * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if + * @ref XXH_SIZE_OPT >= 1, this will automatically be defined. + */ +# define XXH_NO_INLINE_HINTS 0 + +/*! + * @def XXH3_INLINE_SECRET + * @brief Determines whether to inline the XXH3 withSecret code. + * + * When the secret size is known, the compiler can improve the performance + * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret(). + * + * However, if the secret size is not known, it doesn't have any benefit. This + * happens when xxHash is compiled into a global symbol. Therefore, if + * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0. + * + * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers + * that are *sometimes* force inline on -Og, and it is impossible to automatically + * detect this optimization level. + */ +# define XXH3_INLINE_SECRET 0 + +/*! + * @def XXH32_ENDJMP + * @brief Whether to use a jump for `XXH32_finalize`. + * + * For performance, `XXH32_finalize` uses multiple branches in the finalizer. + * This is generally preferable for performance, + * but depending on exact architecture, a jmp may be preferable. + * + * This setting is only possibly making a difference for very small inputs. + */ +# define XXH32_ENDJMP 0 + +/*! + * @internal + * @brief Redefines old internal names. + * + * For compatibility with code that uses xxHash's internals before the names + * were changed to improve namespacing. There is no other reason to use this. + */ +# define XXH_OLD_NAMES +# undef XXH_OLD_NAMES /* don't actually use, it is ugly. */ + +/*! + * @def XXH_NO_STREAM + * @brief Disables the streaming API. + * + * When xxHash is not inlined and the streaming functions are not used, disabling + * the streaming functions can improve code size significantly, especially with + * the @ref XXH3_family which tends to make constant folded copies of itself. + */ +# define XXH_NO_STREAM +# undef XXH_NO_STREAM /* don't actually */ +#endif /* XXH_DOXYGEN */ +/*! + * @} + */ + +#ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */ + /* prefer __packed__ structures (method 1) for GCC + * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy + * which for some reason does unaligned loads. */ +# if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED)) +# define XXH_FORCE_MEMORY_ACCESS 1 +# endif +#endif + +#ifndef XXH_SIZE_OPT + /* default to 1 for -Os or -Oz */ +# if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__) +# define XXH_SIZE_OPT 1 +# else +# define XXH_SIZE_OPT 0 +# endif +#endif + +#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */ + /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */ +# if XXH_SIZE_OPT >= 1 || \ + defined(__i386) || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \ + || defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) || defined(_M_ARM) /* visual */ +# define XXH_FORCE_ALIGN_CHECK 0 +# else +# define XXH_FORCE_ALIGN_CHECK 1 +# endif +#endif + +#ifndef XXH_NO_INLINE_HINTS +# if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__) /* -O0, -fno-inline */ +# define XXH_NO_INLINE_HINTS 1 +# else +# define XXH_NO_INLINE_HINTS 0 +# endif +#endif + +#ifndef XXH3_INLINE_SECRET +# if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \ + || !defined(XXH_INLINE_ALL) +# define XXH3_INLINE_SECRET 0 +# else +# define XXH3_INLINE_SECRET 1 +# endif +#endif + +#ifndef XXH32_ENDJMP +/* generally preferable for performance */ +# define XXH32_ENDJMP 0 +#endif + +/*! + * @defgroup impl Implementation + * @{ + */ + + +/* ************************************* +* Includes & Memory related functions +***************************************/ +#if defined(XXH_NO_STREAM) +/* nothing */ +#elif defined(XXH_NO_STDLIB) + +/* When requesting to disable any mention of stdlib, + * the library loses the ability to invoked malloc / free. + * In practice, it means that functions like `XXH*_createState()` + * will always fail, and return NULL. + * This flag is useful in situations where + * xxhash.h is integrated into some kernel, embedded or limited environment + * without access to dynamic allocation. + */ + +static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; } +static void XXH_free(void* p) { (void)p; } + +#else + +/* + * Modify the local functions below should you wish to use + * different memory routines for malloc() and free() + */ +#include + +/*! + * @internal + * @brief Modify this function to use a different routine than malloc(). + */ +static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); } + +/*! + * @internal + * @brief Modify this function to use a different routine than free(). + */ +static void XXH_free(void* p) { free(p); } + +#endif /* XXH_NO_STDLIB */ + +#ifndef XXH_memcpy +/*! + * @internal + * @brief XXH_memcpy() macro can be redirected at compile time + */ +# include +# define XXH_memcpy memcpy +#endif + +#ifndef XXH_memset +/*! + * @internal + * @brief XXH_memset() macro can be redirected at compile time + */ +# include +# define XXH_memset memset +#endif + +#ifndef XXH_memcmp +/*! + * @internal + * @brief XXH_memcmp() macro can be redirected at compile time + * Note: only needed by XXH128. + */ +# include +# define XXH_memcmp memcmp +#endif + + + +#include /* ULLONG_MAX */ + + +/* ************************************* +* Compiler Specific Options +***************************************/ +#ifdef _MSC_VER /* Visual Studio warning fix */ +# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ +#endif + +#if XXH_NO_INLINE_HINTS /* disable inlining hints */ +# if defined(__GNUC__) || defined(__clang__) +# define XXH_FORCE_INLINE static __attribute__((__unused__)) +# else +# define XXH_FORCE_INLINE static +# endif +# define XXH_NO_INLINE static +/* enable inlining hints */ +#elif defined(__GNUC__) || defined(__clang__) +# define XXH_FORCE_INLINE static __inline__ __attribute__((__always_inline__, __unused__)) +# define XXH_NO_INLINE static __attribute__((__noinline__)) +#elif defined(_MSC_VER) /* Visual Studio */ +# define XXH_FORCE_INLINE static __forceinline +# define XXH_NO_INLINE static __declspec(noinline) +#elif defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) /* C99 */ +# define XXH_FORCE_INLINE static inline +# define XXH_NO_INLINE static +#else +# define XXH_FORCE_INLINE static +# define XXH_NO_INLINE static +#endif + +#if defined(XXH_INLINE_ALL) +# define XXH_STATIC XXH_FORCE_INLINE +#else +# define XXH_STATIC static +#endif + +#if XXH3_INLINE_SECRET +# define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE +#else +# define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE +#endif + +#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */ +# define XXH_RESTRICT /* disable */ +#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* >= C99 */ +# define XXH_RESTRICT restrict +#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \ + || (defined (__clang__)) \ + || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \ + || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300)) +/* + * There are a LOT more compilers that recognize __restrict but this + * covers the major ones. + */ +# define XXH_RESTRICT __restrict +#else +# define XXH_RESTRICT /* disable */ +#endif + +/* ************************************* +* Debug +***************************************/ +/*! + * @ingroup tuning + * @def XXH_DEBUGLEVEL + * @brief Sets the debugging level. + * + * XXH_DEBUGLEVEL is expected to be defined externally, typically via the + * compiler's command line options. The value must be a number. + */ +#ifndef XXH_DEBUGLEVEL +# ifdef DEBUGLEVEL /* backwards compat */ +# define XXH_DEBUGLEVEL DEBUGLEVEL +# else +# define XXH_DEBUGLEVEL 0 +# endif +#endif + +#if (XXH_DEBUGLEVEL>=1) +# include /* note: can still be disabled with NDEBUG */ +# define XXH_ASSERT(c) assert(c) +#else +# if defined(__INTEL_COMPILER) +# define XXH_ASSERT(c) XXH_ASSUME((unsigned char) (c)) +# else +# define XXH_ASSERT(c) XXH_ASSUME(c) +# endif +#endif + +/* note: use after variable declarations */ +#ifndef XXH_STATIC_ASSERT +# if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */ +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0) +# elif defined(__cplusplus) && (__cplusplus >= 201103L) /* C++11 */ +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0) +# else +# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0) +# endif +# define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c) +#endif + +/*! + * @internal + * @def XXH_COMPILER_GUARD(var) + * @brief Used to prevent unwanted optimizations for @p var. + * + * It uses an empty GCC inline assembly statement with a register constraint + * which forces @p var into a general purpose register (eg eax, ebx, ecx + * on x86) and marks it as modified. + * + * This is used in a few places to avoid unwanted autovectorization (e.g. + * XXH32_round()). All vectorization we want is explicit via intrinsics, + * and _usually_ isn't wanted elsewhere. + * + * We also use it to prevent unwanted constant folding for AArch64 in + * XXH3_initCustomSecret_scalar(). + */ +#if defined(__GNUC__) || defined(__clang__) +# define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var)) +#else +# define XXH_COMPILER_GUARD(var) ((void)0) +#endif + +/* Specifically for NEON vectors which use the "w" constraint, on + * Clang. */ +#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__) +# define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var)) +#else +# define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0) +#endif + +/* ************************************* +* Basic Types +***************************************/ +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# ifdef _AIX +# include +# else +# include +# endif + typedef uint8_t xxh_u8; +#else + typedef unsigned char xxh_u8; +#endif +typedef XXH32_hash_t xxh_u32; + +#ifdef XXH_OLD_NAMES +# warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly" +# define BYTE xxh_u8 +# define U8 xxh_u8 +# define U32 xxh_u32 +#endif + +/* *** Memory access *** */ + +/*! + * @internal + * @fn xxh_u32 XXH_read32(const void* ptr) + * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit native endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readLE32(const void* ptr) + * @brief Reads an unaligned 32-bit little endian integer from @p ptr. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit little endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readBE32(const void* ptr) + * @brief Reads an unaligned 32-bit big endian integer from @p ptr. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * + * @param ptr The pointer to read from. + * @return The 32-bit big endian integer from the bytes at @p ptr. + */ + +/*! + * @internal + * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align) + * @brief Like @ref XXH_readLE32(), but has an option for aligned reads. + * + * Affected by @ref XXH_FORCE_MEMORY_ACCESS. + * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is + * always @ref XXH_alignment::XXH_unaligned. + * + * @param ptr The pointer to read from. + * @param align Whether @p ptr is aligned. + * @pre + * If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte + * aligned. + * @return The 32-bit little endian integer from the bytes at @p ptr. + */ + +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) +/* + * Manual byteshift. Best for old compilers which don't inline memcpy. + * We actually directly use XXH_readLE32 and XXH_readBE32. + */ +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) + +/* + * Force direct memory access. Only works on CPU which support unaligned memory + * access in hardware. + */ +static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; } + +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) + +/* + * __attribute__((aligned(1))) is supported by gcc and clang. Originally the + * documentation claimed that it only increased the alignment, but actually it + * can decrease it on gcc, clang, and icc: + * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502, + * https://gcc.godbolt.org/z/xYez1j67Y. + */ +#ifdef XXH_OLD_NAMES +typedef union { xxh_u32 u32; } __attribute__((__packed__)) unalign; +#endif +static xxh_u32 XXH_read32(const void* ptr) +{ + typedef __attribute__((__aligned__(1))) xxh_u32 xxh_unalign32; + return *((const xxh_unalign32*)ptr); +} + +#else + +/* + * Portable and safe solution. Generally efficient. + * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html + */ +static xxh_u32 XXH_read32(const void* memPtr) +{ + xxh_u32 val; + XXH_memcpy(&val, memPtr, sizeof(val)); + return val; +} + +#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ + + +/* *** Endianness *** */ + +/*! + * @ingroup tuning + * @def XXH_CPU_LITTLE_ENDIAN + * @brief Whether the target is little endian. + * + * Defined to 1 if the target is little endian, or 0 if it is big endian. + * It can be defined externally, for example on the compiler command line. + * + * If it is not defined, + * a runtime check (which is usually constant folded) is used instead. + * + * @note + * This is not necessarily defined to an integer constant. + * + * @see XXH_isLittleEndian() for the runtime check. + */ +#ifndef XXH_CPU_LITTLE_ENDIAN +/* + * Try to detect endianness automatically, to avoid the nonstandard behavior + * in `XXH_isLittleEndian()` + */ +# if defined(_WIN32) /* Windows is always little endian */ \ + || defined(__LITTLE_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 1 +# elif defined(__BIG_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 0 +# else +/*! + * @internal + * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN. + * + * Most compilers will constant fold this. + */ +static int XXH_isLittleEndian(void) +{ + /* + * Portable and well-defined behavior. + * Don't use static: it is detrimental to performance. + */ + const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 }; + return one.c[0]; +} +# define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian() +# endif +#endif + + + + +/* **************************************** +* Compiler-specific Functions and Macros +******************************************/ +#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) + +#ifdef __has_builtin +# define XXH_HAS_BUILTIN(x) __has_builtin(x) +#else +# define XXH_HAS_BUILTIN(x) 0 +#endif + + + +/* + * C23 and future versions have standard "unreachable()". + * Once it has been implemented reliably we can add it as an + * additional case: + * + * ``` + * #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) + * # include + * # ifdef unreachable + * # define XXH_UNREACHABLE() unreachable() + * # endif + * #endif + * ``` + * + * Note C++23 also has std::unreachable() which can be detected + * as follows: + * ``` + * #if defined(__cpp_lib_unreachable) && (__cpp_lib_unreachable >= 202202L) + * # include + * # define XXH_UNREACHABLE() std::unreachable() + * #endif + * ``` + * NB: `__cpp_lib_unreachable` is defined in the `` header. + * We don't use that as including `` in `extern "C"` blocks + * doesn't work on GCC12 + */ + +#if XXH_HAS_BUILTIN(__builtin_unreachable) +# define XXH_UNREACHABLE() __builtin_unreachable() + +#elif defined(_MSC_VER) +# define XXH_UNREACHABLE() __assume(0) + +#else +# define XXH_UNREACHABLE() +#endif + +#if XXH_HAS_BUILTIN(__builtin_assume) +# define XXH_ASSUME(c) __builtin_assume(c) +#else +# define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); } +#endif + +/*! + * @internal + * @def XXH_rotl32(x,r) + * @brief 32-bit rotate left. + * + * @param x The 32-bit integer to be rotated. + * @param r The number of bits to rotate. + * @pre + * @p r > 0 && @p r < 32 + * @note + * @p x and @p r may be evaluated multiple times. + * @return The rotated result. + */ +#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \ + && XXH_HAS_BUILTIN(__builtin_rotateleft64) +# define XXH_rotl32 __builtin_rotateleft32 +# define XXH_rotl64 __builtin_rotateleft64 +#elif XXH_HAS_BUILTIN(__builtin_stdc_rotate_left) +# define XXH_rotl32 __builtin_stdc_rotate_left +# define XXH_rotl64 __builtin_stdc_rotate_left +/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */ +#elif defined(_MSC_VER) +# define XXH_rotl32(x,r) _rotl(x,r) +# define XXH_rotl64(x,r) _rotl64(x,r) +#else +# define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +# define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r)))) +#endif + +/*! + * @internal + * @fn xxh_u32 XXH_swap32(xxh_u32 x) + * @brief A 32-bit byteswap. + * + * @param x The 32-bit integer to byteswap. + * @return @p x, byteswapped. + */ +#if defined(_MSC_VER) /* Visual Studio */ +# define XXH_swap32 _byteswap_ulong +#elif XXH_GCC_VERSION >= 403 +# define XXH_swap32 __builtin_bswap32 +#else +static xxh_u32 XXH_swap32 (xxh_u32 x) +{ + return ((x << 24) & 0xff000000 ) | + ((x << 8) & 0x00ff0000 ) | + ((x >> 8) & 0x0000ff00 ) | + ((x >> 24) & 0x000000ff ); +} +#endif + + +/* *************************** +* Memory reads +*****************************/ + +/*! + * @internal + * @brief Enum to indicate whether a pointer is aligned. + */ +typedef enum { + XXH_aligned, /*!< Aligned */ + XXH_unaligned /*!< Possibly unaligned */ +} XXH_alignment; + +/* + * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. + * + * This is ideal for older compilers which don't inline memcpy. + */ +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) + +XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[0] + | ((xxh_u32)bytePtr[1] << 8) + | ((xxh_u32)bytePtr[2] << 16) + | ((xxh_u32)bytePtr[3] << 24); +} + +XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[3] + | ((xxh_u32)bytePtr[2] << 8) + | ((xxh_u32)bytePtr[1] << 16) + | ((xxh_u32)bytePtr[0] << 24); +} + +#else +XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr)); +} + +static xxh_u32 XXH_readBE32(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr); +} +#endif + +XXH_FORCE_INLINE xxh_u32 +XXH_readLE32_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) { + return XXH_readLE32(ptr); + } else { + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr); + } +} + + +/* ************************************* +* Misc +***************************************/ +/*! @ingroup public */ +XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; } + + +/* ******************************************************************* +* 32-bit hash functions +*********************************************************************/ +/*! + * @} + * @defgroup XXH32_impl XXH32 implementation + * @ingroup impl + * + * Details on the XXH32 implementation. + * @{ + */ + /* #define instead of static const, to be used as initializers */ +#define XXH_PRIME32_1 0x9E3779B1U /*!< 0b10011110001101110111100110110001 */ +#define XXH_PRIME32_2 0x85EBCA77U /*!< 0b10000101111010111100101001110111 */ +#define XXH_PRIME32_3 0xC2B2AE3DU /*!< 0b11000010101100101010111000111101 */ +#define XXH_PRIME32_4 0x27D4EB2FU /*!< 0b00100111110101001110101100101111 */ +#define XXH_PRIME32_5 0x165667B1U /*!< 0b00010110010101100110011110110001 */ + +#ifdef XXH_OLD_NAMES +# define PRIME32_1 XXH_PRIME32_1 +# define PRIME32_2 XXH_PRIME32_2 +# define PRIME32_3 XXH_PRIME32_3 +# define PRIME32_4 XXH_PRIME32_4 +# define PRIME32_5 XXH_PRIME32_5 +#endif + +/*! + * @internal + * @brief Normal stripe processing routine. + * + * This shuffles the bits so that any bit from @p input impacts several bits in + * @p acc. + * + * @param acc The accumulator lane. + * @param input The stripe of input to mix. + * @return The mixed accumulator lane. + */ +static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) +{ + acc += input * XXH_PRIME32_2; + acc = XXH_rotl32(acc, 13); + acc *= XXH_PRIME32_1; +#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE) + /* + * UGLY HACK: + * A compiler fence is used to prevent GCC and Clang from + * autovectorizing the XXH32 loop (pragmas and attributes don't work for some + * reason) without globally disabling SSE4.1. + * + * The reason we want to avoid vectorization is because despite working on + * 4 integers at a time, there are multiple factors slowing XXH32 down on + * SSE4: + * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on + * newer chips!) making it slightly slower to multiply four integers at + * once compared to four integers independently. Even when pmulld was + * fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE + * just to multiply unless doing a long operation. + * + * - Four instructions are required to rotate, + * movqda tmp, v // not required with VEX encoding + * pslld tmp, 13 // tmp <<= 13 + * psrld v, 19 // x >>= 19 + * por v, tmp // x |= tmp + * compared to one for scalar: + * roll v, 13 // reliably fast across the board + * shldl v, v, 13 // Sandy Bridge and later prefer this for some reason + * + * - Instruction level parallelism is actually more beneficial here because + * the SIMD actually serializes this operation: While v1 is rotating, v2 + * can load data, while v3 can multiply. SSE forces them to operate + * together. + * + * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing + * the loop. NEON is only faster on the A53, and with the newer cores, it is less + * than half the speed. + * + * Additionally, this is used on WASM SIMD128 because it JITs to the same + * SIMD instructions and has the same issue. + */ + XXH_COMPILER_GUARD(acc); +#endif + return acc; +} + +/*! + * @internal + * @brief Mixes all bits to finalize the hash. + * + * The final mix ensures that all input bits have a chance to impact any bit in + * the output digest, resulting in an unbiased distribution. + * + * @param hash The hash to avalanche. + * @return The avalanched hash. + */ +static xxh_u32 XXH32_avalanche(xxh_u32 hash) +{ + hash ^= hash >> 15; + hash *= XXH_PRIME32_2; + hash ^= hash >> 13; + hash *= XXH_PRIME32_3; + hash ^= hash >> 16; + return hash; +} + +#define XXH_get32bits(p) XXH_readLE32_align(p, align) + +/*! + * @internal + * @brief Sets up the initial accumulator state for XXH32(). + */ +XXH_FORCE_INLINE void +XXH32_initAccs(xxh_u32 *acc, xxh_u32 seed) +{ + XXH_ASSERT(acc != NULL); + acc[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2; + acc[1] = seed + XXH_PRIME32_2; + acc[2] = seed + 0; + acc[3] = seed - XXH_PRIME32_1; +} + +/*! + * @internal + * @brief Consumes a block of data for XXH32(). + * + * @return the end input pointer. + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH32_consumeLong( + xxh_u32 *XXH_RESTRICT acc, + xxh_u8 const *XXH_RESTRICT input, + size_t len, + XXH_alignment align +) +{ + const xxh_u8* const bEnd = input + len; + const xxh_u8* const limit = bEnd - 15; + XXH_ASSERT(acc != NULL); + XXH_ASSERT(input != NULL); + XXH_ASSERT(len >= 16); + do { + acc[0] = XXH32_round(acc[0], XXH_get32bits(input)); input += 4; + acc[1] = XXH32_round(acc[1], XXH_get32bits(input)); input += 4; + acc[2] = XXH32_round(acc[2], XXH_get32bits(input)); input += 4; + acc[3] = XXH32_round(acc[3], XXH_get32bits(input)); input += 4; + } while (input < limit); + + return input; +} + +/*! + * @internal + * @brief Merges the accumulator lanes together for XXH32() + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u32 +XXH32_mergeAccs(const xxh_u32 *acc) +{ + XXH_ASSERT(acc != NULL); + return XXH_rotl32(acc[0], 1) + XXH_rotl32(acc[1], 7) + + XXH_rotl32(acc[2], 12) + XXH_rotl32(acc[3], 18); +} + +/*! + * @internal + * @brief Processes the last 0-15 bytes of @p ptr. + * + * There may be up to 15 bytes remaining to consume from the input. + * This final stage will digest them to ensure that all input bytes are present + * in the final mix. + * + * @param hash The hash to finalize. + * @param ptr The pointer to the remaining input. + * @param len The remaining length, modulo 16. + * @param align Whether @p ptr is aligned. + * @return The finalized hash. + * @see XXH64_finalize(). + */ +static XXH_PUREF xxh_u32 +XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ +#define XXH_PROCESS1 do { \ + hash += (*ptr++) * XXH_PRIME32_5; \ + hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1; \ +} while (0) + +#define XXH_PROCESS4 do { \ + hash += XXH_get32bits(ptr) * XXH_PRIME32_3; \ + ptr += 4; \ + hash = XXH_rotl32(hash, 17) * XXH_PRIME32_4; \ +} while (0) + + if (ptr==NULL) XXH_ASSERT(len == 0); + + /* Compact rerolled version; generally faster */ + if (!XXH32_ENDJMP) { + len &= 15; + while (len >= 4) { + XXH_PROCESS4; + len -= 4; + } + while (len > 0) { + XXH_PROCESS1; + --len; + } + return XXH32_avalanche(hash); + } else { + switch(len&15) /* or switch(bEnd - p) */ { + case 12: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 8: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 4: XXH_PROCESS4; + return XXH32_avalanche(hash); + + case 13: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 9: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 5: XXH_PROCESS4; + XXH_PROCESS1; + return XXH32_avalanche(hash); + + case 14: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 10: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 6: XXH_PROCESS4; + XXH_PROCESS1; + XXH_PROCESS1; + return XXH32_avalanche(hash); + + case 15: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 11: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 7: XXH_PROCESS4; + XXH_FALLTHROUGH; /* fallthrough */ + case 3: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 2: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 1: XXH_PROCESS1; + XXH_FALLTHROUGH; /* fallthrough */ + case 0: return XXH32_avalanche(hash); + } + XXH_ASSERT(0); + return hash; /* reaching this point is deemed impossible */ + } +} + +#ifdef XXH_OLD_NAMES +# define PROCESS1 XXH_PROCESS1 +# define PROCESS4 XXH_PROCESS4 +#else +# undef XXH_PROCESS1 +# undef XXH_PROCESS4 +#endif + +/*! + * @internal + * @brief The implementation for @ref XXH32(). + * + * @param input , len , seed Directly passed from @ref XXH32(). + * @param align Whether @p input is aligned. + * @return The calculated hash. + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u32 +XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align) +{ + xxh_u32 h32; + + if (input==NULL) XXH_ASSERT(len == 0); + + if (len>=16) { + xxh_u32 acc[4]; + XXH32_initAccs(acc, seed); + + input = XXH32_consumeLong(acc, input, len, align); + + h32 = XXH32_mergeAccs(acc); + } else { + h32 = seed + XXH_PRIME32_5; + } + + h32 += (xxh_u32)len; + + return XXH32_finalize(h32, input, len&15, align); +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed) +{ +#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2 + /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ + XXH32_state_t state; + XXH32_reset(&state, seed); + XXH32_update(&state, (const xxh_u8*)input, len); + return XXH32_digest(&state); +#else + if (XXH_FORCE_ALIGN_CHECK) { + if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */ + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); + } } + + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); +#endif +} + + + +/******* Hash streaming *******/ +#ifndef XXH_NO_STREAM +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void) +{ + return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t)); +} +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr) +{ + XXH_free(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState) +{ + XXH_memcpy(dstState, srcState, sizeof(*dstState)); +} + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed) +{ + XXH_ASSERT(statePtr != NULL); + XXH_memset(statePtr, 0, sizeof(*statePtr)); + XXH32_initAccs(statePtr->acc, seed); + return XXH_OK; +} + + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH_errorcode +XXH32_update(XXH32_state_t* state, const void* input, size_t len) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + state->total_len_32 += (XXH32_hash_t)len; + state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16)); + + XXH_ASSERT(state->bufferedSize < sizeof(state->buffer)); + if (len < sizeof(state->buffer) - state->bufferedSize) { /* fill in tmp buffer */ + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + { const xxh_u8* xinput = (const xxh_u8*)input; + const xxh_u8* const bEnd = xinput + len; + + if (state->bufferedSize) { /* non-empty buffer: complete first */ + XXH_memcpy(state->buffer + state->bufferedSize, xinput, sizeof(state->buffer) - state->bufferedSize); + xinput += sizeof(state->buffer) - state->bufferedSize; + /* then process one round */ + (void)XXH32_consumeLong(state->acc, state->buffer, sizeof(state->buffer), XXH_aligned); + state->bufferedSize = 0; + } + + XXH_ASSERT(xinput <= bEnd); + if ((size_t)(bEnd - xinput) >= sizeof(state->buffer)) { + /* Process the remaining data */ + xinput = XXH32_consumeLong(state->acc, xinput, (size_t)(bEnd - xinput), XXH_unaligned); + } + + if (xinput < bEnd) { + /* Copy the leftover to the tmp buffer */ + XXH_memcpy(state->buffer, xinput, (size_t)(bEnd-xinput)); + state->bufferedSize = (unsigned)(bEnd-xinput); + } + } + + return XXH_OK; +} + + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state) +{ + xxh_u32 h32; + + if (state->large_len) { + h32 = XXH32_mergeAccs(state->acc); + } else { + h32 = state->acc[2] /* == seed */ + XXH_PRIME32_5; + } + + h32 += state->total_len_32; + + return XXH32_finalize(h32, state->buffer, state->bufferedSize, XXH_aligned); +} +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash); + XXH_memcpy(dst, &hash, sizeof(*dst)); +} +/*! @ingroup XXH32_family */ +XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src) +{ + return XXH_readBE32(src); +} + + +#ifndef XXH_NO_LONG_LONG + +/* ******************************************************************* +* 64-bit hash functions +*********************************************************************/ +/*! + * @} + * @ingroup impl + * @{ + */ +/******* Memory access *******/ + +typedef XXH64_hash_t xxh_u64; + +#ifdef XXH_OLD_NAMES +# define U64 xxh_u64 +#endif + +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) +/* + * Manual byteshift. Best for old compilers which don't inline memcpy. + * We actually directly use XXH_readLE64 and XXH_readBE64. + */ +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) + +/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ +static xxh_u64 XXH_read64(const void* memPtr) +{ + return *(const xxh_u64*) memPtr; +} + +#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) + +/* + * __attribute__((aligned(1))) is supported by gcc and clang. Originally the + * documentation claimed that it only increased the alignment, but actually it + * can decrease it on gcc, clang, and icc: + * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502, + * https://gcc.godbolt.org/z/xYez1j67Y. + */ +#ifdef XXH_OLD_NAMES +typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((__packed__)) unalign64; +#endif +static xxh_u64 XXH_read64(const void* ptr) +{ + typedef __attribute__((__aligned__(1))) xxh_u64 xxh_unalign64; + return *((const xxh_unalign64*)ptr); +} + +#else + +/* + * Portable and safe solution. Generally efficient. + * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html + */ +static xxh_u64 XXH_read64(const void* memPtr) +{ + xxh_u64 val; + XXH_memcpy(&val, memPtr, sizeof(val)); + return val; +} + +#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ + +#if defined(_MSC_VER) /* Visual Studio */ +# define XXH_swap64 _byteswap_uint64 +#elif XXH_GCC_VERSION >= 403 +# define XXH_swap64 __builtin_bswap64 +#else +static xxh_u64 XXH_swap64(xxh_u64 x) +{ + return ((x << 56) & 0xff00000000000000ULL) | + ((x << 40) & 0x00ff000000000000ULL) | + ((x << 24) & 0x0000ff0000000000ULL) | + ((x << 8) & 0x000000ff00000000ULL) | + ((x >> 8) & 0x00000000ff000000ULL) | + ((x >> 24) & 0x0000000000ff0000ULL) | + ((x >> 40) & 0x000000000000ff00ULL) | + ((x >> 56) & 0x00000000000000ffULL); +} +#endif + + +/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */ +#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3)) + +XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[0] + | ((xxh_u64)bytePtr[1] << 8) + | ((xxh_u64)bytePtr[2] << 16) + | ((xxh_u64)bytePtr[3] << 24) + | ((xxh_u64)bytePtr[4] << 32) + | ((xxh_u64)bytePtr[5] << 40) + | ((xxh_u64)bytePtr[6] << 48) + | ((xxh_u64)bytePtr[7] << 56); +} + +XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr) +{ + const xxh_u8* bytePtr = (const xxh_u8 *)memPtr; + return bytePtr[7] + | ((xxh_u64)bytePtr[6] << 8) + | ((xxh_u64)bytePtr[5] << 16) + | ((xxh_u64)bytePtr[4] << 24) + | ((xxh_u64)bytePtr[3] << 32) + | ((xxh_u64)bytePtr[2] << 40) + | ((xxh_u64)bytePtr[1] << 48) + | ((xxh_u64)bytePtr[0] << 56); +} + +#else +XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr)); +} + +static xxh_u64 XXH_readBE64(const void* ptr) +{ + return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr); +} +#endif + +XXH_FORCE_INLINE xxh_u64 +XXH_readLE64_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) + return XXH_readLE64(ptr); + else + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr); +} + + +/******* xxh64 *******/ +/*! + * @} + * @defgroup XXH64_impl XXH64 implementation + * @ingroup impl + * + * Details on the XXH64 implementation. + * @{ + */ +/* #define rather that static const, to be used as initializers */ +#define XXH_PRIME64_1 0x9E3779B185EBCA87ULL /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */ +#define XXH_PRIME64_2 0xC2B2AE3D27D4EB4FULL /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */ +#define XXH_PRIME64_3 0x165667B19E3779F9ULL /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */ +#define XXH_PRIME64_4 0x85EBCA77C2B2AE63ULL /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */ +#define XXH_PRIME64_5 0x27D4EB2F165667C5ULL /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */ + +#ifdef XXH_OLD_NAMES +# define PRIME64_1 XXH_PRIME64_1 +# define PRIME64_2 XXH_PRIME64_2 +# define PRIME64_3 XXH_PRIME64_3 +# define PRIME64_4 XXH_PRIME64_4 +# define PRIME64_5 XXH_PRIME64_5 +#endif + +/*! @copydoc XXH32_round */ +static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input) +{ + acc += input * XXH_PRIME64_2; + acc = XXH_rotl64(acc, 31); + acc *= XXH_PRIME64_1; +#if (defined(__AVX512F__)) && !defined(XXH_ENABLE_AUTOVECTORIZE) + /* + * DISABLE AUTOVECTORIZATION: + * A compiler fence is used to prevent GCC and Clang from + * autovectorizing the XXH64 loop (pragmas and attributes don't work for some + * reason) without globally disabling AVX512. + * + * Autovectorization of XXH64 tends to be detrimental, + * though the exact outcome may change depending on exact cpu and compiler version. + * For information, it has been reported as detrimental for Skylake-X, + * but possibly beneficial for Zen4. + * + * The default is to disable auto-vectorization, + * but you can select to enable it instead using `XXH_ENABLE_AUTOVECTORIZE` build variable. + */ + XXH_COMPILER_GUARD(acc); +#endif + return acc; +} + +static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val) +{ + val = XXH64_round(0, val); + acc ^= val; + acc = acc * XXH_PRIME64_1 + XXH_PRIME64_4; + return acc; +} + +/*! @copydoc XXH32_avalanche */ +static xxh_u64 XXH64_avalanche(xxh_u64 hash) +{ + hash ^= hash >> 33; + hash *= XXH_PRIME64_2; + hash ^= hash >> 29; + hash *= XXH_PRIME64_3; + hash ^= hash >> 32; + return hash; +} + + +#define XXH_get64bits(p) XXH_readLE64_align(p, align) + +/*! + * @internal + * @brief Sets up the initial accumulator state for XXH64(). + */ +XXH_FORCE_INLINE void +XXH64_initAccs(xxh_u64 *acc, xxh_u64 seed) +{ + XXH_ASSERT(acc != NULL); + acc[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2; + acc[1] = seed + XXH_PRIME64_2; + acc[2] = seed + 0; + acc[3] = seed - XXH_PRIME64_1; +} + +/*! + * @internal + * @brief Consumes a block of data for XXH64(). + * + * @return the end input pointer. + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH64_consumeLong( + xxh_u64 *XXH_RESTRICT acc, + xxh_u8 const *XXH_RESTRICT input, + size_t len, + XXH_alignment align +) +{ + const xxh_u8* const bEnd = input + len; + const xxh_u8* const limit = bEnd - 31; + XXH_ASSERT(acc != NULL); + XXH_ASSERT(input != NULL); + XXH_ASSERT(len >= 32); + do { + /* reroll on 32-bit */ + if (sizeof(void *) < sizeof(xxh_u64)) { + size_t i; + for (i = 0; i < 4; i++) { + acc[i] = XXH64_round(acc[i], XXH_get64bits(input)); + input += 8; + } + } else { + acc[0] = XXH64_round(acc[0], XXH_get64bits(input)); input += 8; + acc[1] = XXH64_round(acc[1], XXH_get64bits(input)); input += 8; + acc[2] = XXH64_round(acc[2], XXH_get64bits(input)); input += 8; + acc[3] = XXH64_round(acc[3], XXH_get64bits(input)); input += 8; + } + } while (input < limit); + + return input; +} + +/*! + * @internal + * @brief Merges the accumulator lanes together for XXH64() + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u64 +XXH64_mergeAccs(const xxh_u64 *acc) +{ + XXH_ASSERT(acc != NULL); + { + xxh_u64 h64 = XXH_rotl64(acc[0], 1) + XXH_rotl64(acc[1], 7) + + XXH_rotl64(acc[2], 12) + XXH_rotl64(acc[3], 18); + /* reroll on 32-bit */ + if (sizeof(void *) < sizeof(xxh_u64)) { + size_t i; + for (i = 0; i < 4; i++) { + h64 = XXH64_mergeRound(h64, acc[i]); + } + } else { + h64 = XXH64_mergeRound(h64, acc[0]); + h64 = XXH64_mergeRound(h64, acc[1]); + h64 = XXH64_mergeRound(h64, acc[2]); + h64 = XXH64_mergeRound(h64, acc[3]); + } + return h64; + } +} + +/*! + * @internal + * @brief Processes the last 0-31 bytes of @p ptr. + * + * There may be up to 31 bytes remaining to consume from the input. + * This final stage will digest them to ensure that all input bytes are present + * in the final mix. + * + * @param hash The hash to finalize. + * @param ptr The pointer to the remaining input. + * @param len The remaining length, modulo 32. + * @param align Whether @p ptr is aligned. + * @return The finalized hash + * @see XXH32_finalize(). + */ +XXH_STATIC XXH_PUREF xxh_u64 +XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ + if (ptr==NULL) XXH_ASSERT(len == 0); + len &= 31; + while (len >= 8) { + xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr)); + ptr += 8; + hash ^= k1; + hash = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4; + len -= 8; + } + if (len >= 4) { + hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1; + ptr += 4; + hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3; + len -= 4; + } + while (len > 0) { + hash ^= (*ptr++) * XXH_PRIME64_5; + hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1; + --len; + } + return XXH64_avalanche(hash); +} + +#ifdef XXH_OLD_NAMES +# define PROCESS1_64 XXH_PROCESS1_64 +# define PROCESS4_64 XXH_PROCESS4_64 +# define PROCESS8_64 XXH_PROCESS8_64 +#else +# undef XXH_PROCESS1_64 +# undef XXH_PROCESS4_64 +# undef XXH_PROCESS8_64 +#endif + +/*! + * @internal + * @brief The implementation for @ref XXH64(). + * + * @param input , len , seed Directly passed from @ref XXH64(). + * @param align Whether @p input is aligned. + * @return The calculated hash. + */ +XXH_FORCE_INLINE XXH_PUREF xxh_u64 +XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align) +{ + xxh_u64 h64; + if (input==NULL) XXH_ASSERT(len == 0); + + if (len>=32) { /* Process a large block of data */ + xxh_u64 acc[4]; + XXH64_initAccs(acc, seed); + + input = XXH64_consumeLong(acc, input, len, align); + + h64 = XXH64_mergeAccs(acc); + } else { + h64 = seed + XXH_PRIME64_5; + } + + h64 += (xxh_u64) len; + + return XXH64_finalize(h64, input, len, align); +} + + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ +#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2 + /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ + XXH64_state_t state; + XXH64_reset(&state, seed); + XXH64_update(&state, (const xxh_u8*)input, len); + return XXH64_digest(&state); +#else + if (XXH_FORCE_ALIGN_CHECK) { + if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */ + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); + } } + + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); + +#endif +} + +/******* Hash Streaming *******/ +#ifndef XXH_NO_STREAM +/*! @ingroup XXH64_family*/ +XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void) +{ + return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t)); +} +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr) +{ + XXH_free(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState) +{ + XXH_memcpy(dstState, srcState, sizeof(*dstState)); +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed) +{ + XXH_ASSERT(statePtr != NULL); + XXH_memset(statePtr, 0, sizeof(*statePtr)); + XXH64_initAccs(statePtr->acc, seed); + return XXH_OK; +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH_errorcode +XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + state->total_len += len; + + XXH_ASSERT(state->bufferedSize <= sizeof(state->buffer)); + if (len < sizeof(state->buffer) - state->bufferedSize) { /* fill in tmp buffer */ + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + { const xxh_u8* xinput = (const xxh_u8*)input; + const xxh_u8* const bEnd = xinput + len; + + if (state->bufferedSize) { /* non-empty buffer => complete first */ + XXH_memcpy(state->buffer + state->bufferedSize, xinput, sizeof(state->buffer) - state->bufferedSize); + xinput += sizeof(state->buffer) - state->bufferedSize; + /* and process one round */ + (void)XXH64_consumeLong(state->acc, state->buffer, sizeof(state->buffer), XXH_aligned); + state->bufferedSize = 0; + } + + XXH_ASSERT(xinput <= bEnd); + if ((size_t)(bEnd - xinput) >= sizeof(state->buffer)) { + /* Process the remaining data */ + xinput = XXH64_consumeLong(state->acc, xinput, (size_t)(bEnd - xinput), XXH_unaligned); + } + + if (xinput < bEnd) { + /* Copy the leftover to the tmp buffer */ + XXH_memcpy(state->buffer, xinput, (size_t)(bEnd-xinput)); + state->bufferedSize = (unsigned)(bEnd-xinput); + } + } + + return XXH_OK; +} + + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state) +{ + xxh_u64 h64; + + if (state->total_len >= 32) { + h64 = XXH64_mergeAccs(state->acc); + } else { + h64 = state->acc[2] /*seed*/ + XXH_PRIME64_5; + } + + h64 += (xxh_u64) state->total_len; + + return XXH64_finalize(h64, state->buffer, (size_t)state->total_len, XXH_aligned); +} +#endif /* !XXH_NO_STREAM */ + +/******* Canonical representation *******/ + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash); + XXH_memcpy(dst, &hash, sizeof(*dst)); +} + +/*! @ingroup XXH64_family */ +XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src) +{ + return XXH_readBE64(src); +} + +#ifndef XXH_NO_XXH3 + +/* ********************************************************************* +* XXH3 +* New generation hash designed for speed on small keys and vectorization +************************************************************************ */ +/*! + * @} + * @defgroup XXH3_impl XXH3 implementation + * @ingroup impl + * @{ + */ + +/* === Compiler specifics === */ + + +#if (defined(__GNUC__) && (__GNUC__ >= 3)) \ + || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \ + || defined(__clang__) +# define XXH_likely(x) __builtin_expect(x, 1) +# define XXH_unlikely(x) __builtin_expect(x, 0) +#else +# define XXH_likely(x) (x) +# define XXH_unlikely(x) (x) +#endif + +#ifndef XXH_HAS_INCLUDE +# ifdef __has_include +/* + * Not defined as XXH_HAS_INCLUDE(x) (function-like) because + * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion) + */ +# define XXH_HAS_INCLUDE __has_include +# else +# define XXH_HAS_INCLUDE(x) 0 +# endif +#endif + +#if defined(__GNUC__) || defined(__clang__) +# if defined(__ARM_FEATURE_SVE) +# include +# endif +# if defined(__ARM_NEON__) || defined(__ARM_NEON) \ + || (defined(_M_ARM) && _M_ARM >= 7) \ + || defined(_M_ARM64) || defined(_M_ARM64EC) \ + || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE()) /* WASM SIMD128 via SIMDe */ +# define inline __inline__ /* circumvent a clang bug */ +# include +# undef inline +# elif defined(__AVX2__) +# include +# elif defined(__SSE2__) +# include +# elif defined(__loongarch_asx) +# include +# include +# elif defined(__loongarch_sx) +# include +# endif +#endif + +#if defined(_MSC_VER) +# include +#endif + +/* + * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while + * remaining a true 64-bit/128-bit hash function. + * + * This is done by prioritizing a subset of 64-bit operations that can be + * emulated without too many steps on the average 32-bit machine. + * + * For example, these two lines seem similar, and run equally fast on 64-bit: + * + * xxh_u64 x; + * x ^= (x >> 47); // good + * x ^= (x >> 13); // bad + * + * However, to a 32-bit machine, there is a major difference. + * + * x ^= (x >> 47) looks like this: + * + * x.lo ^= (x.hi >> (47 - 32)); + * + * while x ^= (x >> 13) looks like this: + * + * // note: funnel shifts are not usually cheap. + * x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13)); + * x.hi ^= (x.hi >> 13); + * + * The first one is significantly faster than the second, simply because the + * shift is larger than 32. This means: + * - All the bits we need are in the upper 32 bits, so we can ignore the lower + * 32 bits in the shift. + * - The shift result will always fit in the lower 32 bits, and therefore, + * we can ignore the upper 32 bits in the xor. + * + * Thanks to this optimization, XXH3 only requires these features to be efficient: + * + * - Usable unaligned access + * - A 32-bit or 64-bit ALU + * - If 32-bit, a decent ADC instruction + * - A 32 or 64-bit multiply with a 64-bit result + * - For the 128-bit variant, a decent byteswap helps short inputs. + * + * The first two are already required by XXH32, and almost all 32-bit and 64-bit + * platforms which can run XXH32 can run XXH3 efficiently. + * + * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one + * notable exception. + * + * First of all, Thumb-1 lacks support for the UMULL instruction which + * performs the important long multiply. This means numerous __aeabi_lmul + * calls. + * + * Second of all, the 8 functional registers are just not enough. + * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need + * Lo registers, and this shuffling results in thousands more MOVs than A32. + * + * A32 and T32 don't have this limitation. They can access all 14 registers, + * do a 32->64 multiply with UMULL, and the flexible operand allowing free + * shifts is helpful, too. + * + * Therefore, we do a quick sanity check. + * + * If compiling Thumb-1 for a target which supports ARM instructions, we will + * emit a warning, as it is not a "sane" platform to compile for. + * + * Usually, if this happens, it is because of an accident and you probably need + * to specify -march, as you likely meant to compile for a newer architecture. + * + * Credit: large sections of the vectorial and asm source code paths + * have been contributed by @easyaspi314 + */ +#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM) +# warning "XXH3 is highly inefficient without ARM or Thumb-2." +#endif + +/* ========================================== + * Vectorization detection + * ========================================== */ + +#ifdef XXH_DOXYGEN +/*! + * @ingroup tuning + * @brief Overrides the vectorization implementation chosen for XXH3. + * + * Can be defined to 0 to disable SIMD, + * or any other authorized value of @ref XXH_VECTOR. + * + * If this is not defined, it uses predefined macros to determine the best + * implementation. + */ +# define XXH_VECTOR XXH_SCALAR +/*! + * @ingroup tuning + * @brief Selects the minimum alignment for XXH3's accumulators. + * + * When using SIMD, this should match the alignment required for said vector + * type, so, for example, 32 for AVX2. + * + * Default: Auto detected. + */ +# define XXH_ACC_ALIGN 8 +#endif + +/* Actual definition */ +#ifndef XXH_DOXYGEN +#endif + +#ifndef XXH_VECTOR /* can be defined on command line */ +# if defined(__ARM_FEATURE_SVE) +# define XXH_VECTOR XXH_SVE +# elif ( \ + defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \ + || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \ + || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE()) /* wasm simd128 via SIMDe */ \ + ) && ( \ + defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ + ) +# define XXH_VECTOR XXH_NEON +# elif defined(__AVX512F__) +# define XXH_VECTOR XXH_AVX512 +# elif defined(__AVX2__) +# define XXH_VECTOR XXH_AVX2 +# elif defined(__SSE2__) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2)) +# define XXH_VECTOR XXH_SSE2 +# elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \ + || (defined(__s390x__) && defined(__VEC__)) \ + && defined(__GNUC__) /* TODO: IBM XL */ +# define XXH_VECTOR XXH_VSX +# elif defined(__loongarch_asx) +# define XXH_VECTOR XXH_LASX +# elif defined(__loongarch_sx) +# define XXH_VECTOR XXH_LSX +# else +# define XXH_VECTOR XXH_SCALAR +# endif +#endif + +/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */ +#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE) +# ifdef _MSC_VER +# pragma warning(once : 4606) +# else +# warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead." +# endif +# undef XXH_VECTOR +# define XXH_VECTOR XXH_SCALAR +#endif + +/* + * Controls the alignment of the accumulator, + * for compatibility with aligned vector loads, which are usually faster. + */ +#ifndef XXH_ACC_ALIGN +# if defined(XXH_X86DISPATCH) +# define XXH_ACC_ALIGN 64 /* for compatibility with avx512 */ +# elif XXH_VECTOR == XXH_SCALAR /* scalar */ +# define XXH_ACC_ALIGN 8 +# elif XXH_VECTOR == XXH_SSE2 /* sse2 */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_AVX2 /* avx2 */ +# define XXH_ACC_ALIGN 32 +# elif XXH_VECTOR == XXH_NEON /* neon */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_VSX /* vsx */ +# define XXH_ACC_ALIGN 16 +# elif XXH_VECTOR == XXH_AVX512 /* avx512 */ +# define XXH_ACC_ALIGN 64 +# elif XXH_VECTOR == XXH_SVE /* sve */ +# define XXH_ACC_ALIGN 64 +# elif XXH_VECTOR == XXH_LASX /* lasx */ +# define XXH_ACC_ALIGN 64 +# elif XXH_VECTOR == XXH_LSX /* lsx */ +# define XXH_ACC_ALIGN 64 +# endif +#endif + +#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \ + || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512 +# define XXH_SEC_ALIGN XXH_ACC_ALIGN +#elif XXH_VECTOR == XXH_SVE +# define XXH_SEC_ALIGN XXH_ACC_ALIGN +#else +# define XXH_SEC_ALIGN 8 +#endif + +#if defined(__GNUC__) || defined(__clang__) +# define XXH_ALIASING __attribute__((__may_alias__)) +#else +# define XXH_ALIASING /* nothing */ +#endif + +/* + * UGLY HACK: + * GCC usually generates the best code with -O3 for xxHash. + * + * However, when targeting AVX2, it is overzealous in its unrolling resulting + * in code roughly 3/4 the speed of Clang. + * + * There are other issues, such as GCC splitting _mm256_loadu_si256 into + * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which + * only applies to Sandy and Ivy Bridge... which don't even support AVX2. + * + * That is why when compiling the AVX2 version, it is recommended to use either + * -O2 -mavx2 -march=haswell + * or + * -O2 -mavx2 -mno-avx256-split-unaligned-load + * for decent performance, or to use Clang instead. + * + * Fortunately, we can control the first one with a pragma that forces GCC into + * -O2, but the other one we can't control without "failed to inline always + * inline function due to target mismatch" warnings. + */ +#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \ + && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \ + && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */ +# pragma GCC push_options +# pragma GCC optimize("-O2") +#endif + +#if XXH_VECTOR == XXH_NEON + +/* + * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3 + * optimizes out the entire hashLong loop because of the aliasing violation. + * + * However, GCC is also inefficient at load-store optimization with vld1q/vst1q, + * so the only option is to mark it as aliasing. + */ +typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING; + +/*! + * @internal + * @brief `vld1q_u64` but faster and alignment-safe. + * + * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only + * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86). + * + * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it + * prohibits load-store optimizations. Therefore, a direct dereference is used. + * + * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe + * unaligned load. + */ +#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) +XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */ +{ + return *(xxh_aliasing_uint64x2_t const *)ptr; +} +#else +XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) +{ + return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr)); +} +#endif + +/*! + * @internal + * @brief `vmlal_u32` on low and high halves of a vector. + * + * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with + * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32` + * with `vmlal_u32`. + */ +#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11 +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + /* Inline assembly is the only way */ + __asm__("umlal %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs)); + return acc; +} +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + /* This intrinsic works as expected */ + return vmlal_high_u32(acc, lhs, rhs); +} +#else +/* Portable intrinsic versions */ +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs)); +} +/*! @copydoc XXH_vmlal_low_u32 + * Assume the compiler converts this to vmlal_high_u32 on aarch64 */ +XXH_FORCE_INLINE uint64x2_t +XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs) +{ + return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs)); +} +#endif + +/*! + * @ingroup tuning + * @brief Controls the NEON to scalar ratio for XXH3 + * + * This can be set to 2, 4, 6, or 8. + * + * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used. + * + * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those + * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU + * bandwidth. + * + * This is even more noticeable on the more advanced cores like the Cortex-A76 which + * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once. + * + * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes + * and 2 scalar lanes, which is chosen by default. + * + * This does not apply to Apple processors or 32-bit processors, which run better with + * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes. + * + * This change benefits CPUs with large micro-op buffers without negatively affecting + * most other CPUs: + * + * | Chipset | Dispatch type | NEON only | 6:2 hybrid | Diff. | + * |:----------------------|:--------------------|----------:|-----------:|------:| + * | Snapdragon 730 (A76) | 2 NEON/8 micro-ops | 8.8 GB/s | 10.1 GB/s | ~16% | + * | Snapdragon 835 (A73) | 2 NEON/3 micro-ops | 5.1 GB/s | 5.3 GB/s | ~5% | + * | Marvell PXA1928 (A53) | In-order dual-issue | 1.9 GB/s | 1.9 GB/s | 0% | + * | Apple M1 | 4 NEON/8 micro-ops | 37.3 GB/s | 36.1 GB/s | ~-3% | + * + * It also seems to fix some bad codegen on GCC, making it almost as fast as clang. + * + * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning + * it effectively becomes worse 4. + * + * @see XXH3_accumulate_512_neon() + */ +# ifndef XXH3_NEON_LANES +# if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \ + && !defined(__APPLE__) && XXH_SIZE_OPT <= 0 +# define XXH3_NEON_LANES 6 +# else +# define XXH3_NEON_LANES XXH_ACC_NB +# endif +# endif +#endif /* XXH_VECTOR == XXH_NEON */ + +/* + * VSX and Z Vector helpers. + * + * This is very messy, and any pull requests to clean this up are welcome. + * + * There are a lot of problems with supporting VSX and s390x, due to + * inconsistent intrinsics, spotty coverage, and multiple endiannesses. + */ +#if XXH_VECTOR == XXH_VSX +/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`, + * and `pixel`. This is a problem for obvious reasons. + * + * These keywords are unnecessary; the spec literally says they are + * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd + * after including the header. + * + * We use pragma push_macro/pop_macro to keep the namespace clean. */ +# pragma push_macro("bool") +# pragma push_macro("vector") +# pragma push_macro("pixel") +/* silence potential macro redefined warnings */ +# undef bool +# undef vector +# undef pixel + +# if defined(__s390x__) +# include +# else +# include +# endif + +/* Restore the original macro values, if applicable. */ +# pragma pop_macro("pixel") +# pragma pop_macro("vector") +# pragma pop_macro("bool") + +typedef __vector unsigned long long xxh_u64x2; +typedef __vector unsigned char xxh_u8x16; +typedef __vector unsigned xxh_u32x4; + +/* + * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue. + */ +typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING; + +# ifndef XXH_VSX_BE +# if defined(__BIG_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +# define XXH_VSX_BE 1 +# elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__ +# warning "-maltivec=be is not recommended. Please use native endianness." +# define XXH_VSX_BE 1 +# else +# define XXH_VSX_BE 0 +# endif +# endif /* !defined(XXH_VSX_BE) */ + +# if XXH_VSX_BE +# if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__)) +# define XXH_vec_revb vec_revb +# else +/*! + * A polyfill for POWER9's vec_revb(). + */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val) +{ + xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00, + 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 }; + return vec_perm(val, val, vByteSwap); +} +# endif +# endif /* XXH_VSX_BE */ + +/*! + * Performs an unaligned vector load and byte swaps it on big endian. + */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr) +{ + xxh_u64x2 ret; + XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2)); +# if XXH_VSX_BE + ret = XXH_vec_revb(ret); +# endif + return ret; +} + +/* + * vec_mulo and vec_mule are very problematic intrinsics on PowerPC + * + * These intrinsics weren't added until GCC 8, despite existing for a while, + * and they are endian dependent. Also, their meaning swap depending on version. + * */ +# if defined(__s390x__) + /* s390x is always big endian, no issue on this platform */ +# define XXH_vec_mulo vec_mulo +# define XXH_vec_mule vec_mule +# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__) +/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */ + /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */ +# define XXH_vec_mulo __builtin_altivec_vmulouw +# define XXH_vec_mule __builtin_altivec_vmuleuw +# else +/* gcc needs inline assembly */ +/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */ +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b) +{ + xxh_u64x2 result; + __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b)); + return result; +} +XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b) +{ + xxh_u64x2 result; + __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b)); + return result; +} +# endif /* XXH_vec_mulo, XXH_vec_mule */ +#endif /* XXH_VECTOR == XXH_VSX */ + +#if XXH_VECTOR == XXH_SVE +#define ACCRND(acc, offset) \ +do { \ + svuint64_t input_vec = svld1_u64(mask, xinput + offset); \ + svuint64_t secret_vec = svld1_u64(mask, xsecret + offset); \ + svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec); \ + svuint64_t swapped = svtbl_u64(input_vec, kSwap); \ + svuint64_t mixed_lo = svextw_u64_x(mask, mixed); \ + svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32); \ + svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \ + acc = svadd_u64_x(mask, acc, mul); \ +} while (0) +#endif /* XXH_VECTOR == XXH_SVE */ + +/* prefetch + * can be disabled, by declaring XXH_NO_PREFETCH build macro */ +#if defined(XXH_NO_PREFETCH) +# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */ +#else +# if XXH_SIZE_OPT >= 1 +# define XXH_PREFETCH(ptr) (void)(ptr) +# elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */ +# include /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ +# define XXH_PREFETCH(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) +# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) +# define XXH_PREFETCH(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) +# else +# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */ +# endif +#endif /* XXH_NO_PREFETCH */ + + +/* ========================================== + * XXH3 default settings + * ========================================== */ + +#define XXH_SECRET_DEFAULT_SIZE 192 /* minimum XXH3_SECRET_SIZE_MIN */ + +#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN) +# error "default keyset is not large enough" +#endif + +/*! + * @internal + * @def XXH3_kSecret + * @brief Pseudorandom secret taken directly from FARSH. */ +XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = { + 0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c, + 0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f, + 0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21, + 0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c, + 0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3, + 0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8, + 0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d, + 0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64, + 0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb, + 0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e, + 0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce, + 0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e, +}; + +static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL; /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */ +static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL; /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */ + +#ifdef XXH_OLD_NAMES +# define kSecret XXH3_kSecret +#endif + +#ifdef XXH_DOXYGEN +/*! + * @brief Calculates a 32-bit to 64-bit long multiply. + * + * Implemented as a macro. + * + * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't + * need to (but it shouldn't need to anyways, it is about 7 instructions to do + * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we + * use that instead of the normal method. + * + * If you are compiling for platforms like Thumb-1 and don't have a better option, + * you may also want to write your own long multiply routine here. + * + * @param x, y Numbers to be multiplied + * @return 64-bit product of the low 32 bits of @p x and @p y. + */ +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64(xxh_u64 x, xxh_u64 y) +{ + return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF); +} +#elif defined(_MSC_VER) && defined(_M_IX86) +# define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y)) +#else +/* + * Downcast + upcast is usually better than masking on older compilers like + * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers. + * + * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands + * and perform a full 64x64 multiply -- entirely redundant on 32-bit. + */ +# define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y)) +#endif + +/*! + * @brief Calculates a 64->128-bit long multiply. + * + * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar + * version. + * + * @param lhs , rhs The 64-bit integers to be multiplied + * @return The 128-bit result represented in an @ref XXH128_hash_t. + */ +static XXH128_hash_t +XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs) +{ + /* + * GCC/Clang __uint128_t method. + * + * On most 64-bit targets, GCC and Clang define a __uint128_t type. + * This is usually the best way as it usually uses a native long 64-bit + * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64. + * + * Usually. + * + * Despite being a 32-bit platform, Clang (and emscripten) define this type + * despite not having the arithmetic for it. This results in a laggy + * compiler builtin call which calculates a full 128-bit multiply. + * In that case it is best to use the portable one. + * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677 + */ +#if (defined(__GNUC__) || defined(__clang__)) && defined(DISABLED) \ + && defined(__SIZEOF_INT128__) \ + || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128) + + __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs; + XXH128_hash_t r128; + r128.low64 = (xxh_u64)(product); + r128.high64 = (xxh_u64)(product >> 64); + return r128; + + /* + * MSVC for x64's _umul128 method. + * + * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct); + * + * This compiles to single operand MUL on x64. + */ +#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC) + +#ifndef _MSC_VER +# pragma intrinsic(_umul128) +#endif + xxh_u64 product_high; + xxh_u64 const product_low = _umul128(lhs, rhs, &product_high); + XXH128_hash_t r128; + r128.low64 = product_low; + r128.high64 = product_high; + return r128; + + /* + * MSVC for ARM64's __umulh method. + * + * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method. + */ +#elif defined(_M_ARM64) || defined(_M_ARM64EC) + +#ifndef _MSC_VER +# pragma intrinsic(__umulh) +#endif + XXH128_hash_t r128; + r128.low64 = lhs * rhs; + r128.high64 = __umulh(lhs, rhs); + return r128; + +#else + /* + * Portable scalar method. Optimized for 32-bit and 64-bit ALUs. + * + * This is a fast and simple grade school multiply, which is shown below + * with base 10 arithmetic instead of base 0x100000000. + * + * 9 3 // D2 lhs = 93 + * x 7 5 // D2 rhs = 75 + * ---------- + * 1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15 + * 4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45 + * 2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21 + * + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63 + * --------- + * 2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27 + * + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67 + * --------- + * 6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975 + * + * The reasons for adding the products like this are: + * 1. It avoids manual carry tracking. Just like how + * (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX. + * This avoids a lot of complexity. + * + * 2. It hints for, and on Clang, compiles to, the powerful UMAAL + * instruction available in ARM's Digital Signal Processing extension + * in 32-bit ARMv6 and later, which is shown below: + * + * void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm) + * { + * xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm; + * *RdLo = (xxh_u32)(product & 0xFFFFFFFF); + * *RdHi = (xxh_u32)(product >> 32); + * } + * + * This instruction was designed for efficient long multiplication, and + * allows this to be calculated in only 4 instructions at speeds + * comparable to some 64-bit ALUs. + * + * 3. It isn't terrible on other platforms. Usually this will be a couple + * of 32-bit ADD/ADCs. + */ + + /* First calculate all of the cross products. */ + xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF); + xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32, rhs & 0xFFFFFFFF); + xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32); + xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32, rhs >> 32); + + /* Now add the products together. These will never overflow. */ + xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi; + xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32) + hi_hi; + xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF); + + XXH128_hash_t r128; + r128.low64 = lower; + r128.high64 = upper; + return r128; +#endif +} + +/*! + * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it. + * + * The reason for the separate function is to prevent passing too many structs + * around by value. This will hopefully inline the multiply, but we don't force it. + * + * @param lhs , rhs The 64-bit integers to multiply + * @return The low 64 bits of the product XOR'd by the high 64 bits. + * @see XXH_mult64to128() + */ +static xxh_u64 +XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs) +{ + XXH128_hash_t product = XXH_mult64to128(lhs, rhs); + return product.low64 ^ product.high64; +} + +/*! Seems to produce slightly better code on GCC for some reason. */ +XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift) +{ + XXH_ASSERT(0 <= shift && shift < 64); + return v64 ^ (v64 >> shift); +} + +/* + * This is a fast avalanche stage, + * suitable when input bits are already partially mixed + */ +static XXH64_hash_t XXH3_avalanche(xxh_u64 h64) +{ + h64 = XXH_xorshift64(h64, 37); + h64 *= PRIME_MX1; + h64 = XXH_xorshift64(h64, 32); + return h64; +} + +/* + * This is a stronger avalanche, + * inspired by Pelle Evensen's rrmxmx + * preferable when input has not been previously mixed + */ +static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) +{ + /* this mix is inspired by Pelle Evensen's rrmxmx */ + h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24); + h64 *= PRIME_MX2; + h64 ^= (h64 >> 35) + len ; + h64 *= PRIME_MX2; + return XXH_xorshift64(h64, 28); +} + + +/* ========================================== + * Short keys + * ========================================== + * One of the shortcomings of XXH32 and XXH64 was that their performance was + * sub-optimal on short lengths. It used an iterative algorithm which strongly + * favored lengths that were a multiple of 4 or 8. + * + * Instead of iterating over individual inputs, we use a set of single shot + * functions which piece together a range of lengths and operate in constant time. + * + * Additionally, the number of multiplies has been significantly reduced. This + * reduces latency, especially when emulating 64-bit multiplies on 32-bit. + * + * Depending on the platform, this may or may not be faster than XXH32, but it + * is almost guaranteed to be faster than XXH64. + */ + +/* + * At very short lengths, there isn't enough input to fully hide secrets, or use + * the entire secret. + * + * There is also only a limited amount of mixing we can do before significantly + * impacting performance. + * + * Therefore, we use different sections of the secret and always mix two secret + * samples with an XOR. This should have no effect on performance on the + * seedless or withSeed variants because everything _should_ be constant folded + * by modern compilers. + * + * The XOR mixing hides individual parts of the secret and increases entropy. + * + * This adds an extra layer of strength for custom secrets. + */ +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(1 <= len && len <= 3); + XXH_ASSERT(secret != NULL); + /* + * len = 1: combined = { input[0], 0x01, input[0], input[0] } + * len = 2: combined = { input[1], 0x02, input[0], input[1] } + * len = 3: combined = { input[2], 0x03, input[0], input[1] } + */ + { xxh_u8 const c1 = input[0]; + xxh_u8 const c2 = input[len >> 1]; + xxh_u8 const c3 = input[len - 1]; + xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) + | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8); + xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed; + xxh_u64 const keyed = (xxh_u64)combined ^ bitflip; + return XXH64_avalanche(keyed); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(4 <= len && len <= 8); + seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32; + { xxh_u32 const input1 = XXH_readLE32(input); + xxh_u32 const input2 = XXH_readLE32(input + len - 4); + xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed; + xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32); + xxh_u64 const keyed = input64 ^ bitflip; + return XXH3_rrmxmx(keyed, len); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(9 <= len && len <= 16); + { xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed; + xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed; + xxh_u64 const input_lo = XXH_readLE64(input) ^ bitflip1; + xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2; + xxh_u64 const acc = len + + XXH_swap64(input_lo) + input_hi + + XXH3_mul128_fold64(input_lo, input_hi); + return XXH3_avalanche(acc); + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(len <= 16); + { if (XXH_likely(len > 8)) return XXH3_len_9to16_64b(input, len, secret, seed); + if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed); + if (len) return XXH3_len_1to3_64b(input, len, secret, seed); + return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64))); + } +} + +/* + * DISCLAIMER: There are known *seed-dependent* multicollisions here due to + * multiplication by zero, affecting hashes of lengths 17 to 240. + * + * However, they are very unlikely. + * + * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all + * unseeded non-cryptographic hashes, it does not attempt to defend itself + * against specially crafted inputs, only random inputs. + * + * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes + * cancelling out the secret is taken an arbitrary number of times (addressed + * in XXH3_accumulate_512), this collision is very unlikely with random inputs + * and/or proper seeding: + * + * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a + * function that is only called up to 16 times per hash with up to 240 bytes of + * input. + * + * This is not too bad for a non-cryptographic hash function, especially with + * only 64 bit outputs. + * + * The 128-bit variant (which trades some speed for strength) is NOT affected + * by this, although it is always a good idea to use a proper seed if you care + * about strength. + */ +XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input, + const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64) +{ +#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \ + && defined(__i386__) && defined(__SSE2__) /* x86 + SSE2 */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable like XXH32 hack */ + /* + * UGLY HACK: + * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in + * slower code. + * + * By forcing seed64 into a register, we disrupt the cost model and + * cause it to scalarize. See `XXH32_round()` + * + * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600, + * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on + * GCC 9.2, despite both emitting scalar code. + * + * GCC generates much better scalar code than Clang for the rest of XXH3, + * which is why finding a more optimal codepath is an interest. + */ + XXH_COMPILER_GUARD(seed64); +#endif + { xxh_u64 const input_lo = XXH_readLE64(input); + xxh_u64 const input_hi = XXH_readLE64(input+8); + return XXH3_mul128_fold64( + input_lo ^ (XXH_readLE64(secret) + seed64), + input_hi ^ (XXH_readLE64(secret+8) - seed64) + ); + } +} + +/* For mid range keys, XXH3 uses a Mum-hash variant. */ +XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(16 < len && len <= 128); + + { xxh_u64 acc = len * XXH_PRIME64_1; +#if XXH_SIZE_OPT >= 1 + /* Smaller and cleaner, but slightly slower. */ + unsigned int i = (unsigned int)(len - 1) / 32; + do { + acc += XXH3_mix16B(input+16 * i, secret+32*i, seed); + acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed); + } while (i-- != 0); +#else + if (len > 32) { + if (len > 64) { + if (len > 96) { + acc += XXH3_mix16B(input+48, secret+96, seed); + acc += XXH3_mix16B(input+len-64, secret+112, seed); + } + acc += XXH3_mix16B(input+32, secret+64, seed); + acc += XXH3_mix16B(input+len-48, secret+80, seed); + } + acc += XXH3_mix16B(input+16, secret+32, seed); + acc += XXH3_mix16B(input+len-32, secret+48, seed); + } + acc += XXH3_mix16B(input+0, secret+0, seed); + acc += XXH3_mix16B(input+len-16, secret+16, seed); +#endif + return XXH3_avalanche(acc); + } +} + +XXH_NO_INLINE XXH_PUREF XXH64_hash_t +XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + + #define XXH3_MIDSIZE_STARTOFFSET 3 + #define XXH3_MIDSIZE_LASTOFFSET 17 + + { xxh_u64 acc = len * XXH_PRIME64_1; + xxh_u64 acc_end; + unsigned int const nbRounds = (unsigned int)len / 16; + unsigned int i; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + for (i=0; i<8; i++) { + acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed); + } + /* last bytes */ + acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed); + XXH_ASSERT(nbRounds >= 8); + acc = XXH3_avalanche(acc); +#if defined(__clang__) /* Clang */ \ + && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */ + /* + * UGLY HACK: + * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86. + * In everywhere else, it uses scalar code. + * + * For 64->128-bit multiplies, even if the NEON was 100% optimal, it + * would still be slower than UMAAL (see XXH_mult64to128). + * + * Unfortunately, Clang doesn't handle the long multiplies properly and + * converts them to the nonexistent "vmulq_u64" intrinsic, which is then + * scalarized into an ugly mess of VMOV.32 instructions. + * + * This mess is difficult to avoid without turning autovectorization + * off completely, but they are usually relatively minor and/or not + * worth it to fix. + * + * This loop is the easiest to fix, as unlike XXH32, this pragma + * _actually works_ because it is a loop vectorization instead of an + * SLP vectorization. + */ + #pragma clang loop vectorize(disable) +#endif + for (i=8 ; i < nbRounds; i++) { + /* + * Prevents clang for unrolling the acc loop and interleaving with this one. + */ + XXH_COMPILER_GUARD(acc); + acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed); + } + return XXH3_avalanche(acc + acc_end); + } +} + + +/* ======= Long Keys ======= */ + +#define XXH_STRIPE_LEN 64 +#define XXH_SECRET_CONSUME_RATE 8 /* nb of secret bytes consumed at each accumulation */ +#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64)) + +#ifdef XXH_OLD_NAMES +# define STRIPE_LEN XXH_STRIPE_LEN +# define ACC_NB XXH_ACC_NB +#endif + +#ifndef XXH_PREFETCH_DIST +# ifdef __clang__ +# define XXH_PREFETCH_DIST 320 +# else +# if (XXH_VECTOR == XXH_AVX512) +# define XXH_PREFETCH_DIST 512 +# else +# define XXH_PREFETCH_DIST 384 +# endif +# endif /* __clang__ */ +#endif /* XXH_PREFETCH_DIST */ + +/* + * These macros are to generate an XXH3_accumulate() function. + * The two arguments select the name suffix and target attribute. + * + * The name of this symbol is XXH3_accumulate_() and it calls + * XXH3_accumulate_512_(). + * + * It may be useful to hand implement this function if the compiler fails to + * optimize the inline function. + */ +#define XXH3_ACCUMULATE_TEMPLATE(name) \ +void \ +XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc, \ + const xxh_u8* XXH_RESTRICT input, \ + const xxh_u8* XXH_RESTRICT secret, \ + size_t nbStripes) \ +{ \ + size_t n; \ + for (n = 0; n < nbStripes; n++ ) { \ + const xxh_u8* const in = input + n*XXH_STRIPE_LEN; \ + XXH_PREFETCH(in + XXH_PREFETCH_DIST); \ + XXH3_accumulate_512_##name( \ + acc, \ + in, \ + secret + n*XXH_SECRET_CONSUME_RATE); \ + } \ +} + + +XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64) +{ + if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64); + XXH_memcpy(dst, &v64, sizeof(v64)); +} + +/* Several intrinsic functions below are supposed to accept __int64 as argument, + * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ . + * However, several environments do not define __int64 type, + * requiring a workaround. + */ +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) + typedef int64_t xxh_i64; +#else + /* the following type must have a width of 64-bit */ + typedef long long xxh_i64; +#endif + + +/* + * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized. + * + * It is a hardened version of UMAC, based off of FARSH's implementation. + * + * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD + * implementations, and it is ridiculously fast. + * + * We harden it by mixing the original input to the accumulators as well as the product. + * + * This means that in the (relatively likely) case of a multiply by zero, the + * original input is preserved. + * + * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve + * cross-pollination, as otherwise the upper and lower halves would be + * essentially independent. + * + * This doesn't matter on 64-bit hashes since they all get merged together in + * the end, so we skip the extra step. + * + * Both XXH3_64bits and XXH3_128bits use this subroutine. + */ + +#if (XXH_VECTOR == XXH_AVX512) \ + || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0) + +#ifndef XXH_TARGET_AVX512 +# define XXH_TARGET_AVX512 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + __m512i* const xacc = (__m512i *) acc; + XXH_ASSERT((((size_t)acc) & 63) == 0); + XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i)); + + { + /* data_vec = input[0]; */ + __m512i const data_vec = _mm512_loadu_si512 (input); + /* key_vec = secret[0]; */ + __m512i const key_vec = _mm512_loadu_si512 (secret); + /* data_key = data_vec ^ key_vec; */ + __m512i const data_key = _mm512_xor_si512 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m512i const product = _mm512_mul_epu32 (data_key, data_key_lo); + /* xacc[0] += swap(data_vec); */ + __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2)); + __m512i const sum = _mm512_add_epi64(*xacc, data_swap); + /* xacc[0] += product; */ + *xacc = _mm512_add_epi64(product, sum); + } +} +XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512) + +/* + * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing. + * + * Multiplication isn't perfect, as explained by Google in HighwayHash: + * + * // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to + * // varying degrees. In descending order of goodness, bytes + * // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32. + * // As expected, the upper and lower bytes are much worse. + * + * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291 + * + * Since our algorithm uses a pseudorandom secret to add some variance into the + * mix, we don't need to (or want to) mix as often or as much as HighwayHash does. + * + * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid + * extraction. + * + * Both XXH3_64bits and XXH3_128bits use this subroutine. + */ + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 63) == 0); + XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i)); + { __m512i* const xacc = (__m512i*) acc; + const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1); + + /* xacc[0] ^= (xacc[0] >> 47) */ + __m512i const acc_vec = *xacc; + __m512i const shifted = _mm512_srli_epi64 (acc_vec, 47); + /* xacc[0] ^= secret; */ + __m512i const key_vec = _mm512_loadu_si512 (secret); + __m512i const data_key = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */); + + /* xacc[0] *= XXH_PRIME32_1; */ + __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32); + __m512i const prod_lo = _mm512_mul_epu32 (data_key, prime32); + __m512i const prod_hi = _mm512_mul_epu32 (data_key_hi, prime32); + *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32)); + } +} + +XXH_FORCE_INLINE XXH_TARGET_AVX512 void +XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0); + XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64); + XXH_ASSERT(((size_t)customSecret & 63) == 0); + (void)(&XXH_writeLE64); + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i); + __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64); + __m512i const seed = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos); + + const __m512i* const src = (const __m512i*) ((const void*) XXH3_kSecret); + __m512i* const dest = ( __m512i*) customSecret; + int i; + XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dest & 63) == 0); + for (i=0; i < nbRounds; ++i) { + dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_AVX2) \ + || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0) + +#ifndef XXH_TARGET_AVX2 +# define XXH_TARGET_AVX2 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void +XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { __m256i* const xacc = (__m256i *) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xinput = (const __m256i *) input; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xsecret = (const __m256i *) secret; + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) { + /* data_vec = xinput[i]; */ + __m256i const data_vec = _mm256_loadu_si256 (xinput+i); + /* key_vec = xsecret[i]; */ + __m256i const key_vec = _mm256_loadu_si256 (xsecret+i); + /* data_key = data_vec ^ key_vec; */ + __m256i const data_key = _mm256_xor_si256 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m256i const product = _mm256_mul_epu32 (data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2)); + __m256i const sum = _mm256_add_epi64(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = _mm256_add_epi64(product, sum); + } } +} +XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2) + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void +XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { __m256i* const xacc = (__m256i*) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */ + const __m256i* const xsecret = (const __m256i *) secret; + const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1); + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m256i const acc_vec = xacc[i]; + __m256i const shifted = _mm256_srli_epi64 (acc_vec, 47); + __m256i const data_vec = _mm256_xor_si256 (acc_vec, shifted); + /* xacc[i] ^= xsecret; */ + __m256i const key_vec = _mm256_loadu_si256 (xsecret+i); + __m256i const data_key = _mm256_xor_si256 (data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32); + __m256i const prod_lo = _mm256_mul_epu32 (data_key, prime32); + __m256i const prod_hi = _mm256_mul_epu32 (data_key_hi, prime32); + xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32)); + } + } +} + +XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0); + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6); + XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64); + (void)(&XXH_writeLE64); + XXH_PREFETCH(customSecret); + { __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64); + + const __m256i* const src = (const __m256i*) ((const void*) XXH3_kSecret); + __m256i* dest = ( __m256i*) customSecret; + +# if defined(__GNUC__) || defined(__clang__) + /* + * On GCC & Clang, marking 'dest' as modified will cause the compiler: + * - do not extract the secret from sse registers in the internal loop + * - use less common registers, and avoid pushing these reg into stack + */ + XXH_COMPILER_GUARD(dest); +# endif + XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dest & 31) == 0); + + /* GCC -O2 need unroll loop manually */ + dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed); + dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed); + dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed); + dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed); + dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed); + dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed); + } +} + +#endif + +/* x86dispatch always generates SSE2 */ +#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH) + +#ifndef XXH_TARGET_SSE2 +# define XXH_TARGET_SSE2 /* disable attribute target */ +#endif + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void +XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + /* SSE2 is just a half-scale version of the AVX2 version. */ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { __m128i* const xacc = (__m128i *) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xinput = (const __m128i *) input; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xsecret = (const __m128i *) secret; + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) { + /* data_vec = xinput[i]; */ + __m128i const data_vec = _mm_loadu_si128 (xinput+i); + /* key_vec = xsecret[i]; */ + __m128i const key_vec = _mm_loadu_si128 (xsecret+i); + /* data_key = data_vec ^ key_vec; */ + __m128i const data_key = _mm_xor_si128 (data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1)); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m128i const product = _mm_mul_epu32 (data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2)); + __m128i const sum = _mm_add_epi64(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = _mm_add_epi64(product, sum); + } } +} +XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2) + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void +XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { __m128i* const xacc = (__m128i*) acc; + /* Unaligned. This is mainly for pointer arithmetic, and because + * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */ + const __m128i* const xsecret = (const __m128i *) secret; + const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1); + + size_t i; + for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m128i const acc_vec = xacc[i]; + __m128i const shifted = _mm_srli_epi64 (acc_vec, 47); + __m128i const data_vec = _mm_xor_si128 (acc_vec, shifted); + /* xacc[i] ^= xsecret[i]; */ + __m128i const key_vec = _mm_loadu_si128 (xsecret+i); + __m128i const data_key = _mm_xor_si128 (data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1)); + __m128i const prod_lo = _mm_mul_epu32 (data_key, prime32); + __m128i const prod_hi = _mm_mul_epu32 (data_key_hi, prime32); + xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32)); + } + } +} + +XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0); + (void)(&XXH_writeLE64); + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i); + +# if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900 + /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */ + XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) }; + __m128i const seed = _mm_load_si128((__m128i const*)seed64x2); +# else + __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64); +# endif + int i; + + const void* const src16 = XXH3_kSecret; + __m128i* dst16 = (__m128i*) customSecret; +# if defined(__GNUC__) || defined(__clang__) + /* + * On GCC & Clang, marking 'dest' as modified will cause the compiler: + * - do not extract the secret from sse registers in the internal loop + * - use less common registers, and avoid pushing these reg into stack + */ + XXH_COMPILER_GUARD(dst16); +# endif + XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */ + XXH_ASSERT(((size_t)dst16 & 15) == 0); + + for (i=0; i < nbRounds; ++i) { + dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_NEON) + +/* forward declarations for the scalar routines */ +XXH_FORCE_INLINE void +XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input, + void const* XXH_RESTRICT secret, size_t lane); + +XXH_FORCE_INLINE void +XXH3_scalarScrambleRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT secret, size_t lane); + +/*! + * @internal + * @brief The bulk processing loop for NEON and WASM SIMD128. + * + * The NEON code path is actually partially scalar when running on AArch64. This + * is to optimize the pipelining and can have up to 15% speedup depending on the + * CPU, and it also mitigates some GCC codegen issues. + * + * @see XXH3_NEON_LANES for configuring this and details about this optimization. + * + * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit + * integers instead of the other platforms which mask full 64-bit vectors, + * so the setup is more complicated than just shifting right. + * + * Additionally, there is an optimization for 4 lanes at once noted below. + * + * Since, as stated, the most optimal amount of lanes for Cortexes is 6, + * there needs to be *three* versions of the accumulate operation used + * for the remaining 2 lanes. + * + * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap + * nearly perfectly. + */ + +XXH_FORCE_INLINE void +XXH3_accumulate_512_neon( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0); + { /* GCC for darwin arm64 does not like aliasing here */ + xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc; + /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */ + uint8_t const* xinput = (const uint8_t *) input; + uint8_t const* xsecret = (const uint8_t *) secret; + + size_t i; +#ifdef __wasm_simd128__ + /* + * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret + * is constant propagated, which results in it converting it to this + * inside the loop: + * + * a = v128.load(XXH3_kSecret + 0 + $secret_offset, offset = 0) + * b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0) + * ... + * + * This requires a full 32-bit address immediate (and therefore a 6 byte + * instruction) as well as an add for each offset. + * + * Putting an asm guard prevents it from folding (at the cost of losing + * the alignment hint), and uses the free offset in `v128.load` instead + * of adding secret_offset each time which overall reduces code size by + * about a kilobyte and improves performance. + */ + XXH_COMPILER_GUARD(xsecret); +#endif + /* Scalar lanes use the normal scalarRound routine */ + for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) { + XXH3_scalarRound(acc, input, secret, i); + } + i = 0; + /* 4 NEON lanes at a time. */ + for (; i+1 < XXH3_NEON_LANES / 2; i+=2) { + /* data_vec = xinput[i]; */ + uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput + (i * 16)); + uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput + ((i+1) * 16)); + /* key_vec = xsecret[i]; */ + uint64x2_t key_vec_1 = XXH_vld1q_u64(xsecret + (i * 16)); + uint64x2_t key_vec_2 = XXH_vld1q_u64(xsecret + ((i+1) * 16)); + /* data_swap = swap(data_vec) */ + uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1); + uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1); + /* data_key = data_vec ^ key_vec; */ + uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1); + uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2); + + /* + * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a + * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to + * get one vector with the low 32 bits of each lane, and one vector + * with the high 32 bits of each lane. + * + * The intrinsic returns a double vector because the original ARMv7-a + * instruction modified both arguments in place. AArch64 and SIMD128 emit + * two instructions from this intrinsic. + * + * [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ] + * [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ] + */ + uint32x4x2_t unzipped = vuzpq_u32( + vreinterpretq_u32_u64(data_key_1), + vreinterpretq_u32_u64(data_key_2) + ); + /* data_key_lo = data_key & 0xFFFFFFFF */ + uint32x4_t data_key_lo = unzipped.val[0]; + /* data_key_hi = data_key >> 32 */ + uint32x4_t data_key_hi = unzipped.val[1]; + /* + * Then, we can split the vectors horizontally and multiply which, as for most + * widening intrinsics, have a variant that works on both high half vectors + * for free on AArch64. A similar instruction is available on SIMD128. + * + * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi + */ + uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi); + uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi); + /* + * Clang reorders + * a += b * c; // umlal swap.2d, dkl.2s, dkh.2s + * c += a; // add acc.2d, acc.2d, swap.2d + * to + * c += a; // add acc.2d, acc.2d, swap.2d + * c += b * c; // umlal acc.2d, dkl.2s, dkh.2s + * + * While it would make sense in theory since the addition is faster, + * for reasons likely related to umlal being limited to certain NEON + * pipelines, this is worse. A compiler guard fixes this. + */ + XXH_COMPILER_GUARD_CLANG_NEON(sum_1); + XXH_COMPILER_GUARD_CLANG_NEON(sum_2); + /* xacc[i] = acc_vec + sum; */ + xacc[i] = vaddq_u64(xacc[i], sum_1); + xacc[i+1] = vaddq_u64(xacc[i+1], sum_2); + } + /* Operate on the remaining NEON lanes 2 at a time. */ + for (; i < XXH3_NEON_LANES / 2; i++) { + /* data_vec = xinput[i]; */ + uint64x2_t data_vec = XXH_vld1q_u64(xinput + (i * 16)); + /* key_vec = xsecret[i]; */ + uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16)); + /* acc_vec_2 = swap(data_vec) */ + uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1); + /* data_key = data_vec ^ key_vec; */ + uint64x2_t data_key = veorq_u64(data_vec, key_vec); + /* For two lanes, just use VMOVN and VSHRN. */ + /* data_key_lo = data_key & 0xFFFFFFFF; */ + uint32x2_t data_key_lo = vmovn_u64(data_key); + /* data_key_hi = data_key >> 32; */ + uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32); + /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */ + uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi); + /* Same Clang workaround as before */ + XXH_COMPILER_GUARD_CLANG_NEON(sum); + /* xacc[i] = acc_vec + sum; */ + xacc[i] = vaddq_u64 (xacc[i], sum); + } + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + + { xxh_aliasing_uint64x2_t* xacc = (xxh_aliasing_uint64x2_t*) acc; + uint8_t const* xsecret = (uint8_t const*) secret; + + size_t i; + /* WASM uses operator overloads and doesn't need these. */ +#ifndef __wasm_simd128__ + /* { prime32_1, prime32_1 } */ + uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1); + /* { 0, prime32_1, 0, prime32_1 } */ + uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32)); +#endif + + /* AArch64 uses both scalar and neon at the same time */ + for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) { + XXH3_scalarScrambleRound(acc, secret, i); + } + for (i=0; i < XXH3_NEON_LANES / 2; i++) { + /* xacc[i] ^= (xacc[i] >> 47); */ + uint64x2_t acc_vec = xacc[i]; + uint64x2_t shifted = vshrq_n_u64(acc_vec, 47); + uint64x2_t data_vec = veorq_u64(acc_vec, shifted); + + /* xacc[i] ^= xsecret[i]; */ + uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16)); + uint64x2_t data_key = veorq_u64(data_vec, key_vec); + /* xacc[i] *= XXH_PRIME32_1 */ +#ifdef __wasm_simd128__ + /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */ + xacc[i] = data_key * XXH_PRIME32_1; +#else + /* + * Expanded version with portable NEON intrinsics + * + * lo(x) * lo(y) + (hi(x) * lo(y) << 32) + * + * prod_hi = hi(data_key) * lo(prime) << 32 + * + * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector + * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits + * and avoid the shift. + */ + uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi); + /* Extract low bits for vmlal_u32 */ + uint32x2_t data_key_lo = vmovn_u64(data_key); + /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */ + xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo); +#endif + } + } +} +#endif + +#if (XXH_VECTOR == XXH_VSX) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_vsx( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + /* presumed aligned */ + xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc; + xxh_u8 const* const xinput = (xxh_u8 const*) input; /* no alignment restriction */ + xxh_u8 const* const xsecret = (xxh_u8 const*) secret; /* no alignment restriction */ + xxh_u64x2 const v32 = { 32, 32 }; + size_t i; + for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) { + /* data_vec = xinput[i]; */ + xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i); + /* key_vec = xsecret[i]; */ + xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i); + xxh_u64x2 const data_key = data_vec ^ key_vec; + /* shuffled = (data_key << 32) | (data_key >> 32); */ + xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32); + /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */ + xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled); + /* acc_vec = xacc[i]; */ + xxh_u64x2 acc_vec = xacc[i]; + acc_vec += product; + + /* swap high and low halves */ +#ifdef __s390x__ + acc_vec += vec_permi(data_vec, data_vec, 2); +#else + acc_vec += vec_xxpermdi(data_vec, data_vec, 2); +#endif + xacc[i] = acc_vec; + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + + { xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc; + const xxh_u8* const xsecret = (const xxh_u8*) secret; + /* constants */ + xxh_u64x2 const v32 = { 32, 32 }; + xxh_u64x2 const v47 = { 47, 47 }; + xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 }; + size_t i; + for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) { + /* xacc[i] ^= (xacc[i] >> 47); */ + xxh_u64x2 const acc_vec = xacc[i]; + xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47); + + /* xacc[i] ^= xsecret[i]; */ + xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16*i); + xxh_u64x2 const data_key = data_vec ^ key_vec; + + /* xacc[i] *= XXH_PRIME32_1 */ + /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF); */ + xxh_u64x2 const prod_even = XXH_vec_mule((xxh_u32x4)data_key, prime); + /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32); */ + xxh_u64x2 const prod_odd = XXH_vec_mulo((xxh_u32x4)data_key, prime); + xacc[i] = prod_odd + (prod_even << v32); + } } +} + +#endif + +#if (XXH_VECTOR == XXH_SVE) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_sve( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + uint64_t *xacc = (uint64_t *)acc; + const uint64_t *xinput = (const uint64_t *)(const void *)input; + const uint64_t *xsecret = (const uint64_t *)(const void *)secret; + svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1); + uint64_t element_count = svcntd(); + if (element_count >= 8) { + svbool_t mask = svptrue_pat_b64(SV_VL8); + svuint64_t vacc = svld1_u64(mask, xacc); + ACCRND(vacc, 0); + svst1_u64(mask, xacc, vacc); + } else if (element_count == 2) { /* sve128 */ + svbool_t mask = svptrue_pat_b64(SV_VL2); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 2); + svuint64_t acc2 = svld1_u64(mask, xacc + 4); + svuint64_t acc3 = svld1_u64(mask, xacc + 6); + ACCRND(acc0, 0); + ACCRND(acc1, 2); + ACCRND(acc2, 4); + ACCRND(acc3, 6); + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 2, acc1); + svst1_u64(mask, xacc + 4, acc2); + svst1_u64(mask, xacc + 6, acc3); + } else { + svbool_t mask = svptrue_pat_b64(SV_VL4); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 4); + ACCRND(acc0, 0); + ACCRND(acc1, 4); + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 4, acc1); + } +} + +XXH_FORCE_INLINE void +XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc, + const xxh_u8* XXH_RESTRICT input, + const xxh_u8* XXH_RESTRICT secret, + size_t nbStripes) +{ + if (nbStripes != 0) { + uint64_t *xacc = (uint64_t *)acc; + const uint64_t *xinput = (const uint64_t *)(const void *)input; + const uint64_t *xsecret = (const uint64_t *)(const void *)secret; + svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1); + uint64_t element_count = svcntd(); + if (element_count >= 8) { + svbool_t mask = svptrue_pat_b64(SV_VL8); + svuint64_t vacc = svld1_u64(mask, xacc + 0); + do { + /* svprfd(svbool_t, void *, enum svfprop); */ + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(vacc, 0); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, vacc); + } else if (element_count == 2) { /* sve128 */ + svbool_t mask = svptrue_pat_b64(SV_VL2); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 2); + svuint64_t acc2 = svld1_u64(mask, xacc + 4); + svuint64_t acc3 = svld1_u64(mask, xacc + 6); + do { + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(acc0, 0); + ACCRND(acc1, 2); + ACCRND(acc2, 4); + ACCRND(acc3, 6); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 2, acc1); + svst1_u64(mask, xacc + 4, acc2); + svst1_u64(mask, xacc + 6, acc3); + } else { + svbool_t mask = svptrue_pat_b64(SV_VL4); + svuint64_t acc0 = svld1_u64(mask, xacc + 0); + svuint64_t acc1 = svld1_u64(mask, xacc + 4); + do { + svprfd(mask, xinput + 128, SV_PLDL1STRM); + ACCRND(acc0, 0); + ACCRND(acc1, 4); + xinput += 8; + xsecret += 1; + nbStripes--; + } while (nbStripes != 0); + + svst1_u64(mask, xacc + 0, acc0); + svst1_u64(mask, xacc + 4, acc1); + } + } +} + +#endif + +#if (XXH_VECTOR == XXH_LSX) +#define _LSX_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_lsx( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { + __m128i* const xacc = (__m128i *) acc; + const __m128i* const xinput = (const __m128i *) input; + const __m128i* const xsecret = (const __m128i *) secret; + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) { + /* data_vec = xinput[i]; */ + __m128i const data_vec = __lsx_vld(xinput + i, 0); + /* key_vec = xsecret[i]; */ + __m128i const key_vec = __lsx_vld(xsecret + i, 0); + /* data_key = data_vec ^ key_vec; */ + __m128i const data_key = __lsx_vxor_v(data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m128i const data_key_lo = __lsx_vsrli_d(data_key, 32); + // __m128i const data_key_lo = __lsx_vsrli_d(data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m128i const product = __lsx_vmulwev_d_wu(data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m128i const data_swap = __lsx_vshuf4i_w(data_vec, _LSX_SHUFFLE(1, 0, 3, 2)); + __m128i const sum = __lsx_vadd_d(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = __lsx_vadd_d(product, sum); + } + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(lsx) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_lsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 15) == 0); + { + __m128i* const xacc = (__m128i*) acc; + const __m128i* const xsecret = (const __m128i *) secret; + const __m128i prime32 = __lsx_vreplgr2vr_d(XXH_PRIME32_1); + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m128i const acc_vec = xacc[i]; + __m128i const shifted = __lsx_vsrli_d(acc_vec, 47); + __m128i const data_vec = __lsx_vxor_v(acc_vec, shifted); + /* xacc[i] ^= xsecret[i]; */ + __m128i const key_vec = __lsx_vld(xsecret + i, 0); + __m128i const data_key = __lsx_vxor_v(data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + xacc[i] = __lsx_vmul_d(data_key, prime32); + } + } +} + +#endif + +#if (XXH_VECTOR == XXH_LASX) +#define _LASX_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) + +XXH_FORCE_INLINE void +XXH3_accumulate_512_lasx( void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { + __m256i* const xacc = (__m256i *) acc; + const __m256i* const xinput = (const __m256i *) input; + const __m256i* const xsecret = (const __m256i *) secret; + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) { + /* data_vec = xinput[i]; */ + __m256i const data_vec = __lasx_xvld(xinput + i, 0); + /* key_vec = xsecret[i]; */ + __m256i const key_vec = __lasx_xvld(xsecret + i, 0); + /* data_key = data_vec ^ key_vec; */ + __m256i const data_key = __lasx_xvxor_v(data_vec, key_vec); + /* data_key_lo = data_key >> 32; */ + __m256i const data_key_lo = __lasx_xvsrli_d(data_key, 32); + // __m256i const data_key_lo = __lasx_xvsrli_d(data_key, 32); + /* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */ + __m256i const product = __lasx_xvmulwev_d_wu(data_key, data_key_lo); + /* xacc[i] += swap(data_vec); */ + __m256i const data_swap = __lasx_xvshuf4i_w(data_vec, _LASX_SHUFFLE(1, 0, 3, 2)); + __m256i const sum = __lasx_xvadd_d(xacc[i], data_swap); + /* xacc[i] += product; */ + xacc[i] = __lasx_xvadd_d(product, sum); + } + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(lasx) + +XXH_FORCE_INLINE void +XXH3_scrambleAcc_lasx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + XXH_ASSERT((((size_t)acc) & 31) == 0); + { + __m256i* const xacc = (__m256i*) acc; + const __m256i* const xsecret = (const __m256i *) secret; + const __m256i prime32 = __lasx_xvreplgr2vr_d(XXH_PRIME32_1); + + for (size_t i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) { + /* xacc[i] ^= (xacc[i] >> 47) */ + __m256i const acc_vec = xacc[i]; + __m256i const shifted = __lasx_xvsrli_d(acc_vec, 47); + __m256i const data_vec = __lasx_xvxor_v(acc_vec, shifted); + /* xacc[i] ^= xsecret[i]; */ + __m256i const key_vec = __lasx_xvld(xsecret + i, 0); + __m256i const data_key = __lasx_xvxor_v(data_vec, key_vec); + + /* xacc[i] *= XXH_PRIME32_1; */ + xacc[i] = __lasx_xvmul_d(data_key, prime32); + } + } +} + +#endif + +/* scalar variants - universal */ + +#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__)) +/* + * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they + * emit an excess mask and a full 64-bit multiply-add (MADD X-form). + * + * While this might not seem like much, as AArch64 is a 64-bit architecture, only + * big Cortex designs have a full 64-bit multiplier. + * + * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit + * multiplies expand to 2-3 multiplies in microcode. This has a major penalty + * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline. + * + * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does + * not have this penalty and does the mask automatically. + */ +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc) +{ + xxh_u64 ret; + /* note: %x = 64-bit register, %w = 32-bit register */ + __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc)); + return ret; +} +#else +XXH_FORCE_INLINE xxh_u64 +XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc) +{ + return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc; +} +#endif + +/*! + * @internal + * @brief Scalar round for @ref XXH3_accumulate_512_scalar(). + * + * This is extracted to its own function because the NEON path uses a combination + * of NEON and scalar. + */ +XXH_FORCE_INLINE void +XXH3_scalarRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT input, + void const* XXH_RESTRICT secret, + size_t lane) +{ + xxh_u64* xacc = (xxh_u64*) acc; + xxh_u8 const* xinput = (xxh_u8 const*) input; + xxh_u8 const* xsecret = (xxh_u8 const*) secret; + XXH_ASSERT(lane < XXH_ACC_NB); + XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0); + { + xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8); + xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8); + xacc[lane ^ 1] += data_val; /* swap adjacent lanes */ + xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]); + } +} + +/*! + * @internal + * @brief Processes a 64 byte block of data using the scalar path. + */ +XXH_FORCE_INLINE void +XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc, + const void* XXH_RESTRICT input, + const void* XXH_RESTRICT secret) +{ + size_t i; + /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */ +#if defined(__GNUC__) && !defined(__clang__) \ + && (defined(__arm__) || defined(__thumb2__)) \ + && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \ + && XXH_SIZE_OPT <= 0 +# pragma GCC unroll 8 +#endif + for (i=0; i < XXH_ACC_NB; i++) { + XXH3_scalarRound(acc, input, secret, i); + } +} +XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar) + +/*! + * @internal + * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar(). + * + * This is extracted to its own function because the NEON path uses a combination + * of NEON and scalar. + */ +XXH_FORCE_INLINE void +XXH3_scalarScrambleRound(void* XXH_RESTRICT acc, + void const* XXH_RESTRICT secret, + size_t lane) +{ + xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */ + const xxh_u8* const xsecret = (const xxh_u8*) secret; /* no alignment restriction */ + XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0); + XXH_ASSERT(lane < XXH_ACC_NB); + { + xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8); + xxh_u64 acc64 = xacc[lane]; + acc64 = XXH_xorshift64(acc64, 47); + acc64 ^= key64; + acc64 *= XXH_PRIME32_1; + xacc[lane] = acc64; + } +} + +/*! + * @internal + * @brief Scrambles the accumulators after a large chunk has been read + */ +XXH_FORCE_INLINE void +XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret) +{ + size_t i; + for (i=0; i < XXH_ACC_NB; i++) { + XXH3_scalarScrambleRound(acc, secret, i); + } +} + +XXH_FORCE_INLINE void +XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64) +{ + /* + * We need a separate pointer for the hack below, + * which requires a non-const pointer. + * Any decent compiler will optimize this out otherwise. + */ + const xxh_u8* kSecretPtr = XXH3_kSecret; + XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0); + +#if defined(__GNUC__) && defined(__aarch64__) + /* + * UGLY HACK: + * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are + * placed sequentially, in order, at the top of the unrolled loop. + * + * While MOVK is great for generating constants (2 cycles for a 64-bit + * constant compared to 4 cycles for LDR), it fights for bandwidth with + * the arithmetic instructions. + * + * I L S + * MOVK + * MOVK + * MOVK + * MOVK + * ADD + * SUB STR + * STR + * By forcing loads from memory (as the asm line causes the compiler to assume + * that XXH3_kSecretPtr has been changed), the pipelines are used more + * efficiently: + * I L S + * LDR + * ADD LDR + * SUB STR + * STR + * + * See XXH3_NEON_LANES for details on the pipsline. + * + * XXH3_64bits_withSeed, len == 256, Snapdragon 835 + * without hack: 2654.4 MB/s + * with hack: 3202.9 MB/s + */ + XXH_COMPILER_GUARD(kSecretPtr); +#endif + { int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16; + int i; + for (i=0; i < nbRounds; i++) { + /* + * The asm hack causes the compiler to assume that kSecretPtr aliases with + * customSecret, and on aarch64, this prevented LDP from merging two + * loads together for free. Putting the loads together before the stores + * properly generates LDP. + */ + xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i) + seed64; + xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64; + XXH_writeLE64((xxh_u8*)customSecret + 16*i, lo); + XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi); + } } +} + + +typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t); +typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*); +typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64); + + +#if (XXH_VECTOR == XXH_AVX512) + +#define XXH3_accumulate_512 XXH3_accumulate_512_avx512 +#define XXH3_accumulate XXH3_accumulate_avx512 +#define XXH3_scrambleAcc XXH3_scrambleAcc_avx512 +#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512 + +#elif (XXH_VECTOR == XXH_AVX2) + +#define XXH3_accumulate_512 XXH3_accumulate_512_avx2 +#define XXH3_accumulate XXH3_accumulate_avx2 +#define XXH3_scrambleAcc XXH3_scrambleAcc_avx2 +#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2 + +#elif (XXH_VECTOR == XXH_SSE2) + +#define XXH3_accumulate_512 XXH3_accumulate_512_sse2 +#define XXH3_accumulate XXH3_accumulate_sse2 +#define XXH3_scrambleAcc XXH3_scrambleAcc_sse2 +#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2 + +#elif (XXH_VECTOR == XXH_NEON) + +#define XXH3_accumulate_512 XXH3_accumulate_512_neon +#define XXH3_accumulate XXH3_accumulate_neon +#define XXH3_scrambleAcc XXH3_scrambleAcc_neon +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_VSX) + +#define XXH3_accumulate_512 XXH3_accumulate_512_vsx +#define XXH3_accumulate XXH3_accumulate_vsx +#define XXH3_scrambleAcc XXH3_scrambleAcc_vsx +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_SVE) +#define XXH3_accumulate_512 XXH3_accumulate_512_sve +#define XXH3_accumulate XXH3_accumulate_sve +#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_LASX) +#define XXH3_accumulate_512 XXH3_accumulate_512_lasx +#define XXH3_accumulate XXH3_accumulate_lasx +#define XXH3_scrambleAcc XXH3_scrambleAcc_lasx +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#elif (XXH_VECTOR == XXH_LSX) +#define XXH3_accumulate_512 XXH3_accumulate_512_lsx +#define XXH3_accumulate XXH3_accumulate_lsx +#define XXH3_scrambleAcc XXH3_scrambleAcc_lsx +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#else /* scalar */ + +#define XXH3_accumulate_512 XXH3_accumulate_512_scalar +#define XXH3_accumulate XXH3_accumulate_scalar +#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar +#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar + +#endif + +#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */ +# undef XXH3_initCustomSecret +# define XXH3_initCustomSecret XXH3_initCustomSecret_scalar +#endif + +XXH_FORCE_INLINE void +XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc, + const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE; + size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock; + size_t const nb_blocks = (len - 1) / block_len; + + size_t n; + + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); + + for (n = 0; n < nb_blocks; n++) { + f_acc(acc, input + n*block_len, secret, nbStripesPerBlock); + f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN); + } + + /* last partial block */ + XXH_ASSERT(len > XXH_STRIPE_LEN); + { size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN; + XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE)); + f_acc(acc, input + nb_blocks*block_len, secret, nbStripes); + + /* last stripe */ + { const xxh_u8* const p = input + len - XXH_STRIPE_LEN; +#define XXH_SECRET_LASTACC_START 7 /* not aligned on 8, last secret is different from acc & scrambler */ + XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START); + } } +} + +XXH_FORCE_INLINE xxh_u64 +XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret) +{ + return XXH3_mul128_fold64( + acc[0] ^ XXH_readLE64(secret), + acc[1] ^ XXH_readLE64(secret+8) ); +} + +static XXH_PUREF XXH64_hash_t +XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start) +{ + xxh_u64 result64 = start; + size_t i = 0; + + for (i = 0; i < 4; i++) { + result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i); +#if defined(__clang__) /* Clang */ \ + && (defined(__arm__) || defined(__thumb__)) /* ARMv7 */ \ + && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \ + && !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */ + /* + * UGLY HACK: + * Prevent autovectorization on Clang ARMv7-a. Exact same problem as + * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b. + * XXH3_64bits, len == 256, Snapdragon 835: + * without hack: 2063.7 MB/s + * with hack: 2560.7 MB/s + */ + XXH_COMPILER_GUARD(result64); +#endif + } + + return XXH3_avalanche(result64); +} + +/* do not align on 8, so that the secret is different from the accumulator */ +#define XXH_SECRET_MERGEACCS_START 11 + +static XXH_PUREF XXH64_hash_t +XXH3_finalizeLong_64b(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 len) +{ + return XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START, len * XXH_PRIME64_1); +} + +#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \ + XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 } + +XXH_FORCE_INLINE XXH64_hash_t +XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len, + const void* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC; + + XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble); + + /* converge into final hash */ + XXH_STATIC_ASSERT(sizeof(acc) == 64); + XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_64b(acc, (const xxh_u8*)secret, (xxh_u64)len); +} + +/* + * It's important for performance to transmit secret's size (when it's static) + * so that the compiler can properly optimize the vectorized loop. + * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set. + * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE + * breaks -Og, this is XXH_NO_INLINE. + */ +XXH3_WITH_SECRET_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; + return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * It's preferable for performance that XXH3_hashLong is not inlined, + * as it results in a smaller function for small data, easier to the instruction cache. + * Note that inside this no_inline function, we do inline the internal loop, + * and provide a statically defined secret size to allow optimization of vector loop. + */ +XXH_NO_INLINE XXH_PUREF XXH64_hash_t +XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * XXH3_hashLong_64b_withSeed(): + * Generate a custom key based on alteration of default XXH3_kSecret with the seed, + * and then use this key for long mode hashing. + * + * This operation is decently fast but nonetheless costs a little bit of time. + * Try to avoid it whenever possible (typically when seed==0). + * + * It's important for performance that XXH3_hashLong is not inlined. Not sure + * why (uop cache maybe?), but the difference is large and easily measurable. + */ +XXH_FORCE_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len, + XXH64_hash_t seed, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble, + XXH3_f_initCustomSecret f_initSec) +{ +#if XXH_SIZE_OPT <= 0 + if (seed == 0) + return XXH3_hashLong_64b_internal(input, len, + XXH3_kSecret, sizeof(XXH3_kSecret), + f_acc, f_scramble); +#endif + { XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE]; + f_initSec(secret, seed); + return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret), + f_acc, f_scramble); + } +} + +/* + * It's important for performance that XXH3_hashLong is not inlined. + */ +XXH_NO_INLINE XXH64_hash_t +XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + return XXH3_hashLong_64b_withSeed_internal(input, len, seed, + XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret); +} + + +typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t, + XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t); + +XXH_FORCE_INLINE XXH64_hash_t +XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen, + XXH3_hashLong64_f f_hashLong) +{ + XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN); + /* + * If an action is to be taken if `secretLen` condition is not respected, + * it should be done here. + * For now, it's a contract pre-condition. + * Adding a check and a branch here would cost performance at every hash. + * Also, note that function signature doesn't offer room to return an error. + */ + if (len <= 16) + return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64); + if (len <= 128) + return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen); +} + + +/* === Public entry point === */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length) +{ + return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed) +{ + return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed); +} + +XXH_PUBLIC_API XXH64_hash_t +XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + if (length <= XXH3_MIDSIZE_MAX) + return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL); + return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize); +} + + +/* === XXH3 streaming === */ +#ifndef XXH_NO_STREAM +/* + * Malloc's a pointer that is always aligned to @align. + * + * This must be freed with `XXH_alignedFree()`. + * + * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte + * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2 + * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON. + * + * This underalignment previously caused a rather obvious crash which went + * completely unnoticed due to XXH3_createState() not actually being tested. + * Credit to RedSpah for noticing this bug. + * + * The alignment is done manually: Functions like posix_memalign or _mm_malloc + * are avoided: To maintain portability, we would have to write a fallback + * like this anyways, and besides, testing for the existence of library + * functions without relying on external build tools is impossible. + * + * The method is simple: Overallocate, manually align, and store the offset + * to the original behind the returned pointer. + * + * Align must be a power of 2 and 8 <= align <= 128. + */ +static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align) +{ + XXH_ASSERT(align <= 128 && align >= 8); /* range check */ + XXH_ASSERT((align & (align-1)) == 0); /* power of 2 */ + XXH_ASSERT(s != 0 && s < (s + align)); /* empty/overflow */ + { /* Overallocate to make room for manual realignment and an offset byte */ + xxh_u8* base = (xxh_u8*)XXH_malloc(s + align); + if (base != NULL) { + /* + * Get the offset needed to align this pointer. + * + * Even if the returned pointer is aligned, there will always be + * at least one byte to store the offset to the original pointer. + */ + size_t offset = align - ((size_t)base & (align - 1)); /* base % align */ + /* Add the offset for the now-aligned pointer */ + xxh_u8* ptr = base + offset; + + XXH_ASSERT((size_t)ptr % align == 0); + + /* Store the offset immediately before the returned pointer. */ + ptr[-1] = (xxh_u8)offset; + return ptr; + } + return NULL; + } +} +/* + * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass + * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout. + */ +static void XXH_alignedFree(void* p) +{ + if (p != NULL) { + xxh_u8* ptr = (xxh_u8*)p; + /* Get the offset byte we added in XXH_malloc. */ + xxh_u8 offset = ptr[-1]; + /* Free the original malloc'd pointer */ + xxh_u8* base = ptr - offset; + XXH_free(base); + } +} +/*! @ingroup XXH3_family */ +/*! + * @brief Allocate an @ref XXH3_state_t. + * + * @return An allocated pointer of @ref XXH3_state_t on success. + * @return `NULL` on failure. + * + * @note Must be freed with XXH3_freeState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void) +{ + XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64); + if (state==NULL) return NULL; + XXH3_INITSTATE(state); + return state; +} + +/*! @ingroup XXH3_family */ +/*! + * @brief Frees an @ref XXH3_state_t. + * + * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState(). + * + * @return @ref XXH_OK. + * + * @note Must be allocated with XXH3_createState(). + * + * @see @ref streaming_example "Streaming Example" + */ +XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr) +{ + XXH_alignedFree(statePtr); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API void +XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state) +{ + XXH_memcpy(dst_state, src_state, sizeof(*dst_state)); +} + +static void +XXH3_reset_internal(XXH3_state_t* statePtr, + XXH64_hash_t seed, + const void* secret, size_t secretSize) +{ + size_t const initStart = offsetof(XXH3_state_t, bufferedSize); + size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart; + XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart); + XXH_ASSERT(statePtr != NULL); + /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */ + XXH_memset((char*)statePtr + initStart, 0, initLength); + statePtr->acc[0] = XXH_PRIME32_3; + statePtr->acc[1] = XXH_PRIME64_1; + statePtr->acc[2] = XXH_PRIME64_2; + statePtr->acc[3] = XXH_PRIME64_3; + statePtr->acc[4] = XXH_PRIME64_4; + statePtr->acc[5] = XXH_PRIME32_2; + statePtr->acc[6] = XXH_PRIME64_5; + statePtr->acc[7] = XXH_PRIME32_1; + statePtr->seed = seed; + statePtr->useSeed = (seed != 0); + statePtr->extSecret = (const unsigned char*)secret; + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); + statePtr->secretLimit = secretSize - XXH_STRIPE_LEN; + statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr) +{ + if (statePtr == NULL) return XXH_ERROR; + XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + if (statePtr == NULL) return XXH_ERROR; + XXH3_reset_internal(statePtr, 0, secret, secretSize); + if (secret == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed) +{ + if (statePtr == NULL) return XXH_ERROR; + if (seed==0) return XXH3_64bits_reset(statePtr); + if ((seed != statePtr->seed) || (statePtr->extSecret != NULL)) + XXH3_initCustomSecret(statePtr->customSecret, seed); + XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE); + return XXH_OK; +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64) +{ + if (statePtr == NULL) return XXH_ERROR; + if (secret == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; + XXH3_reset_internal(statePtr, seed64, secret, secretSize); + statePtr->useSeed = 1; /* always, even if seed64==0 */ + return XXH_OK; +} + +/*! + * @internal + * @brief Processes a large input for XXH3_update() and XXH3_digest_long(). + * + * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block. + * + * @param acc Pointer to the 8 accumulator lanes + * @param nbStripesSoFarPtr In/out pointer to the number of leftover stripes in the block* + * @param nbStripesPerBlock Number of stripes in a block + * @param input Input pointer + * @param nbStripes Number of stripes to process + * @param secret Secret pointer + * @param secretLimit Offset of the last block in @p secret + * @param f_acc Pointer to an XXH3_accumulate implementation + * @param f_scramble Pointer to an XXH3_scrambleAcc implementation + * @return Pointer past the end of @p input after processing + */ +XXH_FORCE_INLINE const xxh_u8 * +XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc, + size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock, + const xxh_u8* XXH_RESTRICT input, size_t nbStripes, + const xxh_u8* XXH_RESTRICT secret, size_t secretLimit, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE; + /* Process full blocks */ + if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) { + /* Process the initial partial block... */ + size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr; + + do { + /* Accumulate and scramble */ + f_acc(acc, input, initialSecret, nbStripesThisIter); + f_scramble(acc, secret + secretLimit); + input += nbStripesThisIter * XXH_STRIPE_LEN; + nbStripes -= nbStripesThisIter; + /* Then continue the loop with the full block size */ + nbStripesThisIter = nbStripesPerBlock; + initialSecret = secret; + } while (nbStripes >= nbStripesPerBlock); + *nbStripesSoFarPtr = 0; + } + /* Process a partial block */ + if (nbStripes > 0) { + f_acc(acc, input, initialSecret, nbStripes); + input += nbStripes * XXH_STRIPE_LEN; + *nbStripesSoFarPtr += nbStripes; + } + /* Return end pointer */ + return input; +} + +#ifndef XXH3_STREAM_USE_STACK +# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */ +# define XXH3_STREAM_USE_STACK 1 +# endif +#endif +/* This function accepts f_acc and f_scramble as function pointers, + * making it possible to implement multiple variants with different acc & scramble stages. + * This is notably useful to implement multiple vector variants with different intrinsics. + */ +XXH_FORCE_INLINE XXH_errorcode +XXH3_update(XXH3_state_t* XXH_RESTRICT const state, + const xxh_u8* XXH_RESTRICT input, size_t len, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + if (input==NULL) { + XXH_ASSERT(len == 0); + return XXH_OK; + } + + XXH_ASSERT(state != NULL); + { const xxh_u8* const bEnd = input + len; + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; +#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1 + /* For some reason, gcc and MSVC seem to suffer greatly + * when operating accumulators directly into state. + * Operating into stack space seems to enable proper optimization. + * clang, on the other hand, doesn't seem to need this trick */ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; + XXH_memcpy(acc, state->acc, sizeof(acc)); +#else + xxh_u64* XXH_RESTRICT const acc = state->acc; +#endif + state->totalLen += len; + XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE); + + /* small input : just fill in tmp buffer */ + if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) { + XXH_memcpy(state->buffer + state->bufferedSize, input, len); + state->bufferedSize += (XXH32_hash_t)len; + return XXH_OK; + } + + /* total input is now > XXH3_INTERNALBUFFER_SIZE */ + #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN) + XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0); /* clean multiple */ + + /* + * Internal buffer is partially filled (always, except at beginning) + * Complete it, then consume it. + */ + if (state->bufferedSize) { + size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize; + XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize); + input += loadSize; + XXH3_consumeStripes(acc, + &state->nbStripesSoFar, state->nbStripesPerBlock, + state->buffer, XXH3_INTERNALBUFFER_STRIPES, + secret, state->secretLimit, + f_acc, f_scramble); + state->bufferedSize = 0; + } + XXH_ASSERT(input < bEnd); + if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) { + size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN; + input = XXH3_consumeStripes(acc, + &state->nbStripesSoFar, state->nbStripesPerBlock, + input, nbStripes, + secret, state->secretLimit, + f_acc, f_scramble); + XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN); + + } + /* Some remaining input (always) : buffer it */ + XXH_ASSERT(input < bEnd); + XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE); + XXH_ASSERT(state->bufferedSize == 0); + XXH_memcpy(state->buffer, input, (size_t)(bEnd-input)); + state->bufferedSize = (XXH32_hash_t)(bEnd-input); +#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1 + /* save stack accumulators into state */ + XXH_memcpy(state->acc, acc, sizeof(acc)); +#endif + } + + return XXH_OK; +} + +/* + * Both XXH3_64bits_update and XXH3_128bits_update use this routine. + */ +XXH_NO_INLINE XXH_errorcode +XXH3_update_regular(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_update(state, (const xxh_u8*)input, len, + XXH3_accumulate, XXH3_scrambleAcc); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_update_regular(state, input, len); +} + + +XXH_FORCE_INLINE void +XXH3_digest_long (XXH64_hash_t* acc, + const XXH3_state_t* state, + const unsigned char* secret) +{ + xxh_u8 lastStripe[XXH_STRIPE_LEN]; + const xxh_u8* lastStripePtr; + + /* + * Digest on a local copy. This way, the state remains unaltered, and it can + * continue ingesting more input afterwards. + */ + XXH_memcpy(acc, state->acc, sizeof(state->acc)); + if (state->bufferedSize >= XXH_STRIPE_LEN) { + /* Consume remaining stripes then point to remaining data in buffer */ + size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN; + size_t nbStripesSoFar = state->nbStripesSoFar; + XXH3_consumeStripes(acc, + &nbStripesSoFar, state->nbStripesPerBlock, + state->buffer, nbStripes, + secret, state->secretLimit, + XXH3_accumulate, XXH3_scrambleAcc); + lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN; + } else { /* bufferedSize < XXH_STRIPE_LEN */ + /* Copy to temp buffer */ + size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize; + XXH_ASSERT(state->bufferedSize > 0); /* there is always some input buffered */ + XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize); + XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize); + lastStripePtr = lastStripe; + } + /* Last stripe */ + XXH3_accumulate_512(acc, + lastStripePtr, + secret + state->secretLimit - XXH_SECRET_LASTACC_START); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state) +{ + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; + if (state->totalLen > XXH3_MIDSIZE_MAX) { + XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB]; + XXH3_digest_long(acc, state, secret); + return XXH3_finalizeLong_64b(acc, secret, (xxh_u64)state->totalLen); + } + /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */ + if (state->useSeed) + return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed); + return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen), + secret, state->secretLimit + XXH_STRIPE_LEN); +} +#endif /* !XXH_NO_STREAM */ + + +/* ========================================== + * XXH3 128 bits (a.k.a XXH128) + * ========================================== + * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant, + * even without counting the significantly larger output size. + * + * For example, extra steps are taken to avoid the seed-dependent collisions + * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B). + * + * This strength naturally comes at the cost of some speed, especially on short + * lengths. Note that longer hashes are about as fast as the 64-bit version + * due to it using only a slight modification of the 64-bit loop. + * + * XXH128 is also more oriented towards 64-bit machines. It is still extremely + * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64). + */ + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + /* A doubled version of 1to3_64b with different constants. */ + XXH_ASSERT(input != NULL); + XXH_ASSERT(1 <= len && len <= 3); + XXH_ASSERT(secret != NULL); + /* + * len = 1: combinedl = { input[0], 0x01, input[0], input[0] } + * len = 2: combinedl = { input[1], 0x02, input[0], input[1] } + * len = 3: combinedl = { input[2], 0x03, input[0], input[1] } + */ + { xxh_u8 const c1 = input[0]; + xxh_u8 const c2 = input[len >> 1]; + xxh_u8 const c3 = input[len - 1]; + xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24) + | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8); + xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13); + xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed; + xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed; + xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl; + xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph; + XXH128_hash_t h128; + h128.low64 = XXH64_avalanche(keyed_lo); + h128.high64 = XXH64_avalanche(keyed_hi); + return h128; + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(4 <= len && len <= 8); + seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32; + { xxh_u32 const input_lo = XXH_readLE32(input); + xxh_u32 const input_hi = XXH_readLE32(input + len - 4); + xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32); + xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed; + xxh_u64 const keyed = input_64 ^ bitflip; + + /* Shift len to the left to ensure it is even, this avoids even multiplies. */ + XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2)); + + m128.high64 += (m128.low64 << 1); + m128.low64 ^= (m128.high64 >> 3); + + m128.low64 = XXH_xorshift64(m128.low64, 35); + m128.low64 *= PRIME_MX2; + m128.low64 = XXH_xorshift64(m128.low64, 28); + m128.high64 = XXH3_avalanche(m128.high64); + return m128; + } +} + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(input != NULL); + XXH_ASSERT(secret != NULL); + XXH_ASSERT(9 <= len && len <= 16); + { xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed; + xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed; + xxh_u64 const input_lo = XXH_readLE64(input); + xxh_u64 input_hi = XXH_readLE64(input + len - 8); + XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1); + /* + * Put len in the middle of m128 to ensure that the length gets mixed to + * both the low and high bits in the 128x64 multiply below. + */ + m128.low64 += (xxh_u64)(len - 1) << 54; + input_hi ^= bitfliph; + /* + * Add the high 32 bits of input_hi to the high 32 bits of m128, then + * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to + * the high 64 bits of m128. + * + * The best approach to this operation is different on 32-bit and 64-bit. + */ + if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */ + /* + * 32-bit optimized version, which is more readable. + * + * On 32-bit, it removes an ADC and delays a dependency between the two + * halves of m128.high64, but it generates an extra mask on 64-bit. + */ + m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2); + } else { + /* + * 64-bit optimized (albeit more confusing) version. + * + * Uses some properties of addition and multiplication to remove the mask: + * + * Let: + * a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF) + * b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000) + * c = XXH_PRIME32_2 + * + * a + (b * c) + * Inverse Property: x + y - x == y + * a + (b * (1 + c - 1)) + * Distributive Property: x * (y + z) == (x * y) + (x * z) + * a + (b * 1) + (b * (c - 1)) + * Identity Property: x * 1 == x + * a + b + (b * (c - 1)) + * + * Substitute a, b, and c: + * input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1)) + * + * Since input_hi.hi + input_hi.lo == input_hi, we get this: + * input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1)) + */ + m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1); + } + /* m128 ^= XXH_swap64(m128 >> 64); */ + m128.low64 ^= XXH_swap64(m128.high64); + + { /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */ + XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2); + h128.high64 += m128.high64 * XXH_PRIME64_2; + + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = XXH3_avalanche(h128.high64); + return h128; + } } +} + +/* + * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN + */ +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed) +{ + XXH_ASSERT(len <= 16); + { if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed); + if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed); + if (len) return XXH3_len_1to3_128b(input, len, secret, seed); + { XXH128_hash_t h128; + xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72); + xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88); + h128.low64 = XXH64_avalanche(seed ^ bitflipl); + h128.high64 = XXH64_avalanche( seed ^ bitfliph); + return h128; + } } +} + +/* + * A bit slower than XXH3_mix16B, but handles multiply by zero better. + */ +XXH_FORCE_INLINE XXH128_hash_t +XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2, + const xxh_u8* secret, XXH64_hash_t seed) +{ + acc.low64 += XXH3_mix16B (input_1, secret+0, seed); + acc.low64 ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8); + acc.high64 += XXH3_mix16B (input_2, secret+16, seed); + acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8); + return acc; +} + + +XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(16 < len && len <= 128); + + { XXH128_hash_t acc; + acc.low64 = len * XXH_PRIME64_1; + acc.high64 = 0; + +#if XXH_SIZE_OPT >= 1 + { + /* Smaller, but slightly slower. */ + unsigned int i = (unsigned int)(len - 1) / 32; + do { + acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed); + } while (i-- != 0); + } +#else + if (len > 32) { + if (len > 64) { + if (len > 96) { + acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed); + } + acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed); + } + acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed); + } + acc = XXH128_mix32B(acc, input, input+len-16, secret, seed); +#endif + { XXH128_hash_t h128; + h128.low64 = acc.low64 + acc.high64; + h128.high64 = (acc.low64 * XXH_PRIME64_1) + + (acc.high64 * XXH_PRIME64_4) + + ((len - seed) * XXH_PRIME64_2); + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64); + return h128; + } + } +} + +XXH_NO_INLINE XXH_PUREF XXH128_hash_t +XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH64_hash_t seed) +{ + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize; + XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX); + + { XXH128_hash_t acc; + unsigned i; + acc.low64 = len * XXH_PRIME64_1; + acc.high64 = 0; + /* + * We set as `i` as offset + 32. We do this so that unchanged + * `len` can be used as upper bound. This reaches a sweet spot + * where both x86 and aarch64 get simple agen and good codegen + * for the loop. + */ + for (i = 32; i < 160; i += 32) { + acc = XXH128_mix32B(acc, + input + i - 32, + input + i - 16, + secret + i - 32, + seed); + } + acc.low64 = XXH3_avalanche(acc.low64); + acc.high64 = XXH3_avalanche(acc.high64); + /* + * NB: `i <= len` will duplicate the last 32-bytes if + * len % 32 was zero. This is an unfortunate necessity to keep + * the hash result stable. + */ + for (i=160; i <= len; i += 32) { + acc = XXH128_mix32B(acc, + input + i - 32, + input + i - 16, + secret + XXH3_MIDSIZE_STARTOFFSET + i - 160, + seed); + } + /* last bytes */ + acc = XXH128_mix32B(acc, + input + len - 16, + input + len - 32, + secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16, + (XXH64_hash_t)0 - seed); + + { XXH128_hash_t h128; + h128.low64 = acc.low64 + acc.high64; + h128.high64 = (acc.low64 * XXH_PRIME64_1) + + (acc.high64 * XXH_PRIME64_4) + + ((len - seed) * XXH_PRIME64_2); + h128.low64 = XXH3_avalanche(h128.low64); + h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64); + return h128; + } + } +} + +static XXH_PUREF XXH128_hash_t +XXH3_finalizeLong_128b(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, size_t secretSize, xxh_u64 len) +{ + XXH128_hash_t h128; + h128.low64 = XXH3_finalizeLong_64b(acc, secret, len); + h128.high64 = XXH3_mergeAccs(acc, secret + secretSize + - XXH_STRIPE_LEN - XXH_SECRET_MERGEACCS_START, + ~(len * XXH_PRIME64_2)); + return h128; +} + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len, + const xxh_u8* XXH_RESTRICT secret, size_t secretSize, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble) +{ + XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC; + + XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble); + + /* converge into final hash */ + XXH_STATIC_ASSERT(sizeof(acc) == 64); + XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_128b(acc, secret, secretSize, (xxh_u64)len); +} + +/* + * It's important for performance that XXH3_hashLong() is not inlined. + */ +XXH_NO_INLINE XXH_PUREF XXH128_hash_t +XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; (void)secret; (void)secretLen; + return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_accumulate, XXH3_scrambleAcc); +} + +/* + * It's important for performance to pass @p secretLen (when it's static) + * to the compiler, so that it can properly optimize the vectorized loop. + * + * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE + * breaks -Og, this is XXH_NO_INLINE. + */ +XXH3_WITH_SECRET_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)seed64; + return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen, + XXH3_accumulate, XXH3_scrambleAcc); +} + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len, + XXH64_hash_t seed64, + XXH3_f_accumulate f_acc, + XXH3_f_scrambleAcc f_scramble, + XXH3_f_initCustomSecret f_initSec) +{ + if (seed64 == 0) + return XXH3_hashLong_128b_internal(input, len, + XXH3_kSecret, sizeof(XXH3_kSecret), + f_acc, f_scramble); + { XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE]; + f_initSec(secret, seed64); + return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret), + f_acc, f_scramble); + } +} + +/* + * It's important for performance that XXH3_hashLong is not inlined. + */ +XXH_NO_INLINE XXH128_hash_t +XXH3_hashLong_128b_withSeed(const void* input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen) +{ + (void)secret; (void)secretLen; + return XXH3_hashLong_128b_withSeed_internal(input, len, seed64, + XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret); +} + +typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t, + XXH64_hash_t, const void* XXH_RESTRICT, size_t); + +XXH_FORCE_INLINE XXH128_hash_t +XXH3_128bits_internal(const void* input, size_t len, + XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen, + XXH3_hashLong128_f f_hl128) +{ + XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN); + /* + * If an action is to be taken if `secret` conditions are not respected, + * it should be done here. + * For now, it's a contract pre-condition. + * Adding a check and a branch here would cost performance at every hash. + */ + if (len <= 16) + return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64); + if (len <= 128) + return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64); + return f_hl128(input, len, seed64, secret, secretLen); +} + + +/* === Public XXH128 API === */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_128bits_internal(input, len, 0, + XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_hashLong_128b_default); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_128bits_internal(input, len, 0, + (const xxh_u8*)secret, secretSize, + XXH3_hashLong_128b_withSecret); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_128bits_internal(input, len, seed, + XXH3_kSecret, sizeof(XXH3_kSecret), + XXH3_hashLong_128b_withSeed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + if (len <= XXH3_MIDSIZE_MAX) + return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL); + return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed) +{ + return XXH3_128bits_withSeed(input, len, seed); +} + + +/* === XXH3 128-bit streaming === */ +#ifndef XXH_NO_STREAM +/* + * All initialization and update functions are identical to 64-bit streaming variant. + * The only difference is the finalization routine. + */ + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr) +{ + return XXH3_64bits_reset(statePtr); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize) +{ + return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed) +{ + return XXH3_64bits_reset_withSeed(statePtr, seed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed) +{ + return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len) +{ + return XXH3_update_regular(state, input, len); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state) +{ + const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret; + if (state->totalLen > XXH3_MIDSIZE_MAX) { + XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB]; + XXH3_digest_long(acc, state, secret); + XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START); + return XXH3_finalizeLong_128b(acc, secret, state->secretLimit + XXH_STRIPE_LEN, (xxh_u64)state->totalLen); + } + /* len <= XXH3_MIDSIZE_MAX : short code */ + if (state->useSeed) + return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed); + return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen), + secret, state->secretLimit + XXH_STRIPE_LEN); +} +#endif /* !XXH_NO_STREAM */ +/* 128-bit utility functions */ + +/* return : 1 is equal, 0 if different */ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2) +{ + /* note : XXH128_hash_t is compact, it has no padding byte */ + return !(XXH_memcmp(&h1, &h2, sizeof(h1))); +} + +/* This prototype is compatible with stdlib's qsort(). + * @return : >0 if *h128_1 > *h128_2 + * <0 if *h128_1 < *h128_2 + * =0 if *h128_1 == *h128_2 */ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2) +{ + XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1; + XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2; + int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64); + /* note : bets that, in most cases, hash values are different */ + if (hcmp) return hcmp; + return (h1.low64 > h2.low64) - (h2.low64 > h1.low64); +} + + +/*====== Canonical representation ======*/ +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API void +XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) { + hash.high64 = XXH_swap64(hash.high64); + hash.low64 = XXH_swap64(hash.low64); + } + XXH_memcpy(dst, &hash.high64, sizeof(hash.high64)); + XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64)); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH128_hash_t +XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src) +{ + XXH128_hash_t h; + h.high64 = XXH_readBE64(src); + h.low64 = XXH_readBE64(src->digest + 8); + return h; +} + + + +/* ========================================== + * Secret generators + * ========================================== + */ +#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x)) + +XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128) +{ + XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 ); + XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 ); +} + +/*! @ingroup XXH3_family */ +XXH_PUBLIC_API XXH_errorcode +XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize) +{ +#if (XXH_DEBUGLEVEL >= 1) + XXH_ASSERT(secretBuffer != NULL); + XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); +#else + /* production mode, assert() are disabled */ + if (secretBuffer == NULL) return XXH_ERROR; + if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR; +#endif + + if (customSeedSize == 0) { + customSeed = XXH3_kSecret; + customSeedSize = XXH_SECRET_DEFAULT_SIZE; + } +#if (XXH_DEBUGLEVEL >= 1) + XXH_ASSERT(customSeed != NULL); +#else + if (customSeed == NULL) return XXH_ERROR; +#endif + + /* Fill secretBuffer with a copy of customSeed - repeat as needed */ + { size_t pos = 0; + while (pos < secretSize) { + size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize); + XXH_memcpy((char*)secretBuffer + pos, customSeed, toCopy); + pos += toCopy; + } } + + { size_t const nbSeg16 = secretSize / 16; + size_t n; + XXH128_canonical_t scrambler; + XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0)); + for (n=0; n +# include +# include +# include +# include +#endif + +#if __linux__ +# define VK_USE_PLATFORM_XCB_KHR +#elif _WIN32 +# define VK_USE_PLATFORM_WIN32_KHR +#endif + +#define VK_NO_PROTOTYPES + +#include + +#define VMA_STATIC_VULKAN_FUNCTIONS 0 +#define VMA_DYNAMIC_VULKAN_FUNCTIONS 1 +#define VMA_VULKAN_VERSION 1003000 + +#include "../external/vma/vk_mem_alloc.h" + +#include "../external/xxhash/xxhash.h" diff --git a/src/main.d b/src/main.d new file mode 100644 index 0000000..1eb6169 --- /dev/null +++ b/src/main.d @@ -0,0 +1,18 @@ +public import includes; +import std.stdio; +import aliases; +import core.memory; +import u = util : Result; +import p = platform; +import a = alloc; +import vk = vulkan : Vulkan; + +void main() +{ + p.Window window = p.CreateWindow("Video Game", 1920, 1080); + + Result!(Vulkan) result = vk.Init(&window, u.MB(24), u.MB(32)); + Vulkan vulkan = result.value; + + vk.Destroy(&vulkan); +} diff --git a/src/main.o b/src/main.o new file mode 100644 index 0000000..887e9df Binary files /dev/null and b/src/main.o differ diff --git a/src/math.d b/src/math.d new file mode 100644 index 0000000..6f397a7 --- /dev/null +++ b/src/math.d @@ -0,0 +1,6 @@ +import aliases; + +T AlignPow2(T)(T v, T a) +{ + return (v + a - 1) & ~(a - 1); +}; diff --git a/src/platform.d b/src/platform.d new file mode 100644 index 0000000..8ae3f0d --- /dev/null +++ b/src/platform.d @@ -0,0 +1,174 @@ +import aliases; +import includes; +import std.stdio; + +version(linux) +{ +import core.sys.posix.dlfcn; + +struct Window +{ + Display *display; + xcb_connection_t *conn; + xcb_window_t window; + xcb_atom_t close_event; + xcb_atom_t minimize_event; + u16 w; + u16 h; +}; + +struct Library +{ + void* ptr; +}; + +struct Function +{ + void* ptr; +}; + +void CheckErr(Window *window, xcb_void_cookie_t *cookie, xcb_generic_error_t *err, string msg) +{ + assert(err == null, msg); + free(err); +}; + +Window CreateWindow(string name, u16 width, u16 height) +{ + Window window = { + w: width, + h: height, + }; + + assert(width > 0 && height > 0, "CreateWindow error: width and height must be above 0"); + + window.display = XOpenDisplay(null); + assert(window.display != null, "XOpenDisplay failure"); + + window.conn = XGetXCBConnection(window.display); + assert(window.conn != null, "XGetXCBConnection failure"); + + xcb_void_cookie_t cookie; + xcb_generic_error_t *error; + + xcb_setup_t *setup = xcb_get_setup(window.conn); + xcb_screen_iterator_t iter = xcb_setup_roots_iterator(setup); + xcb_screen_t *screen = iter.data; + + i32 event_mask = XCB_EVENT_MASK_EXPOSURE | + XCB_EVENT_MASK_KEY_PRESS | + XCB_EVENT_MASK_KEY_RELEASE | + XCB_EVENT_MASK_BUTTON_PRESS | + XCB_EVENT_MASK_BUTTON_RELEASE | + XCB_EVENT_MASK_POINTER_MOTION | + XCB_EVENT_MASK_STRUCTURE_NOTIFY; + + i32[] val_win = [screen.black_pixel, event_mask]; + i32 val_mask = XCB_CW_BACK_PIXEL | XCB_CW_EVENT_MASK; + + window.window = xcb_generate_id(window.conn); + + cookie = xcb_create_window_checked( + window.conn, + XCB_COPY_FROM_PARENT, + window.window, + screen.root, + 0, // x pos + 0, // y pos + window.w, // width + window.h, // height + 0, + XCB_WINDOW_CLASS_INPUT_OUTPUT, + screen.root_visual, + val_mask, + val_win.ptr + ); + error = xcb_request_check(window.conn, cookie); + CheckErr(&window, &cookie, error, "xcb_create_window failure"); + + cookie = xcb_map_window_checked(window.conn, window.window); + error = xcb_request_check(window.conn, cookie); + CheckErr(&window, &cookie, error, "xcb_map_window_checked failure"); + + cookie = xcb_change_property_checked( + window.conn, + XCB_PROP_MODE_REPLACE, + window.window, + XCB_ATOM_WM_NAME, + XCB_ATOM_STRING, + 8, + cast(u32)name.length, + name.ptr + ); + error = xcb_request_check(window.conn, cookie); + CheckErr(&window, &cookie, error, "xcb_change_property_checked failure"); + + xcb_intern_atom_cookie_t c_proto = xcb_intern_atom(window.conn, 1, 12, "WM_PROTOCOLS"); + xcb_intern_atom_reply_t *r_proto = xcb_intern_atom_reply(window.conn, c_proto, &error); + CheckErr(&window, &cookie, error, "xcb_intern_atom WM_PROTOCOLS failure"); + + xcb_intern_atom_cookie_t c_close = xcb_intern_atom(window.conn, 0, 16, "WM_DELETE_WINDOW"); + xcb_intern_atom_reply_t *r_close = xcb_intern_atom_reply(window.conn, c_close, &error); + CheckErr(&window, &cookie, error, "xcb_intern_atom WM_DELETE_WINDOW failure"); + + xcb_intern_atom_cookie_t c_minimize = xcb_intern_atom(window.conn, 0, 20, "_NET_WM_STATE_HIDDEN"); + xcb_intern_atom_reply_t *r_minimize = xcb_intern_atom_reply(window.conn, c_minimize, &error); + CheckErr(&window, &cookie, error, "xcb_intern_atom _NET_WM_STATE_HIDDEN failure"); + + + cookie = xcb_change_property_checked( + window.conn, + XCB_PROP_MODE_REPLACE, + window.window, + r_proto.atom, + XCB_ATOM_ATOM, + 32, + 1, + &r_close.atom + ); + error = xcb_request_check(window.conn, cookie); + CheckErr(&window, &cookie, error, "xcb_change_property_checked failure"); + + window.close_event = r_close.atom; + window.minimize_event = r_minimize.atom; + + free(r_proto); + free(r_close); + free(r_minimize); + + xcb_map_window(window.conn, window.window); + + i32 stream_result = xcb_flush(window.conn); + assert(stream_result > 0, "xcb_flush failure"); + + return window; +}; + +Library LoadLibrary(string name) +{ + Library lib = { + ptr: null, + }; + + lib.ptr = dlopen(name.ptr, RTLD_NOW); + + return lib; +}; + +Function LoadFunction(Library lib, string name) +{ + Function fn = { + ptr: null, + }; + + fn.ptr = dlsym(lib.ptr, name.ptr); + + return fn; +}; + +} + +version(Windows) +{ + +} diff --git a/src/renderer.d b/src/renderer.d new file mode 100644 index 0000000..a1ef793 --- /dev/null +++ b/src/renderer.d @@ -0,0 +1,4 @@ +import aliases; +import includes; + +alias Shader = VkShaderModule; diff --git a/src/tests.d b/src/tests.d new file mode 100644 index 0000000..698ecac --- /dev/null +++ b/src/tests.d @@ -0,0 +1,25 @@ +import aliases; +import u = util; + +void +RunTests() +{ + TestHashTable(); +} + +void +TestHashTable() +{ + auto ht = u.CreateHashTable!(u64, u64)(8); + + u64 value = 55; + ht[3] = value; + auto res = ht[3]; + assert(res.ok && value == res.value, "hash table failure"); + + res = ~ht[3]; + assert(res.ok && res.value == value, "hash table delete failure"); + + res = ~ht[3]; + assert(!res.ok, "value not deleted from hash table"); +} diff --git a/src/util.d b/src/util.d new file mode 100644 index 0000000..c537ab7 --- /dev/null +++ b/src/util.d @@ -0,0 +1,314 @@ +import aliases; +import includes; +import std.stdio; +import core.stdc.string : memset; +import a = alloc; + +void +Logf(Args...)(string fmt, Args args) +{ + writefln(fmt, args); +} + +void +Log(string str) +{ + writeln(str); +} + +void +Log(char* str) +{ + writeln(str); +} + +u64 +KB(u64 v) +{ + return v * 1024; +}; + +u64 +MB(u64 v) +{ + return KB(v) * 1024; +}; + +u64 +GB(u64 v) +{ + return MB(v) * 1024; +}; + +bool +BitEq(u64 l, u64 r) +{ + return (l & r) == r; +} + +struct Node(T) +{ + Node!(T)* next; + T value; +} + +struct SLList(T) +{ + Node!(T)* first; + Node!(T)* last; +} + +pragma(inline): bool +CheckNil(T)(Node!(T)* nil, Node!(T)* node) +{ + return node == null || node == nil; +} + +pragma(inline): void +ConcatInPlace(T)(SLList!(T)* list, SLList!(T)* to_concat) +{ + if (to_concat.first) + { + if (list.first) + { + list.last.next = to_concat.first; + list.last = to_concat.last; + } + else + { + list.first = to_concat.first; + list.last = to_concat.last; + } + + memset(to_concat, 0, SLList!(T).sizeof); + } +} + +pragma(inline): Node!(T)* +Pop(T)(SLList!(T)*list, Node!(T)* nil) +{ + Node!(T)* node = list.first; + + if (list.first == list.last) + { + list.first = list.last = nil; + } + else + { + list.first = list.first.next; + } + + return node; +} + +pragma(inline): void +PushFront(T)(SLList!(T)*list, Node!(T)* node, Node!(T)* nil) +{ + if (CheckNil(nil, list.first)) + { + list.first = list.last = node; + node.next = nil; + } + else + { + node.next = list.first; + list.first = node; + } +} + +pragma(inline): void +Push(T)(SLList!(T)*list, Node!(T)* node, Node!(T)* nil) +{ + if (CheckNil(nil, list.first)) + { + list.first = list.last = node; + node.next = nil; + } + else + { + list.last.next = node; + list.last = node; + node.next = nil; + } +} + +struct KVPair(K, V) +{ + K key; + V value; +} + +struct Result(V) +{ + V value; + bool ok; +} + +struct HashTable(K, V) +{ + alias P = KVPair!(K, V); + + SLList!(P) free_lists; + SLList!(P)[] lists; + Node!(P)* nil; + u64 node_count; + u64 list_count; + + void opIndexAssign(V value, K key) + { + Push(&this, key, value); + } + + Result!(V) opIndex(K key) + { + P* pair = Search(&this, key); + assert(pair != null, "HashTable key index failure: Result must be present"); + + Result!(V) result = { ok: false }; + if (pair != null) + { + result.value = pair.value; + result.ok = true; + } + + return result; + } + + Result!(V) opIndexUnary(string s: "~")(K key) + { + return Delete(&this, key); + } +} + +HashTable!(K, V) +CreateHashTable(K, V)(u64 size) +{ + auto nil = a.Alloc!(Node!(KVPair!(K, V))); + auto lists = a.AllocArray!(SLList!(KVPair!(K, V)))(size); + + HashTable!(K, V) table = { + lists: lists, + list_count: size, + nil: nil, + free_lists: { + first: nil, + last: nil, + }, + }; + + foreach(list; table.lists) + { + list.first = nil; + list.last = nil; + } + + return table; +} + +pragma(inline): void +Clear(K, V)(HashTable!(K, V)* ht) +{ + table.count = 0; + foreach(i, list; ht.lists) + { + ConcatInPlace(&ht.free_lists, ht.lists.ptr + i); + } +} + +pragma(inline): Node!(KVPair!(K, V))* +Push(K, V)(HashTable!(K, V)* ht, K key, V value) +{ + alias P = KVPair!(K, V); + alias N = Node!(P); + + N* node = ht.nil; + + if (ht.free_lists.first != ht.nil) + { + node = Pop(&ht.free_lists, ht.nil); + } + else + { + node = a.Alloc!(N); + } + + node.next = ht.nil; + node.value.key = key; + node.value.value = value; + + Push(GetList(ht, key), node, ht.nil); + + ht.node_count += 1; + + return node; +} + +pragma(inline): KVPair!(K, V)* +Search(K, V)(HashTable!(K, V)* ht, K key) +{ + KVPair!(K, V)* result = null; + + auto list = GetList(ht, key); + for(auto node = list.first; node != ht.nil; node = node.next) + { + if (node.value.key == key) + { + result = &node.value; + break; + } + } + + return result; +} + +pragma(inline): SLList!(KVPair!(K, V))* +GetList(K, V)(HashTable!(K, V)* ht, K key) +{ + u64 hash = Hash(&key); + u64 index = hash % ht.list_count; + return ht.lists.ptr + index; +} + +pragma(inline): Result!(V) +Delete(K, V)(HashTable!(K, V)* ht, K key) +{ + Result!(V) result = { ok: false }; + + auto list = GetList(ht, key); + auto prev = ht.nil; + for(auto node = list.first; node != ht.nil; node = node.next) + { + if (node.value.key == key) + { + if (prev != ht.nil) + { + prev.next = node.next; + } + + result.ok = true; + result.value = node.value.value; + + memset(&node.value, 0, node.value.sizeof); + + Push(&ht.free_lists, node, ht.nil); + + break; + } + } + + return result; +} + +const u64 HASH_SEED = 5995; + +pragma(inline): u64 +Hash(T)(T* value) +{ + return XXH3_64bits_withSeed(value, T.sizeof / u8.sizeof, HASH_SEED); +} + +pragma(inline): u64 +Hash(string str) +{ + return XXH3_64bits_withSeed(str.ptr, str.length, HASH_SEED); +} diff --git a/src/vulkan.d b/src/vulkan.d new file mode 100644 index 0000000..d20558d --- /dev/null +++ b/src/vulkan.d @@ -0,0 +1,1457 @@ +import vulkan_funcs; +import vulkan_logging; +import aliases; +import std.stdio; +import std.algorithm.comparison; +import core.stdc.string : strcmp; +import std.format : sformat; +import u = util : HashTable, Result, Logf, Log; +import a = alloc; +import p = platform; +import renderer; + +bool g_VLAYER_SUPPORT = false; + +const FRAME_OVERLAP = 2; + +version(linux) +{ + const string[] VULKAN_LIBS = [ + "libvulkan.so.1", + "libvulkan.so", + ]; +} + +version(Windows) +{ + const string[] VULKAN_LIBS = [ + "vulkan-1.dll", + ]; +} + +const char*[] VK_INSTANCE_LAYERS = []; +const char*[] VK_INSTANCE_LAYERS_DEBUG = [ "VK_LAYER_KHRONOS_validation" ]; + +version(linux) +{ + const char*[] VK_INSTANCE_EXT = [ cast(char*)VK_KHR_SURFACE_EXTENSION_NAME, cast(char*)VK_KHR_XCB_SURFACE_EXTENSION_NAME ]; + const char*[] VK_INSTANCE_EXT_DEBUG = VK_INSTANCE_EXT ~ [ cast(char*)VK_EXT_DEBUG_UTILS_EXTENSION_NAME ]; +} + +version(Windows) +{ + const char*[] VK_INSTANCE_EXT = [ cast(char*)VK_KHR_SURFACE_EXTENSION_NAME, cast(char*)VK_KHR_WIN32_SURFACE_EXTENSION_NAME ]; + const char*[] VK_INSTANCE_EXT_DEBUG = VK_INSTANCE_EXT ~ [ cast(char*)VK_EXT_DEBUG_UTILS_EXTENSION_NAME ]; +} + +const char*[] VK_DEVICE_EXTENSIONS = [ + cast(char*)VK_KHR_SWAPCHAIN_EXTENSION_NAME, + cast(char*)VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, +]; + +const VkFormat[] VK_IMAGE_FORMATS = [ + VK_FORMAT_R16G16B16A16_UNORM, + VK_FORMAT_R8G8B8A8_UNORM, +]; + +const VkImageUsageFlags VK_DRAW_IMAGE_USAGE_FLAGS = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | + VK_IMAGE_USAGE_TRANSFER_DST_BIT | + VK_IMAGE_USAGE_STORAGE_BIT | + VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; + +enum StepInitialized : u32 +{ + Renderer = 1, + Instance, + Debug, + Surface, + Device, + Vma, + MappedBuffers, // TODO + FrameStructures, + Swapchain, + DrawImages, + Descriptors, +} + +alias SI = StepInitialized; + +enum DescType : u32 +{ + Shared = 0, + SampledImage, + Material, + Mesh, + + Max, +} + +alias DT = DescType; + +struct PushConst +{ + u32 texture; +} + +struct Image +{ + VkImage image; + VmaAllocation alloc; + VkFormat format; + VkImageLayout layout; +} + +struct ImageView +{ + Image base; + VkImageView view; +} + +struct DescBindings +{ + u32[] free; + u64 count; + HashTable!(string, u32) lookup_table; +} + +struct Vulkan +{ + a.Arena arena; + a.Arena[FRAME_OVERLAP] frame_arenas; + + u.SLList!(SI) cleanup_list; + + p.Window* window; + + VkDebugUtilsMessengerEXT dbg_msg; + VkInstance instance; + VkSurfaceKHR surface; + VkPhysicalDevice physical_device; + VkDevice device; + VmaAllocator vma; + VkSwapchainKHR swapchain; + + VkSurfaceFormatKHR surface_format; + VkPresentModeKHR present_mode; + VkExtent3D swapchain_extent; + + ImageView[] present_images; + ImageView draw_image; + ImageView depth_image; + + VkCommandPool[FRAME_OVERLAP] cmd_pools; + VkCommandBuffer[FRAME_OVERLAP] cmds; + + VkSemaphore[] swapchain_sems; + VkSemaphore[FRAME_OVERLAP] render_sems; + VkFence[FRAME_OVERLAP] render_fences; + + VkCommandPool imm_pool; + VkCommandBuffer imm_cmd; + VkFence imm_fence; + + VkDescriptorPool desc_pool; + VkDescriptorSet[DT.max] desc_sets; + VkDescriptorSetLayout[DT.max] desc_layouts; + DescBindings[DT.max] desc_bindings; + + VkSampler nearest_sampler; + + VkPipelineLayout pipeline_layout; + + QueueInfo queues; +} + +struct QueueInfo +{ + i32 gfx_index, tfer_index; + VkQueue gfx_queue, tfer_queue; + bool single_queue; +}; + +u.Result!(Vulkan) +Init(p.Window* window, u64 permanent_mem, u64 frame_mem) +{ + bool success = true; + + Vulkan vk = { + arena: a.CreateArena(permanent_mem), + frame_arenas: [ + a.CreateArena(frame_mem), + a.CreateArena(frame_mem), + ], + window: window, + }; + + Push(&vk, SI.Renderer); + + success = LoadGlobalFunctions(); + + if (success) success = InitInstance(&vk); + + if (success) + { + LoadInstanceFunctions(&vk); + EnableVLayers(&vk); + } + + if (success) success = InitSurface(&vk); + if (success) success = InitDevice(&vk); + if (success) success = InitVMA(&vk); + if (success) success = CreateSwapchain(&vk); + if (success) success = CreateDrawImages(&vk); + if (success) success = InitFrameStructures(&vk); + if (success) success = InitDescriptors(&vk); + + u.Result!(Vulkan) result = { + ok: success, + value: vk, + }; + + return result; +} + +Result!(Shader) +BuildShader(Vulkan* vk, u8[] bytes) +{ + Result!(Shader) shader; + + VkShaderModuleCreateInfo shader_info = { + sType: VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, + codeSize: bytes.length, + pCode: cast(uint*)bytes.ptr, + }; + + VkResult result = vkCreateShaderModule(vk.device, &shader_info, null, &shader.value); + shader.ok = VkCheck("vkCreateShaderModule failure", result); + + return shader; +} + +Result!(VkPipeline) +CreateComputePipeline(Vulkan* vk, string shader) +{ + Result!(VkPipeline) pipeline; + + + + return pipeline; +} + +void +Destroy(Vulkan* vk) +{ + alias N = u.Node!(SI); + assert(vk.cleanup_list.first != null, "node null"); + for(N* node = vk.cleanup_list.first; node != null; node = node.next) + { + switch (node.value) + { + case SI.Renderer: + DestroyRenderer(vk); + break; + case SI.Instance: + Destroy(vk.instance); + break; + case SI.Debug: + Destroy(vk.dbg_msg, vk.instance); + break; + case SI.Surface: + Destroy(vk.surface, vk.instance); + break; + case SI.Device: + Destroy(vk.device); + break; + case SI.Vma: + Destroy(vk.vma); + break; + case SI.FrameStructures: + DestroyFS(vk); + break; + case SI.Swapchain: + Destroy(vk.swapchain, vk.present_images, vk.device); + break; + case SI.DrawImages: + Destroy(&vk.draw_image, &vk.depth_image, vk.device, vk.vma); + break; + case SI.Descriptors: + Destroy(vk.desc_pool, vk.desc_layouts, vk.pipeline_layout, vk.nearest_sampler, vk.device); + break; + default: + break; + } + } +} + +bool +InitDescriptors(Vulkan* vk) +{ + Push(vk, SI.Descriptors); + + bool success = true; + + VkDescriptorPoolSize[] pool_sizes = [ + { type: VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, descriptorCount: 4096 }, + { type: VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, descriptorCount: 4096 }, + { type: VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, descriptorCount: 4096 }, + { type: VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, descriptorCount: 4096 }, + { type: VK_DESCRIPTOR_TYPE_SAMPLER, descriptorCount: 4096 }, + ]; + + VkDescriptorPoolCreateInfo pool_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, + poolSizeCount: cast(u32)pool_sizes.length, + pPoolSizes: pool_sizes.ptr, + maxSets: 12, + flags: VK_DESCRIPTOR_POOL_CREATE_UPDATE_AFTER_BIND_BIT, + }; + + VkResult result = vkCreateDescriptorPool(vk.device, &pool_info, null, &vk.desc_pool); + success = VkCheck("vkCreateDescriptorPool failure", result); + + if (success) + { + VkDescriptorBindingFlags[4] shared_binding_flags; + shared_binding_flags[] = VK_DESCRIPTOR_BINDING_UPDATE_AFTER_BIND_BIT; + + VkDescriptorSetLayoutBindingFlagsCreateInfo shared_flag_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO, + bindingCount: cast(u32)shared_binding_flags.length, + pBindingFlags: shared_binding_flags.ptr, + }; + + VkDescriptorSetLayoutBinding[] shared_bindings = [ + { binding: 0, descriptorType: VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, descriptorCount: 1, stageFlags: VK_SHADER_STAGE_ALL }, + { binding: 1, descriptorType: VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, descriptorCount: 1, stageFlags: VK_SHADER_STAGE_ALL }, + { binding: 2, descriptorType: VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, descriptorCount: 1, stageFlags: VK_SHADER_STAGE_ALL }, + { binding: 3, descriptorType: VK_DESCRIPTOR_TYPE_SAMPLER, descriptorCount: 1, stageFlags: VK_SHADER_STAGE_ALL }, + ]; + + VkDescriptorSetLayoutCreateInfo shared_set_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + pNext: &shared_flag_info, + bindingCount: cast(u32)shared_bindings.length, + pBindings: shared_bindings.ptr, + flags: VK_DESCRIPTOR_SET_LAYOUT_CREATE_UPDATE_AFTER_BIND_POOL_BIT, + }; + + result = vkCreateDescriptorSetLayout(vk.device, &shared_set_info, null, &vk.desc_layouts[DT.Shared]); + success = VkCheck("vkCreateDescriptorSetLayout failure", result); + } + + if (success) + { + VkDescriptorType[DT.max] type_lookup = [ + DT.SampledImage: VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, + DT.Material: VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, + DT.Mesh: VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, + ]; + + VkDescriptorBindingFlags bindless_flag = VK_DESCRIPTOR_BINDING_UPDATE_AFTER_BIND_BIT | VK_DESCRIPTOR_BINDING_PARTIALLY_BOUND_BIT; + + VkDescriptorSetLayoutBindingFlagsCreateInfo bindless_flag_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO, + bindingCount: 1, + pBindingFlags: &bindless_flag, + }; + + VkDescriptorSetLayoutBinding binding = { + binding: 0, + descriptorCount: 1024, + stageFlags: VK_SHADER_STAGE_ALL, + }; + + VkDescriptorSetLayoutCreateInfo bindless_set_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + pNext: &bindless_flag_info, + bindingCount: 1, + pBindings: &binding, + flags: VK_DESCRIPTOR_SET_LAYOUT_CREATE_UPDATE_AFTER_BIND_POOL_BIT, + }; + + foreach(i; cast(u64)(DT.Shared+1) .. DT.max) + { + binding.descriptorType = type_lookup[i]; + + if (success) + { + result = vkCreateDescriptorSetLayout(vk.device, &bindless_set_info, null, vk.desc_layouts.ptr + i); + success = VkCheck("vkCreateDescriptorSetLayout failure", result); + } + } + } + + if (success) + { + VkDescriptorSetAllocateInfo alloc_info = { + sType: VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, + descriptorSetCount: cast(u32)DT.max, + pSetLayouts: vk.desc_layouts.ptr, + descriptorPool: vk.desc_pool, + }; + + result = vkAllocateDescriptorSets(vk.device, &alloc_info, vk.desc_sets.ptr); + success = VkCheck("vkAllocateDescriptorSets failure", result); + } + + if (success) + { + VkPushConstantRange const_range = { + offset: 0, + size: cast(VkDeviceSize)PushConst.sizeof, + stageFlags: VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_FRAGMENT_BIT | VK_SHADER_STAGE_VERTEX_BIT, + }; + + VkPipelineLayoutCreateInfo layout_info = { + sType: VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + setLayoutCount: cast(u32)DT.max, + pushConstantRangeCount: 1, + pPushConstantRanges: &const_range, + pSetLayouts: vk.desc_layouts.ptr, + }; + + result = vkCreatePipelineLayout(vk.device, &layout_info, null, &vk.pipeline_layout); + success = VkCheck("vkCreatePipelineLayout failure", result); + } + + if (success) + { + foreach(i; cast(u64)DT.min .. cast(u64)DT.max) + { + vk.desc_bindings[i].lookup_table = u.CreateHashTable!(string, u32)(8); + + u32 DESC_MAX_BINDINGS = 512; + vk.desc_bindings[i].free = a.AllocArray!(u32)(&vk.arena, DESC_MAX_BINDINGS); + + u32 free_count = 0; + for(i32 j = DESC_MAX_BINDINGS-1; j >= 0; j -= 1) + { + vk.desc_bindings[i].free[j] = cast(u32)free_count; + free_count += 1; + } + + vk.desc_bindings[i].count = free_count; + } + } + + if (success) + { + VkPhysicalDeviceProperties props; + vkGetPhysicalDeviceProperties(vk.physical_device, &props); + + VkSamplerCreateInfo sampler_info = { + sType: VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + magFilter: VK_FILTER_NEAREST, + minFilter: VK_FILTER_NEAREST, + addressModeU: VK_SAMPLER_ADDRESS_MODE_REPEAT, + addressModeV: VK_SAMPLER_ADDRESS_MODE_REPEAT, + addressModeW: VK_SAMPLER_ADDRESS_MODE_REPEAT, + anisotropyEnable: VK_TRUE, + maxAnisotropy: props.limits.maxSamplerAnisotropy, + borderColor: VK_BORDER_COLOR_INT_OPAQUE_BLACK, + compareOp: VK_COMPARE_OP_ALWAYS, + mipmapMode: VK_SAMPLER_MIPMAP_MODE_LINEAR, + }; + + result = vkCreateSampler(vk.device, &sampler_info, null, &vk.nearest_sampler); + success = VkCheck("vkCreateSampler failure", result); + } + + if (success) + { + VkDescriptorImageInfo sampler_info = { + sampler: vk.nearest_sampler, + }; + + VkWriteDescriptorSet write = { + sType: VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + dstSet: vk.desc_sets[DT.Shared], + dstBinding: 3, + descriptorCount: 1, + descriptorType: VK_DESCRIPTOR_TYPE_SAMPLER, + pImageInfo: &sampler_info, + }; + + vkUpdateDescriptorSets(vk.device, 1, &write, 0, null); + } + + return success; +} + +bool +VkCheck(string message, VkResult result) +{ + bool success = true; + + if (result != VK_SUCCESS) + { + success = false; + char[512] buf; + buf.sformat("%s: %s", message, VkResultStr(result)); + Log(buf.ptr); + } + + return success; +} + +bool +InitFrameStructures(Vulkan* vk) +{ + Push(vk, SI.FrameStructures); + + bool success = true; + a.Arena* arena = &vk.frame_arenas[0]; + + VkSemaphoreCreateInfo sem_info = { sType: VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO }; + + VkCommandPoolCreateInfo pool_info = { + sType: VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, + flags: VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, + }; + + VkFenceCreateInfo fence_info = { + sType: VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, + flags: VK_FENCE_CREATE_SIGNALED_BIT, + }; + + VkCommandBufferAllocateInfo cmd_info = { + sType: VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, + commandBufferCount: 1, + level: VK_COMMAND_BUFFER_LEVEL_PRIMARY, + }; + + vk.swapchain_sems = a.AllocArray!(VkSemaphore)(arena, vk.present_images.length); + + foreach(i; 0 .. vk.swapchain_sems.length) + { + VkResult result = vkCreateSemaphore(vk.device, &sem_info, null, vk.swapchain_sems.ptr + i); + success = VkCheck("vkCreateSemaphore failure", result); + } + + foreach(i; 0 .. FRAME_OVERLAP) + { + VkResult result; + + if (success) + { + result = vkCreateCommandPool(vk.device, &pool_info, null, vk.cmd_pools.ptr + i); + success = VkCheck("vkCreateCommandPool failure", result); + } + + if (success) + { + cmd_info.commandPool = vk.cmd_pools[i]; + result = vkAllocateCommandBuffers(vk.device, &cmd_info, vk.cmds.ptr + i); + success = VkCheck("vkAllocateCommandBuffers failure", result); + } + + if (success) + { + result = vkCreateFence(vk.device, &fence_info, null, vk.render_fences.ptr + i); + success = VkCheck("vkCreateFence failure", result); + } + + if (success) + { + result = vkCreateSemaphore(vk.device, &sem_info, null, vk.render_sems.ptr + i); + success = VkCheck("vkCreateSemaphore failure", result); + } + } + + if (success) + { + VkResult result = vkCreateCommandPool(vk.device, &pool_info, null, &vk.imm_pool); + success = VkCheck("vkCreateCommandPool failure", result); + } + + if (success) + { + cmd_info.commandPool = vk.imm_pool; + VkResult result = vkAllocateCommandBuffers(vk.device, &cmd_info, &vk.imm_cmd); + success = VkCheck("vkAllocateCommandBuffers failure", result); + } + + if (success) + { + VkResult result = vkCreateFence(vk.device, &fence_info, null, &vk.imm_fence); + success = VkCheck("vkCreateFence failure", result); + } + + return success; +} + +VkFormat +GetDrawImageFormat(Vulkan* vk) +{ + VkFormat selected_format; + foreach(format; VK_IMAGE_FORMATS) + { + VkImageFormatProperties props; + VkResult result = vkGetPhysicalDeviceImageFormatProperties( + vk.physical_device, + format, + VK_IMAGE_TYPE_2D, + VK_IMAGE_TILING_OPTIMAL, + VK_DRAW_IMAGE_USAGE_FLAGS, + 0, + &props + ); + if (result == VK_ERROR_FORMAT_NOT_SUPPORTED) + { + continue; + } + + if (result == VK_SUCCESS) + { + selected_format = format; + break; + } + } + + return selected_format; +} + +bool +CreateDrawImages(Vulkan* vk) +{ + Push(vk, SI.DrawImages); + + bool success = true; + + VkFormat draw_format = GetDrawImageFormat(vk); + VkFormat depth_format = VK_FORMAT_D32_SFLOAT; + + VmaAllocationCreateInfo alloc_info = { + usage: VMA_MEMORY_USAGE_GPU_ONLY, + requiredFlags: VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, + }; + + VkImageCreateInfo image_info = { + sType: VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, + imageType: VK_IMAGE_TYPE_2D, + mipLevels: 1, + arrayLayers: 1, + samples: VK_SAMPLE_COUNT_1_BIT, + tiling: VK_IMAGE_TILING_OPTIMAL, + usage: VK_DRAW_IMAGE_USAGE_FLAGS, + extent: vk.swapchain_extent, + format: draw_format, + }; + + VkResult result = vmaCreateImage(vk.vma, &image_info, &alloc_info, &vk.draw_image.base.image, &vk.draw_image.base.alloc, null); + success = VkCheck("vmaCreateImage failure", result); + + if (success) + { + VkImageViewCreateInfo view_info = { + sType: VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + image: vk.draw_image.base.image, + format: draw_format, + viewType: VK_IMAGE_VIEW_TYPE_2D, + subresourceRange: { + aspectMask: VK_IMAGE_ASPECT_COLOR_BIT, + baseMipLevel: 0, + levelCount: 1, + baseArrayLayer: 0, + layerCount: 1, + }, + }; + + result = vkCreateImageView(vk.device, &view_info, null, &vk.draw_image.view); + success = VkCheck("vkCreateImageView failure", result); + } + + if (success) + { + VkImageCreateInfo depth_image_info = { + sType: VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, + imageType: VK_IMAGE_TYPE_2D, + mipLevels: 1, + arrayLayers: 1, + samples: VK_SAMPLE_COUNT_1_BIT, + tiling: VK_IMAGE_TILING_OPTIMAL, + format: depth_format, + usage: VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT, + extent: vk.swapchain_extent, + }; + + result = vmaCreateImage(vk.vma, &depth_image_info, &alloc_info, &vk.depth_image.base.image, &vk.depth_image.base.alloc, null); + success = VkCheck("vmaCreateImage failure", result); + } + + if (success) + { + VkImageViewCreateInfo depth_view_info = { + sType: VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + image: vk.depth_image.base.image, + viewType: VK_IMAGE_VIEW_TYPE_2D, + format: depth_format, + subresourceRange: { + aspectMask: VK_IMAGE_ASPECT_DEPTH_BIT, + baseMipLevel: 0, + levelCount: 1, + baseArrayLayer: 0, + layerCount: 1, + }, + }; + + result = vkCreateImageView(vk.device, &depth_view_info, null, &vk.depth_image.view); + success = VkCheck("vmaCreateImageView failure", result); + } + + vk.draw_image.base.format = draw_format; + vk.draw_image.base.layout = VK_IMAGE_LAYOUT_UNDEFINED; + vk.depth_image.base.format = depth_format; + vk.depth_image.base.layout = VK_IMAGE_LAYOUT_UNDEFINED; + + return success; +} + +void +SelectSwapchainFormats(Vulkan* vk) +{ + a.Arena* arena = &vk.frame_arenas[0]; + + u32 format_count; + vkGetPhysicalDeviceSurfaceFormatsKHR(vk.physical_device, vk.surface, &format_count, null); + VkSurfaceFormatKHR[] formats = a.AllocArray!(VkSurfaceFormatKHR)(arena, format_count); + vkGetPhysicalDeviceSurfaceFormatsKHR(vk.physical_device, vk.surface, &format_count, formats.ptr); + + u32 mode_count; + vkGetPhysicalDeviceSurfacePresentModesKHR(vk.physical_device, vk.surface, &mode_count, null); + VkPresentModeKHR[] modes = a.AllocArray!(VkPresentModeKHR)(arena, mode_count); + vkGetPhysicalDeviceSurfacePresentModesKHR(vk.physical_device, vk.surface, &mode_count, modes.ptr); + + VkPresentModeKHR present_mode = VK_PRESENT_MODE_FIFO_KHR; + foreach(mode; modes) + { + if (mode == VK_PRESENT_MODE_MAILBOX_KHR) + { + present_mode = VK_PRESENT_MODE_MAILBOX_KHR; + break; + } + } + + vk.surface_format = formats[0]; + vk.present_mode = present_mode; +} + +bool +CreateSwapchain(Vulkan* vk) +{ + Push(vk, SI.Swapchain); + + bool success = true; + a.Arena* arena = &vk.frame_arenas[0]; + + VkSurfaceCapabilitiesKHR cap; + vkGetPhysicalDeviceSurfaceCapabilitiesKHR(vk.physical_device, vk.surface, &cap); + + static bool initialized = false; + if (!initialized) + { + SelectSwapchainFormats(vk); + } + + VkSwapchainCreateInfoKHR info = { + sType: VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR, + imageArrayLayers: 1, + imageUsage: VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT, + compositeAlpha: VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR, + clipped: VK_TRUE, + imageSharingMode: VK_SHARING_MODE_EXCLUSIVE, + minImageCount: cap.minImageCount + 1, + surface: vk.surface, + imageFormat: vk.surface_format.format, + imageColorSpace: vk.surface_format.colorSpace, + imageExtent: { + width: clamp(cast(u32)vk.window.w, cap.minImageExtent.width, cap.maxImageExtent.width), + height: clamp(cast(u32)vk.window.h, cap.minImageExtent.height, cap.maxImageExtent.height), + }, + preTransform: cap.currentTransform, + presentMode: vk.present_mode, + }; + + VkResult result = vkCreateSwapchainKHR(vk.device, &info, null, &vk.swapchain); + success = VkCheck("vkCreateSwapchainKHR failure", result); + + u32 count; + vkGetSwapchainImagesKHR(vk.device, vk.swapchain, &count, null); + VkImage[] images = a.AllocArray!(VkImage)(arena, count); + vkGetSwapchainImagesKHR(vk.device, vk.swapchain, &count, images.ptr); + VkImageView[] views = a.AllocArray!(VkImageView)(arena, count); + + vk.present_images = a.AllocArray!(ImageView)(&vk.arena, count); + + VkImageViewCreateInfo view_info = { + sType: VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + viewType: VK_IMAGE_VIEW_TYPE_2D, + components: { + r: VK_COMPONENT_SWIZZLE_IDENTITY, + g: VK_COMPONENT_SWIZZLE_IDENTITY, + b: VK_COMPONENT_SWIZZLE_IDENTITY, + a: VK_COMPONENT_SWIZZLE_IDENTITY, + }, + subresourceRange: { + aspectMask: VK_IMAGE_ASPECT_COLOR_BIT, + baseMipLevel: 0, + levelCount: 1, + baseArrayLayer: 0, + layerCount: 1, + }, + }; + + foreach(i, image; vk.present_images) + { + vk.present_images[i].base.image = images[i]; + view_info.image = images[i]; + view_info.format = vk.surface_format.format; + + result = vkCreateImageView(vk.device, &view_info, null, &vk.present_images[i].view); + success = VkCheck("vkCreateImageView failure", result); + } + + vk.swapchain_extent.width = info.imageExtent.width; + vk.swapchain_extent.height = info.imageExtent.height; + vk.swapchain_extent.depth = 1; + + if (!initialized && success) + { + initialized = true; + } + + return success; +} + +bool +InitVMA(Vulkan* vk) +{ + Push(vk, SI.Vma); + + bool success = true; + + VmaVulkanFunctions vk_functions = { + vkGetInstanceProcAddr: vkGetInstanceProcAddr, + vkGetDeviceProcAddr: vkGetDeviceProcAddr, + }; + + VmaAllocatorCreateInfo info = { + flags: VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT, + vulkanApiVersion: VK_MAKE_API_VERSION(0, 1, 3, 0), + pVulkanFunctions: &vk_functions, + physicalDevice: vk.physical_device, + device: vk.device, + instance: vk.instance, + }; + + VkResult result = vmaCreateAllocator(&info, &vk.vma); + success = VkCheck("vmaCreateAllocator failure", result); + + return success; +} + +bool +InitDevice(Vulkan* vk) +{ + Push(vk, SI.Device); + + bool success = false; + a.Arena* arena = &vk.frame_arenas[0]; + + u32 count; + vkEnumeratePhysicalDevices(vk.instance, &count, null); + VkPhysicalDevice[] devices = a.AllocArray!(VkPhysicalDevice)(arena, count); + vkEnumeratePhysicalDevices(vk.instance, &count, devices.ptr); + + VkPhysicalDevice physical_device = null; + bool discrete_candidate = false; + QueueInfo candidate = { + gfx_index: -1, + tfer_index: -1, + single_queue: false, + }; + + foreach(dev; devices) + { + QueueInfo current = CheckQueueProperties(arena, dev, vk.surface); + b32 discrete = false; + + if (current.gfx_index < 0) + continue; + if (!CheckDeviceProperties(arena, dev, vk.surface, &discrete)) + continue; + if (discrete_candidate && !discrete) + continue; + if (!CheckDeviceFeatures(dev)) + continue; + + discrete_candidate = cast(bool)discrete; + candidate = current; + physical_device = dev; + + if (discrete_candidate && !candidate.single_queue) + continue; + } + + if (physical_device) + { + VkDeviceQueueCreateInfo[2] queue_info; + f32 priority = 1.0f; + count = 1; + + queue_info[0].sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO; + queue_info[0].queueFamilyIndex = candidate.gfx_index; + queue_info[0].queueCount = 1; + queue_info[0].pQueuePriorities = &priority; + queue_info[0].flags = 0; + + if (!candidate.single_queue) + { + queue_info[1].sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO; + queue_info[1].queueFamilyIndex = candidate.tfer_index; + queue_info[1].queueCount = 1; + queue_info[1].pQueuePriorities = &priority; + queue_info[1].flags = 0; + + count += 1; + } + + VkPhysicalDeviceVulkan13Features features_13 = { + sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES, + synchronization2: VK_TRUE, + dynamicRendering: VK_TRUE, + }; + + VkPhysicalDeviceVulkan12Features features_12 = { + sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES, + pNext: &features_13, + descriptorIndexing: VK_TRUE, + bufferDeviceAddress: VK_TRUE, + descriptorBindingUniformBufferUpdateAfterBind: VK_TRUE, + descriptorBindingSampledImageUpdateAfterBind: VK_TRUE, + descriptorBindingStorageImageUpdateAfterBind: VK_TRUE, + descriptorBindingStorageBufferUpdateAfterBind: VK_TRUE, + descriptorBindingPartiallyBound: VK_TRUE, + shaderSampledImageArrayNonUniformIndexing: VK_TRUE, + shaderUniformBufferArrayNonUniformIndexing: VK_TRUE, + runtimeDescriptorArray: VK_TRUE, + }; + + VkPhysicalDeviceVulkan11Features features_11 = { + sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES, + pNext: &features_12, + }; + + VkPhysicalDeviceFeatures2 features = { + sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2, + pNext: &features_11, + features: { + shaderUniformBufferArrayDynamicIndexing: VK_TRUE, + shaderSampledImageArrayDynamicIndexing: VK_TRUE, + shaderStorageBufferArrayDynamicIndexing: VK_TRUE, + shaderStorageImageArrayDynamicIndexing: VK_TRUE, + samplerAnisotropy: VK_TRUE, + }, + }; + + VkDeviceCreateInfo device_info = { + sType: VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO, + pNext: &features, + ppEnabledExtensionNames: VK_DEVICE_EXTENSIONS.ptr, + enabledExtensionCount: cast(u32)VK_DEVICE_EXTENSIONS.length, + queueCreateInfoCount: count, + pQueueCreateInfos: queue_info.ptr, + pEnabledFeatures: null, + }; + + VkResult result = vkCreateDevice(physical_device, &device_info, null, &vk.device); + if (result != VK_SUCCESS) + { + Logf("vkCreateDevices failure: %s", VkResultStr(result)); + } + else + { + LoadDeviceFunctions(vk); + + vkGetDeviceQueue( + vk.device, + candidate.gfx_index, + 0, + &candidate.gfx_queue + ); + + if (!candidate.single_queue) + { + vkGetDeviceQueue( + vk.device, + candidate.tfer_index, + candidate.tfer_index == candidate.gfx_index ? 1 : 0, + &candidate.tfer_queue + ); + } + else + { + candidate.tfer_queue = candidate.gfx_queue; + candidate.tfer_index = candidate.gfx_index; + } + + vk.physical_device = physical_device; + vk.queues = candidate; + + success = true; + } + } + + return success; +} + +bool +CheckDeviceFeatures(VkPhysicalDevice device) +{ + VkPhysicalDeviceFeatures2 features2 = { sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2 }; + VkPhysicalDeviceVulkan12Features features_12 = { sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES }; + VkPhysicalDeviceVulkan13Features features_13 = { sType: VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES }; + + features2.pNext = &features_12; + vkGetPhysicalDeviceFeatures2(device, &features2); + features2.pNext = &features_13; + vkGetPhysicalDeviceFeatures2(device, &features2); + + VkPhysicalDeviceFeatures features = features2.features; + bool result = true; + + result &= cast(bool)features.shaderUniformBufferArrayDynamicIndexing; + result &= cast(bool)features.shaderSampledImageArrayDynamicIndexing; + result &= cast(bool)features.shaderStorageBufferArrayDynamicIndexing; + result &= cast(bool)features.shaderStorageImageArrayDynamicIndexing; + result &= cast(bool)features.samplerAnisotropy; + + result &= cast(bool)features_12.descriptorIndexing; + result &= cast(bool)features_12.bufferDeviceAddress; + result &= cast(bool)features_12.descriptorBindingUniformBufferUpdateAfterBind; + result &= cast(bool)features_12.descriptorBindingSampledImageUpdateAfterBind; + result &= cast(bool)features_12.descriptorBindingStorageImageUpdateAfterBind; + result &= cast(bool)features_12.descriptorBindingStorageBufferUpdateAfterBind; + result &= cast(bool)features_12.descriptorBindingPartiallyBound; + result &= cast(bool)features_12.runtimeDescriptorArray; + result &= cast(bool)features_12.shaderSampledImageArrayNonUniformIndexing; + result &= cast(bool)features_12.shaderUniformBufferArrayNonUniformIndexing; + result &= cast(bool)features_12.timelineSemaphore; + + result &= cast(bool)features_13.synchronization2; + result &= cast(bool)features_13.dynamicRendering; + + return result; +} + +bool +CheckDeviceProperties(a.Arena *arena, VkPhysicalDevice device, VkSurfaceKHR surface, b32* discrete) +{ + bool success = false; + + VkPhysicalDeviceProperties props; + vkGetPhysicalDeviceProperties(device, &props); + + if (VK_API_VERSION_MINOR(props.apiVersion) >= 3) + { + u32 ext_count; + vkEnumerateDeviceExtensionProperties(device, null, &ext_count, null); + VkExtensionProperties[] ext_props = a.AllocArray!(VkExtensionProperties)(arena, ext_count); + vkEnumerateDeviceExtensionProperties(device, null, &ext_count, ext_props.ptr); + + i32 matched = 0; + foreach(prop; ext_props) + { + foreach(ext; VK_DEVICE_EXTENSIONS) + { + if (strcmp(cast(char*)prop.extensionName, ext) == 0) + { + matched += 1; + break; + } + } + } + + if (matched == VK_DEVICE_EXTENSIONS.length) + { + u32 fmt_count, present_count; + vkGetPhysicalDeviceSurfaceFormatsKHR(device, surface, &fmt_count, null); + vkGetPhysicalDeviceSurfacePresentModesKHR(device, surface, &present_count, null); + + *discrete = props.deviceType == VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU; + success = fmt_count && present_count; + } + } + + return success; +} + +QueueInfo +CheckQueueProperties(a.Arena *arena, VkPhysicalDevice device, VkSurfaceKHR surface) +{ + const u32 T_BIT = VK_QUEUE_TRANSFER_BIT; + const u32 C_BIT = VK_QUEUE_COMPUTE_BIT; + const u32 G_BIT = VK_QUEUE_GRAPHICS_BIT; + const u32 S_BIT = VK_QUEUE_SPARSE_BINDING_BIT; + + QueueInfo current = { + gfx_index: -1, + tfer_index: -1, + single_queue: false, + }; + + u32 count; + vkGetPhysicalDeviceQueueFamilyProperties(device, &count, null); + VkQueueFamilyProperties[] properties = a.AllocArray!(VkQueueFamilyProperties)(arena, count); + vkGetPhysicalDeviceQueueFamilyProperties(device, &count, properties.ptr); + + if (count == 1 && properties[0].queueCount == 1 && u.BitEq(properties[0].queueFlags, T_BIT | C_BIT | G_BIT)) + { + current.gfx_index = current.tfer_index = 0; + current.single_queue = true; + } + else + { + bool sparse = false, tfer_only = false; + + foreach(i, prop; properties) + { + b32 surface_support; + vkGetPhysicalDeviceSurfaceSupportKHR(device, cast(u32)i, surface, &surface_support); + + if (current.gfx_index < 0 && surface_support && u.BitEq(prop.queueFlags, G_BIT)) + { + current.gfx_index = cast(i32)i; + continue; + } + + if (u.BitEq(prop.queueFlags, T_BIT | S_BIT) && !u.BitEq(prop.queueFlags, G_BIT | C_BIT)) + { + sparse = true; + tfer_only = true; + current.tfer_index = cast(i32)i; + continue; + } + + if (!(sparse && tfer_only) && u.BitEq(prop.queueFlags, T_BIT | S_BIT)) + { + sparse = true; + current.tfer_index = cast(i32)i; + continue; + } + + if (!sparse && !u.BitEq(prop.queueFlags, T_BIT) && u.BitEq(prop.queueFlags, C_BIT)) + { + tfer_only = true; + current.tfer_index = cast(i32)i; + continue; + } + + if (!sparse && !tfer_only && u.BitEq(prop.queueFlags, C_BIT)) + { + current.tfer_index = cast(i32)i; + } + } + + if (current.tfer_index < 0) + { + current.tfer_index = current.gfx_index; + } + } + + return current; +} + +pragma(inline): void +Push(Vulkan* vk, StepInitialized step) +{ + u.Node!(SI)* node = a.Alloc!(u.Node!(SI)); + node.value = step; + u.PushFront(&vk.cleanup_list, node, null); +} + +void +DestroyRenderer(Vulkan* vk) +{ + foreach(i, arena; vk.frame_arenas) + { + a.Free(vk.frame_arenas.ptr + i); + } + + a.Free(&vk.arena); +} + +void +Destroy(VkInstance instance) +{ + if (instance) + { + vkDestroyInstance(instance, null); + } +} + +void +Destroy(VkDebugUtilsMessengerEXT dbg, VkInstance instance) +{ + if (dbg) + { + vkDestroyDebugUtilsMessengerEXT(instance, dbg, null); + } +} + +void +Destroy(VkSurfaceKHR surface, VkInstance instance) +{ + if (surface) + { + vkDestroySurfaceKHR(instance, surface, null); + } +} + +void +Destroy(VkDevice device) +{ + if (device) + { + vkDestroyDevice(device, null); + } +} + +void +Destroy(VmaAllocator vma) +{ + if (vma) + { + vmaDestroyAllocator(vma); + } +} + +void +Destroy(VkSwapchainKHR swapchain, ImageView[] views, VkDevice device) +{ + foreach(view; views) + { + if (view.view) + { + vkDestroyImageView(device, view.view, null); + } + } + + if (swapchain) + { + vkDestroySwapchainKHR(device, swapchain, null); + } +} + +void +Destroy(ImageView* view, VkDevice device, VmaAllocator vma) +{ + if (view.view) + { + vkDestroyImageView(device, view.view, null); + } + + if (view.base.image) + { + vmaDestroyImage(vma, view.base.image, view.base.alloc); + } +} + +void +Destroy(VkDescriptorPool pool, VkDescriptorSetLayout[] layouts, VkPipelineLayout pipeline_layout, VkSampler sampler, VkDevice device) +{ + if (sampler) + { + vkDestroySampler(device, sampler, null); + } + + if (pipeline_layout) + { + vkDestroyPipelineLayout(device, pipeline_layout, null); + } + + foreach(layout; layouts) + { + if (layout) + { + vkDestroyDescriptorSetLayout(device, layout, null); + } + } + + if (pool) + { + vkDestroyDescriptorPool(device, pool, null); + } +} + +void +Destroy(ImageView* draw, ImageView* depth, VkDevice device, VmaAllocator vma) +{ + Destroy(draw, device, vma); + Destroy(depth, device, vma); +} + +void +DestroyFS(Vulkan* vk) +{ + if (vk.imm_fence) + { + vkDestroyFence(vk.device, vk.imm_fence, null); + } + + if (vk.imm_cmd) + { + vkFreeCommandBuffers(vk.device, vk.imm_pool, 1, &vk.imm_cmd); + } + + if (vk.imm_pool) + { + vkDestroyCommandPool(vk.device, vk.imm_pool, null); + } + + foreach(i, sem; vk.swapchain_sems) + { + if (sem) + { + vkDestroySemaphore(vk.device, sem, null); + } + } + + foreach(i; 0 .. FRAME_OVERLAP) + { + if (vk.render_sems[i]) + { + vkDestroySemaphore(vk.device, vk.render_sems[i], null); + } + + if (vk.render_fences[i]) + { + vkDestroyFence(vk.device, vk.render_fences[i], null); + } + + if (vk.cmd_pools[i]) + { + vkFreeCommandBuffers(vk.device, vk.cmd_pools[i], 1, &vk.cmds[i]); + } + + if (vk.cmd_pools[i]) + { + vkDestroyCommandPool(vk.device, vk.cmd_pools[i], null); + } + } +} + +bool +InitInstance(Vulkan* vk) +{ + Push(vk, SI.Instance); + + bool success = true; + a.Arena* arena = &vk.frame_arenas[0]; + + u32 count; + vkEnumerateInstanceLayerProperties(&count, null); + VkLayerProperties[] layers = a.AllocArray!(VkLayerProperties)(arena, count); + vkEnumerateInstanceLayerProperties(&count, layers.ptr); + + foreach(i, layer; layers) + { + if (strcmp(cast(char*)&layer.layerName, "VK_LAYER_KHRONOS_validation") == 0) + { + g_VLAYER_SUPPORT = true; + break; + } + } + + const char*[] instance_layers = g_VLAYER_SUPPORT && BUILD_DEBUG ? VK_INSTANCE_LAYERS_DEBUG : VK_INSTANCE_LAYERS; + const char*[] instance_ext = g_VLAYER_SUPPORT && BUILD_DEBUG ? VK_INSTANCE_EXT_DEBUG : VK_INSTANCE_EXT; + + VkApplicationInfo app_info = { + sType: VK_STRUCTURE_TYPE_APPLICATION_INFO, + pApplicationName: "Video Game", + applicationVersion: VK_MAKE_API_VERSION(0, 0, 0, 1), + pEngineName: "Gears", + engineVersion: VK_MAKE_API_VERSION(0, 0, 0, 1), + apiVersion: VK_MAKE_API_VERSION(0, 1, 3, 0), + }; + + VkInstanceCreateInfo instance_info = { + sType: VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO, + pApplicationInfo: &app_info, + enabledLayerCount: cast(u32)instance_layers.length, + ppEnabledLayerNames: instance_layers.ptr, + enabledExtensionCount: cast(u32)instance_ext.length, + ppEnabledExtensionNames: instance_ext.ptr, + }; + + VkResult result = vkCreateInstance(&instance_info, null, &vk.instance); + success = VkCheck("vkCreateInstance failure", result); + + return success; +} + +bool +InitSurface(Vulkan* vk) +{ + Push(vk, SI.Surface); + +version(linux) +{ + VkXcbSurfaceCreateInfoKHR surface_info = { + sType: VK_STRUCTURE_TYPE_XCB_SURFACE_CREATE_INFO_KHR, + connection: vk.window.conn, + window: vk.window.window, + }; + + VkResult result = vkCreateXcbSurfaceKHR(vk.instance, &surface_info, null, &vk.surface); +} + +version(Windows) +{ + VkWin32SurfaceCreateInfoKHR surface_info = { + sType: VK_STRUCTURE_TYPE_WIN32_SURFACE_CREATE_INFO_KHR, + hinstance: vk.window.instance, + hwnd: vk.window.handle, + }; + + VkResult result = vkCreateWin32SurfaceKHR(vk.instance, &surface_info, null, &vk.surface); +} + + bool success = VkCheck("InitSurface failure", result); + + return success; +} + + +void +EnableVLayers(Vulkan* vk) +{ + debug + { + Push(vk, SI.Debug); + + if (g_VLAYER_SUPPORT) + { + VkDebugUtilsMessengerCreateInfoEXT info = { + sType: VK_STRUCTURE_TYPE_DEBUG_UTILS_MESSENGER_CREATE_INFO_EXT, + messageSeverity: VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT, + messageType: VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT, + pfnUserCallback: cast(PFN_vkDebugUtilsMessengerCallbackEXT)&DebugCallback, + }; + + VkResult result = vkCreateDebugUtilsMessengerEXT(vk.instance, &info, null, &vk.dbg_msg); + if (result != VK_SUCCESS) + { + Logf("EnableVLayers failed to initialize, will continue without validation: %s", VkResultStr(result)); + } + } + else + { + Logf("EnableVLayers warning: Not supported on current device, continuing without"); + } + } +} + diff --git a/src/vulkan_funcs.d b/src/vulkan_funcs.d new file mode 100644 index 0000000..6a72b86 --- /dev/null +++ b/src/vulkan_funcs.d @@ -0,0 +1,246 @@ +public import includes; +import p = platform; +import vulkan : Vulkan, VULKAN_LIBS; + +// Global Functions + +PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr = null; +PFN_vkCreateInstance vkCreateInstance = null; +PFN_vkEnumerateInstanceLayerProperties vkEnumerateInstanceLayerProperties = null; + +// Instance Functions + +PFN_vkEnumeratePhysicalDevices vkEnumeratePhysicalDevices = null; +PFN_vkCreateDevice vkCreateDevice = null; +PFN_vkGetPhysicalDeviceQueueFamilyProperties vkGetPhysicalDeviceQueueFamilyProperties = null; +PFN_vkGetPhysicalDeviceSurfaceSupportKHR vkGetPhysicalDeviceSurfaceSupportKHR = null; +PFN_vkGetPhysicalDeviceProperties vkGetPhysicalDeviceProperties = null; +PFN_vkGetPhysicalDeviceFeatures2 vkGetPhysicalDeviceFeatures2 = null; +PFN_vkEnumerateDeviceExtensionProperties vkEnumerateDeviceExtensionProperties = null; +PFN_vkGetPhysicalDeviceSurfacePresentModesKHR vkGetPhysicalDeviceSurfacePresentModesKHR = null; +PFN_vkGetPhysicalDeviceSurfaceFormatsKHR vkGetPhysicalDeviceSurfaceFormatsKHR = null; +PFN_vkGetPhysicalDeviceSurfaceCapabilitiesKHR vkGetPhysicalDeviceSurfaceCapabilitiesKHR = null; +PFN_vkGetPhysicalDeviceImageFormatProperties vkGetPhysicalDeviceImageFormatProperties = null; +PFN_vkGetDeviceProcAddr vkGetDeviceProcAddr = null; +PFN_vkDestroyInstance vkDestroyInstance = null; +PFN_vkDestroySurfaceKHR vkDestroySurfaceKHR = null; + +debug +{ + PFN_vkCreateDebugUtilsMessengerEXT vkCreateDebugUtilsMessengerEXT = null; + PFN_vkDestroyDebugUtilsMessengerEXT vkDestroyDebugUtilsMessengerEXT = null; +} + +// Platform Functions + +version(linux) +{ + PFN_vkCreateXcbSurfaceKHR vkCreateXcbSurfaceKHR = null; +}; + +version(Windows) +{ + PFN_vkCreateWin32SurfaceKHR vkCreateWin32SurfaceKHR = null; +}; + + +// Device Functions + +PFN_vkCreateSwapchainKHR vkCreateSwapchainKHR = null; +PFN_vkCreateImage vkCreateImage = null; +PFN_vkCreateImageView vkCreateImageView = null; +PFN_vkGetSwapchainImagesKHR vkGetSwapchainImagesKHR = null; +PFN_vkGetDeviceQueue vkGetDeviceQueue = null; +PFN_vkCreateSemaphore vkCreateSemaphore = null; +PFN_vkAllocateCommandBuffers vkAllocateCommandBuffers = null; +PFN_vkCreateCommandPool vkCreateCommandPool = null; +PFN_vkCreateFence vkCreateFence = null; +PFN_vkCreateDescriptorPool vkCreateDescriptorPool = null; +PFN_vkCreateDescriptorSetLayout vkCreateDescriptorSetLayout = null; +PFN_vkAllocateDescriptorSets vkAllocateDescriptorSets = null; +PFN_vkCreatePipelineLayout vkCreatePipelineLayout = null; +PFN_vkResetDescriptorPool vkResetDescriptorPool = null; +PFN_vkCreateShaderModule vkCreateShaderModule = null; +PFN_vkCreateGraphicsPipelines vkCreateGraphicsPipelines = null; +PFN_vkCreateComputePipelines vkCreateComputePipelines = null; +PFN_vkUpdateDescriptorSets vkUpdateDescriptorSets = null; +PFN_vkDestroyDevice vkDestroyDevice = null; +PFN_vkDestroyDescriptorPool vkDestroyDescriptorPool = null; +PFN_vkDestroySwapchainKHR vkDestroySwapchainKHR = null; +PFN_vkDestroyImage vkDestroyImage = null; +PFN_vkDestroyImageView vkDestroyImageView = null; +PFN_vkDestroyCommandPool vkDestroyCommandPool = null; +PFN_vkDestroySemaphore vkDestroySemaphore = null; +PFN_vkDestroyFence vkDestroyFence = null; +PFN_vkDestroyPipelineLayout vkDestroyPipelineLayout = null; +PFN_vkDestroyPipeline vkDestroyPipeline = null; +PFN_vkWaitForFences vkWaitForFences = null; +PFN_vkBeginCommandBuffer vkBeginCommandBuffer = null; +PFN_vkEndCommandBuffer vkEndCommandBuffer = null; +PFN_vkAcquireNextImageKHR vkAcquireNextImageKHR = null; +PFN_vkCmdBindPipeline vkCmdBindPipeline = null; +PFN_vkCmdBindDescriptorSets vkCmdBindDescriptorSets = null; +PFN_vkCmdDispatch vkCmdDispatch = null; +PFN_vkCmdBeginRendering vkCmdBeginRendering = null; +PFN_vkCmdEndRendering vkCmdEndRendering = null; +PFN_vkCmdSetViewport vkCmdSetViewport = null; +PFN_vkCmdSetScissor vkCmdSetScissor = null; +PFN_vkCmdPushConstants vkCmdPushConstants = null; +PFN_vkCmdBindIndexBuffer vkCmdBindIndexBuffer = null; +PFN_vkCmdBindVertexBuffers vkCmdBindVertexBuffers = null; +PFN_vkCmdDrawIndexed vkCmdDrawIndexed = null; +PFN_vkCmdBlitImage2 vkCmdBlitImage2 = null; +PFN_vkCmdPipelineBarrier2 vkCmdPipelineBarrier2 = null; +PFN_vkCmdCopyBufferToImage vkCmdCopyBufferToImage = null; +PFN_vkCmdCopyBuffer vkCmdCopyBuffer = null; +PFN_vkQueueSubmit2 vkQueueSubmit2 = null; +PFN_vkResetFences vkResetFences = null; +PFN_vkResetCommandBuffer vkResetCommandBuffer = null; +PFN_vkFreeCommandBuffers vkFreeCommandBuffers = null; +PFN_vkDestroyDescriptorSetLayout vkDestroyDescriptorSetLayout = null; +PFN_vkDestroyShaderModule vkDestroyShaderModule = null; +PFN_vkQueuePresentKHR vkQueuePresentKHR = null; +PFN_vkCmdDraw vkCmdDraw = null; +PFN_vkDeviceWaitIdle vkDeviceWaitIdle = null; +PFN_vkCmdClearColorImage vkCmdClearColorImage = null; +PFN_vkCreateSampler vkCreateSampler = null; +PFN_vkDestroySampler vkDestroySampler = null; +PFN_vkGetBufferDeviceAddress vkGetBufferDeviceAddress = null; +PFN_vkWaitSemaphores vkWaitSemaphores = null; +PFN_vkQueueWaitIdle vkQueueWaitIdle = null; + +bool +LoadGlobalFunctions() +{ + p.Library lib; + p.Function fn; + foreach(name; VULKAN_LIBS) + { + lib = p.LoadLibrary(name); + if (lib.ptr) + { + fn = p.LoadFunction(lib, "vkGetInstanceProcAddr"); + vkGetInstanceProcAddr = cast(PFN_vkGetInstanceProcAddr)fn.ptr; + } + } + + if (fn.ptr) + { + vkGetInstanceProcAddr = cast(PFN_vkGetInstanceProcAddr)vkGetInstanceProcAddr(null, "vkGetInstanceProcAddr"); + assert(vkGetInstanceProcAddr != null, "LoadGlobalFunctions failure: Unable to load vkGetInstanceProcAddr"); + + vkCreateInstance = cast(PFN_vkCreateInstance)vkGetInstanceProcAddr(null, "vkCreateInstance"); + assert(vkCreateInstance != null, "LoadGlobalFunctions failure: Unable to load VkCreateInstance"); + + vkEnumerateInstanceLayerProperties = cast(PFN_vkEnumerateInstanceLayerProperties)vkGetInstanceProcAddr(null, "vkEnumerateInstanceLayerProperties"); + assert(vkEnumerateInstanceLayerProperties != null, "LoadGlobalFunctions failure: Unable to load vkEnumerateInstanceLayerProperties"); + } + + return fn.ptr != null; +} + +void +LoadDeviceFunctions(Vulkan* vk) +{ + vkCreateSwapchainKHR = cast(PFN_vkCreateSwapchainKHR)vkGetDeviceProcAddr(vk.device, "vkCreateSwapchainKHR"); + vkCreateImage = cast(PFN_vkCreateImage)vkGetDeviceProcAddr(vk.device, "vkCreateImage"); + vkCreateImageView = cast(PFN_vkCreateImageView)vkGetDeviceProcAddr(vk.device, "vkCreateImageView"); + vkGetSwapchainImagesKHR = cast(PFN_vkGetSwapchainImagesKHR)vkGetDeviceProcAddr(vk.device, "vkGetSwapchainImagesKHR"); + vkGetDeviceQueue = cast(PFN_vkGetDeviceQueue)vkGetDeviceProcAddr(vk.device, "vkGetDeviceQueue"); + vkCreateSemaphore = cast(PFN_vkCreateSemaphore)vkGetDeviceProcAddr(vk.device, "vkCreateSemaphore"); + vkAllocateCommandBuffers = cast(PFN_vkAllocateCommandBuffers)vkGetDeviceProcAddr(vk.device, "vkAllocateCommandBuffers"); + vkCreateCommandPool = cast(PFN_vkCreateCommandPool)vkGetDeviceProcAddr(vk.device, "vkCreateCommandPool"); + vkCreateFence = cast(PFN_vkCreateFence)vkGetDeviceProcAddr(vk.device, "vkCreateFence"); + vkCreateDescriptorPool = cast(PFN_vkCreateDescriptorPool)vkGetDeviceProcAddr(vk.device, "vkCreateDescriptorPool"); + vkCreateDescriptorSetLayout = cast(PFN_vkCreateDescriptorSetLayout)vkGetDeviceProcAddr(vk.device, "vkCreateDescriptorSetLayout"); + vkAllocateDescriptorSets = cast(PFN_vkAllocateDescriptorSets)vkGetDeviceProcAddr(vk.device, "vkAllocateDescriptorSets"); + vkCreatePipelineLayout = cast(PFN_vkCreatePipelineLayout)vkGetDeviceProcAddr(vk.device, "vkCreatePipelineLayout"); + vkResetDescriptorPool = cast(PFN_vkResetDescriptorPool)vkGetDeviceProcAddr(vk.device, "vkResetDescriptorPool"); + vkCreateShaderModule = cast(PFN_vkCreateShaderModule)vkGetDeviceProcAddr(vk.device, "vkCreateShaderModule"); + vkCreateGraphicsPipelines = cast(PFN_vkCreateGraphicsPipelines)vkGetDeviceProcAddr(vk.device, "vkCreateGraphicsPipelines"); + vkCreateComputePipelines = cast(PFN_vkCreateComputePipelines)vkGetDeviceProcAddr(vk.device, "vkCreateComputePipelines"); + vkUpdateDescriptorSets = cast(PFN_vkUpdateDescriptorSets)vkGetDeviceProcAddr(vk.device, "vkUpdateDescriptorSets"); + vkDestroyDevice = cast(PFN_vkDestroyDevice)vkGetDeviceProcAddr(vk.device, "vkDestroyDevice"); + vkDestroyDescriptorPool = cast(PFN_vkDestroyDescriptorPool)vkGetDeviceProcAddr(vk.device, "vkDestroyDescriptorPool"); + vkDestroySwapchainKHR = cast(PFN_vkDestroySwapchainKHR)vkGetDeviceProcAddr(vk.device, "vkDestroySwapchainKHR"); + vkDestroyImage = cast(PFN_vkDestroyImage)vkGetDeviceProcAddr(vk.device, "vkDestroyImage"); + vkDestroyImageView = cast(PFN_vkDestroyImageView)vkGetDeviceProcAddr(vk.device, "vkDestroyImageView"); + vkDestroyCommandPool = cast(PFN_vkDestroyCommandPool)vkGetDeviceProcAddr(vk.device, "vkDestroyCommandPool"); + vkDestroySemaphore = cast(PFN_vkDestroySemaphore)vkGetDeviceProcAddr(vk.device, "vkDestroySemaphore"); + vkDestroyFence = cast(PFN_vkDestroyFence)vkGetDeviceProcAddr(vk.device, "vkDestroyFence"); + vkDestroyPipelineLayout = cast(PFN_vkDestroyPipelineLayout)vkGetDeviceProcAddr(vk.device, "vkDestroyPipelineLayout"); + vkDestroyPipeline = cast(PFN_vkDestroyPipeline)vkGetDeviceProcAddr(vk.device, "vkDestroyPipeline"); + vkWaitForFences = cast(PFN_vkWaitForFences)vkGetDeviceProcAddr(vk.device, "vkWaitForFences"); + vkBeginCommandBuffer = cast(PFN_vkBeginCommandBuffer)vkGetDeviceProcAddr(vk.device, "vkBeginCommandBuffer"); + vkEndCommandBuffer = cast(PFN_vkEndCommandBuffer)vkGetDeviceProcAddr(vk.device, "vkEndCommandBuffer"); + vkAcquireNextImageKHR = cast(PFN_vkAcquireNextImageKHR)vkGetDeviceProcAddr(vk.device, "vkAcquireNextImageKHR"); + vkCmdBindPipeline = cast(PFN_vkCmdBindPipeline)vkGetDeviceProcAddr(vk.device, "vkCmdBindPipeline"); + vkCmdBindDescriptorSets = cast(PFN_vkCmdBindDescriptorSets)vkGetDeviceProcAddr(vk.device, "vkCmdBindDescriptorSets"); + vkCmdDispatch = cast(PFN_vkCmdDispatch)vkGetDeviceProcAddr(vk.device, "vkCmdDispatch"); + vkCmdBeginRendering = cast(PFN_vkCmdBeginRendering)vkGetDeviceProcAddr(vk.device, "vkCmdBeginRendering"); + vkCmdEndRendering = cast(PFN_vkCmdEndRendering)vkGetDeviceProcAddr(vk.device, "vkCmdEndRendering"); + vkCmdSetViewport = cast(PFN_vkCmdSetViewport)vkGetDeviceProcAddr(vk.device, "vkCmdSetViewport"); + vkCmdSetScissor = cast(PFN_vkCmdSetScissor)vkGetDeviceProcAddr(vk.device, "vkCmdSetScissor"); + vkCmdPushConstants = cast(PFN_vkCmdPushConstants)vkGetDeviceProcAddr(vk.device, "vkCmdPushConstants"); + vkCmdBindIndexBuffer = cast(PFN_vkCmdBindIndexBuffer)vkGetDeviceProcAddr(vk.device, "vkCmdBindIndexBuffer"); + vkCmdBindVertexBuffers = cast(PFN_vkCmdBindVertexBuffers)vkGetDeviceProcAddr(vk.device, "vkCmdBindVertexBuffers"); + vkCmdDrawIndexed = cast(PFN_vkCmdDrawIndexed)vkGetDeviceProcAddr(vk.device, "vkCmdDrawIndexed"); + vkCmdBlitImage2 = cast(PFN_vkCmdBlitImage2)vkGetDeviceProcAddr(vk.device, "vkCmdBlitImage2"); + vkCmdPipelineBarrier2 = cast(PFN_vkCmdPipelineBarrier2)vkGetDeviceProcAddr(vk.device, "vkCmdPipelineBarrier2"); + vkCmdCopyBufferToImage = cast(PFN_vkCmdCopyBufferToImage)vkGetDeviceProcAddr(vk.device, "vkCmdCopyBufferToImage"); + vkCmdCopyBuffer = cast(PFN_vkCmdCopyBuffer)vkGetDeviceProcAddr(vk.device, "vkCmdCopyBuffer"); + vkQueueSubmit2 = cast(PFN_vkQueueSubmit2)vkGetDeviceProcAddr(vk.device, "vkQueueSubmit2"); + vkResetFences = cast(PFN_vkResetFences)vkGetDeviceProcAddr(vk.device, "vkResetFences"); + vkResetCommandBuffer = cast(PFN_vkResetCommandBuffer)vkGetDeviceProcAddr(vk.device, "vkResetCommandBuffer"); + vkFreeCommandBuffers = cast(PFN_vkFreeCommandBuffers)vkGetDeviceProcAddr(vk.device, "vkFreeCommandBuffers"); + vkDestroyDescriptorSetLayout = cast(PFN_vkDestroyDescriptorSetLayout)vkGetDeviceProcAddr(vk.device, "vkDestroyDescriptorSetLayout"); + vkDestroyShaderModule = cast(PFN_vkDestroyShaderModule)vkGetDeviceProcAddr(vk.device, "vkDestroyShaderModule"); + vkQueuePresentKHR = cast(PFN_vkQueuePresentKHR)vkGetDeviceProcAddr(vk.device, "vkQueuePresentKHR"); + vkCmdDraw = cast(PFN_vkCmdDraw)vkGetDeviceProcAddr(vk.device, "vkCmdDraw"); + vkDeviceWaitIdle = cast(PFN_vkDeviceWaitIdle)vkGetDeviceProcAddr(vk.device, "vkDeviceWaitIdle"); + vkCmdClearColorImage = cast(PFN_vkCmdClearColorImage)vkGetDeviceProcAddr(vk.device, "vkCmdClearColorImage"); + vkCreateSampler = cast(PFN_vkCreateSampler)vkGetDeviceProcAddr(vk.device, "vkCreateSampler"); + vkDestroySampler = cast(PFN_vkDestroySampler)vkGetDeviceProcAddr(vk.device, "vkDestroySampler"); + vkGetBufferDeviceAddress = cast(PFN_vkGetBufferDeviceAddress)vkGetDeviceProcAddr(vk.device, "vkGetBufferDeviceAddress"); + vkWaitSemaphores = cast(PFN_vkWaitSemaphores)vkGetDeviceProcAddr(vk.device, "vkWaitSemaphores"); + vkQueueWaitIdle = cast(PFN_vkQueueWaitIdle)vkGetDeviceProcAddr(vk.device, "vkQueueWaitIdle"); + + assert(vkCreateSwapchainKHR != null, "LoadDeviceFunctions failure: function pointer is null"); +} + +void +LoadInstanceFunctions(Vulkan* vk) +{ + vkEnumeratePhysicalDevices = cast(PFN_vkEnumeratePhysicalDevices)vkGetInstanceProcAddr(vk.instance, "vkEnumeratePhysicalDevices"); + vkDestroySurfaceKHR = cast(PFN_vkDestroySurfaceKHR)vkGetInstanceProcAddr(vk.instance, "vkDestroySurfaceKHR"); + vkDestroyInstance = cast(PFN_vkDestroyInstance)vkGetInstanceProcAddr(vk.instance, "vkDestroyInstance"); + vkCreateDevice = cast(PFN_vkCreateDevice)vkGetInstanceProcAddr(vk.instance, "vkCreateDevice"); + vkGetPhysicalDeviceQueueFamilyProperties = cast(PFN_vkGetPhysicalDeviceQueueFamilyProperties)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceQueueFamilyProperties"); + vkGetPhysicalDeviceSurfaceSupportKHR = cast(PFN_vkGetPhysicalDeviceSurfaceSupportKHR)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceSurfaceSupportKHR"); + vkGetPhysicalDeviceProperties = cast(PFN_vkGetPhysicalDeviceProperties)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceProperties"); + vkGetPhysicalDeviceFeatures2 = cast(PFN_vkGetPhysicalDeviceFeatures2)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceFeatures2"); + vkEnumerateDeviceExtensionProperties = cast(PFN_vkEnumerateDeviceExtensionProperties)vkGetInstanceProcAddr(vk.instance, "vkEnumerateDeviceExtensionProperties"); + vkGetPhysicalDeviceSurfacePresentModesKHR = cast(PFN_vkGetPhysicalDeviceSurfacePresentModesKHR)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceSurfacePresentModesKHR"); + vkGetPhysicalDeviceSurfaceFormatsKHR = cast(PFN_vkGetPhysicalDeviceSurfaceFormatsKHR)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceSurfaceFormatsKHR"); + vkGetPhysicalDeviceSurfaceCapabilitiesKHR = cast(PFN_vkGetPhysicalDeviceSurfaceCapabilitiesKHR)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceSurfaceCapabilitiesKHR"); + vkGetPhysicalDeviceImageFormatProperties = cast(PFN_vkGetPhysicalDeviceImageFormatProperties)vkGetInstanceProcAddr(vk.instance, "vkGetPhysicalDeviceImageFormatProperties"); + vkGetDeviceProcAddr = cast(PFN_vkGetDeviceProcAddr)vkGetInstanceProcAddr(vk.instance, "vkGetDeviceProcAddr"); + +version(linux) +{ + vkCreateXcbSurfaceKHR = cast(PFN_vkCreateXcbSurfaceKHR)vkGetInstanceProcAddr(vk.instance, "vkCreateXcbSurfaceKHR"); +} + +version(Windows) +{ + vkCreateWin32SurfaceKHR = cast(PFN_vkCreateWin32SurfaceKHR)vkGetInstanceProcAddr(vk.instance, "vkCreateWin32SurfaceKHR"); +} + +debug +{ + vkCreateDebugUtilsMessengerEXT = cast(PFN_vkCreateDebugUtilsMessengerEXT)vkGetInstanceProcAddr(vk.instance, "vkCreateDebugUtilsMessengerEXT"); + vkDestroyDebugUtilsMessengerEXT = cast(PFN_vkDestroyDebugUtilsMessengerEXT)vkGetInstanceProcAddr(vk.instance, "vkDestroyDebugUtilsMessengerEXT"); +} + + assert(vkEnumeratePhysicalDevices != null, "LoadInstanceFunctions failure: failed to load function"); +} diff --git a/src/vulkan_logging.d b/src/vulkan_logging.d new file mode 100644 index 0000000..efd5525 --- /dev/null +++ b/src/vulkan_logging.d @@ -0,0 +1,153 @@ +public import includes; +import std.stdio; +import vulkan : Vulkan; + +VkBool32 +DebugCallback( + VkDebugUtilsMessageSeverityFlagBitsEXT message_severity, + VkDebugUtilsMessageTypeFlagsEXT message_type, + const VkDebugUtilsMessengerCallbackDataEXT* callback_data, + void* user_data +) +{ + string ms, mt; + + switch (message_severity) { + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_VERBOSE_BIT_EXT: + ms = "VERBOSE"; + break; + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_INFO_BIT_EXT: + ms = "INFO"; + break; + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT: + ms = "WARNING"; + break; + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT: + ms = "ERROR"; + break; + default: + ms = "UNKNOWN"; + break; + } + + switch (message_type) { + case VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT: + mt = "General"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT: + mt = "Validation"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT: + mt = "Validation | General"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT: + mt = "Performance"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT: + mt = "General | Performance"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT: + mt = "Validation | Performance"; + break; + case VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT: + mt = "General | Validation | Performance"; + break; + default: + mt = "Unknown"; + break; + } + + writefln("[%s: %s]\n%s\n", ms, mt, callback_data.pMessage); + + return VK_FALSE; +} + +string +VkResultStr(VkResult result) +{ + switch (result) + { + case VK_SUCCESS: + return "VK_SUCCESS"; + case VK_NOT_READY: + return "VK_NOT_READY"; + case VK_TIMEOUT: + return "VK_TIMEOUT"; + case VK_EVENT_SET: + return "VK_EVENT_SET"; + case VK_EVENT_RESET: + return "VK_EVENT_RESET"; + case VK_INCOMPLETE: + return "VK_INCOMPLETE"; + case VK_ERROR_OUT_OF_HOST_MEMORY: + return "VK_ERROR_OUT_OF_HOST_MEMORY"; + case VK_ERROR_OUT_OF_DEVICE_MEMORY: + return "VK_ERROR_OUT_OF_DEVICE_MEMORY"; + case VK_ERROR_INITIALIZATION_FAILED: + return "VK_ERROR_INITIALIZATION_FAILED"; + case VK_ERROR_DEVICE_LOST: + return "VK_ERROR_DEVICE_LOST"; + case VK_ERROR_MEMORY_MAP_FAILED: + return "VK_ERROR_MEMORY_MAP_FAILED"; + case VK_ERROR_LAYER_NOT_PRESENT: + return "VK_ERROR_LAYER_NOT_PRESENT"; + case VK_ERROR_EXTENSION_NOT_PRESENT: + return "VK_ERROR_EXTENSION_NOT_PRESENT"; + case VK_ERROR_FEATURE_NOT_PRESENT: + return "VK_ERROR_FEATURE_NOT_PRESENT"; + case VK_ERROR_INCOMPATIBLE_DRIVER: + return "VK_ERROR_INCOMPATIBLE_DRIVER"; + case VK_ERROR_TOO_MANY_OBJECTS: + return "VK_ERROR_TOO_MANY_OBJECTS"; + case VK_ERROR_FORMAT_NOT_SUPPORTED: + return "VK_ERROR_FORMAT_NOT_SUPPORTED"; + case VK_ERROR_FRAGMENTED_POOL: + return "VK_ERROR_FRAGMENTED_POOL"; + case VK_ERROR_UNKNOWN: + return "VK_ERROR_UNKNOWN"; + case VK_ERROR_OUT_OF_POOL_MEMORY: + return "VK_ERROR_OUT_OF_POOL_MEMORY"; + case VK_ERROR_INVALID_EXTERNAL_HANDLE: + return "VK_ERROR_INVALID_EXTERNAL_HANDLE"; + case VK_ERROR_FRAGMENTATION: + return "VK_ERROR_FRAGMENTATION"; + case VK_ERROR_INVALID_OPAQUE_CAPTURE_ADDRESS: + return "VK_ERROR_INVALID_OPAQUE_CAPTURE_ADDRESS"; + case VK_PIPELINE_COMPILE_REQUIRED: + return "VK_PIPELINE_COMPILE_REQUIRED"; + case VK_ERROR_SURFACE_LOST_KHR: + return "VK_ERROR_SURFACE_LOST_KHR"; + case VK_ERROR_NATIVE_WINDOW_IN_USE_KHR: + return "VK_ERROR_NATIVE_WINDOW_IN_USE_KHR"; + case VK_SUBOPTIMAL_KHR: + return "VK_SUBOPTIMAL_KHR"; + case VK_ERROR_OUT_OF_DATE_KHR: + return "VK_ERROR_OUT_OF_DATE_KHR"; + case VK_ERROR_INCOMPATIBLE_DISPLAY_KHR: + return "VK_ERROR_INCOMPATIBLE_DISPLAY_KHR"; + case VK_ERROR_VALIDATION_FAILED_EXT: + return "VK_ERROR_VALIDATION_FAILED_EXT"; + case VK_ERROR_INVALID_SHADER_NV: + return "VK_ERROR_INVALID_SHADER_NV"; + case VK_ERROR_INVALID_DRM_FORMAT_MODIFIER_PLANE_LAYOUT_EXT: + return "VK_ERROR_INVALID_DRM_FORMAT_MODIFIER_PLANE_LAYOUT_EXT"; + case VK_ERROR_NOT_PERMITTED_KHR: + return "VK_ERROR_NOT_PERMITTED_KHR"; + case VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT: + return "VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT"; + case VK_THREAD_IDLE_KHR: + return "VK_THREAD_IDLE_KHR"; + case VK_THREAD_DONE_KHR: + return "VK_THREAD_DONE_KHR"; + case VK_OPERATION_DEFERRED_KHR: + return "VK_OPERATION_DEFERRED_KHR"; + case VK_OPERATION_NOT_DEFERRED_KHR: + return "VK_OPERATION_NOT_DEFERRED_KHR"; + case VK_ERROR_COMPRESSION_EXHAUSTED_EXT: + return "VK_ERROR_COMPRESSION_EXHAUSTED_EXT"; + case VK_RESULT_MAX_ENUM: + return "VK_RESULT_MAX_ENUM"; + default: + return "UNKNOWN"; + } +}