#ifdef INTELLISENSE_DIRECTIVES # pragma once # include "context_cracking.h" # include "linkage.h" # include "macros.h" # include "platform.h" # include "base_types.h" # include "memory.h" #endif // This provides an alterntive memory strategy to HMH/Casey Muratori/RJF styled arenas // The library is derived from zpl-c which in-turn // is related to the gb headers an thus the Odin-lang memory strategy // Users can override the underlying memory allocator used, even for the HMH arena memory strategy. #ifndef MD__ONES #define MD__ONES ( md_scast( GEN_NS usize, - 1) / MD_U8_MAX ) #define MD__HIGHS ( MD__ONES * ( MD_U8_MAX / 2 + 1 ) ) #define MD__HAS_ZERO( x ) ( ( ( x ) - MD__ONES ) & ~( x ) & MD__HIGHS ) #endif // Return value of md_allocator_type typedef MD_U64 MD_AllocatorType; enum { MD_AllocatorType_Heap = 0, // Genreal md_heap allocator MD_AllocatorType_VArena = 1, // MD_Arena allocator backed by virtual address space MD_AllocatorType_FArena = 2, // Fixed arena backed back by a fixed size block of memory (usually a byte slice) MD_AllocatorType_Arena = 3, // Composite arena used originally by RAD Debugger & Metadesk }; typedef MD_U32 MD_AllocatorMode; enum MD_AllocatorMode { MD_AllocatorMode_Alloc, MD_AllocatorMode_Free, MD_AllocatorMode_FreeAll, MD_AllocatorMode_Resize, MD_AllocatorMode_QueryType, MD_AllocatorMode_QuerySupport, }; typedef MD_U64 MD_AllocatorQueryFlags; enum { MD_AllocatorQuery_Alloc = (1 << 0), MD_AllocatorQuery_Free = (1 << 1), MD_AllocatorQuery_FreeAll = (1 << 2), MD_AllocatorQuery_Resize = (1 << 3), // Supports both grow and shrink MD_AllocatorQuery_ResizeShrink = (1 << 4), MD_AllocatorQuery_ResizeGrow = (1 << 5), }; typedef void*(MD_AllocatorProc)( void* allocator_data, MD_AllocatorMode type, MD_SSIZE size, MD_SSIZE alignment, void* old_memory, MD_SSIZE old_size, MD_U64 flags ); typedef struct MD_AllocatorInfo MD_AllocatorInfo; struct MD_AllocatorInfo { MD_AllocatorProc* proc; void* data; }; // Overridable by the user by defining MD_OVERRIDE_DEFAULT_ALLOCATOR MD_AllocatorInfo md_default_allocator(); enum AllocFlag { MD_ALLOCATOR_FLAG_CLEAR_TO_ZERO = (1 << 0), }; #ifndef MD_DEFAULT_MEMORY_ALIGNMENT # define MD_DEFAULT_MEMORY_ALIGNMENT ( 2 * size_of( void* ) ) #endif #ifndef MD_DEFAULT_ALLOCATOR_FLAGS # define MD_DEFAULT_ALLOCATOR_FLAGS ( MD_ALLOCATOR_FLAG_CLEAR_TO_ZERO ) #endif // Retrieve which type of allocator MD_AllocatorType md_allocator_type(MD_AllocatorInfo a); // Retreive which modes the allocator supports MD_AllocatorQueryFlags md_allocator_query_support(MD_AllocatorInfo a); // Allocate memory with default alignment. void* md_alloc( MD_AllocatorInfo a, MD_SSIZE size ); // Allocate memory with specified alignment. void* md_alloc_align( MD_AllocatorInfo a, MD_SSIZE size, MD_SSIZE alignment ); // Free allocated memory. void md_alloc_free( MD_AllocatorInfo a, void* ptr ); // Free all memory allocated by an allocator. void md_free_all( MD_AllocatorInfo a ); // Resize an allocated memory. void* md_resize( MD_AllocatorInfo a, void* ptr, MD_SSIZE old_size, MD_SSIZE new_size ); // Resize an allocated memory with specified alignment. void* md_resize_align( MD_AllocatorInfo a, void* ptr, MD_SSIZE old_size, MD_SSIZE new_size, MD_SSIZE alignment ); #ifndef md_alloc_item // Allocate memory for an item. #define md_alloc_item(allocator, Type) (Type*)md_memory_zero(md_alloc(allocator, size_of(Type)), size_of(Type)) // Allocate memory for an item. #define md_alloc_item_no_zero( allocator, Type ) (Type*) md_alloc(allocator, size_of(Type)) #endif #ifndef md_alloc_array // Allocate memory for an array of items. #define md_alloc_array( allocator_, Type, count ) (Type*)md_memory_zero(md_alloc( allocator_, size_of(Type) * (count) ), size_of(Type) * (count)) // Allocate memory for an array of items. (Don't zero initialize) #define md_alloc_array_no_zero( allocator_, Type, count ) (Type*) md_alloc( allocator_, size_of(Type) * (count) ) #endif // Allocate/Resize memory using default options. // Use this if you don't need a "fancy" md_resize allocation void* md_default_resize_align( MD_AllocatorInfo a, void* ptr, MD_SSIZE old_size, MD_SSIZE new_size, MD_SSIZE alignment ); #ifdef MD_HEAP_ANALYSIS /* md_heap memory analysis tools */ /* define GEN_HEAP_ANALYSIS to enable this feature */ /* call zpl_heap_stats_init at the beginning of the entry point */ /* you can call zpl_heap_stats_check near the end of the execution to validate any possible leaks */ MD_API void md_heap_stats_init( void ); MD_API MD_SSIZE heap_stats_used_memory( void ); MD_API MD_SSIZE md_heap_stats_alloc_count( void ); MD_API void md_heap_stats_check( void ); #endif MD_API void* md_heap_allocator_proc( void* allocator_data, MD_AllocatorMode mode, MD_SSIZE size, MD_SSIZE alignment, void* old_memory, MD_SSIZE old_size, MD_U64 flags ); #ifndef md_heap // The md_heap allocator backed by the platform vendor's md_malloc & free. #define md_heap() (MD_AllocatorInfo){ md_heap_allocator_proc, md_nullptr } #endif /* Virtual Memory MD_Arena This is separate from the composite arena used by HMH/Casey Muratori/RJF This arena stricly manages one reservation of the process's virtual address space. Segregating this from composite MD_Arena style causes more moremoy to be used for "allocator headers", however it allows users of a library to have greater control over the allocation strategy used from their side instead of the library itself. Like with the composite MD_Arena, the MD_VArena has its struct as the header of the reserve of memory. */ #ifndef MD_VARENA_DEFUALT_RESERVE #define MD_VARENA_DEFAULT_RESERVE MD_MB(64) #endif #ifndef MD_VARENA_DEFUALT_COMMIT #define MD_VARENA_DEFAULT_COMMIT MD_KB(64) #endif typedef MD_U32 MD_VArenaFlags; enum { MD_VArenaFlag_LargePages = (1 << 0), }; typedef struct MD_VArenaParams MD_VArenaParams; struct MD_VArenaParams { MD_U64 base_addr; MD_VArenaFlags flags; MD_U64 reserve_size; MD_U64 commit_size; }; typedef struct MD_VArena MD_VArena; struct MD_VArena { MD_SSIZE reserve_start; MD_SSIZE reserve; MD_SSIZE commit_size; MD_SSIZE committed; MD_SSIZE commit_used; MD_VArenaFlags flags; }; MD_API MD_VArena* md_varena__alloc(MD_VArenaParams params MD_PARAM_DEFAULT); #define md_varena_alloc(...) md_varena__alloc( (MD_VArenaParams){__VA_ARGS__} ) MD_API void md_varena_commit (MD_VArena* vm, MD_SSIZE commit_size); MD_API void md_varena_release(MD_VArena* vm); md_force_inline void varena_rewind(MD_VArena* vm, MD_SSIZE pos) { vm->commit_used = pos; } MD_API void* md_varena_allocator_proc(void* allocator_data, MD_AllocatorMode mode, MD_SSIZE size, MD_SSIZE alignment, void* old_memory, MD_SSIZE old_size, MD_U64 flags); #define md_varena_allocator(vm) (MD_AllocatorInfo) { md_varena_allocator_proc, vm } typedef struct MD_ByteSlice MD_ByteSlice; struct MD_ByteSlice { MD_U8* data; MD_SSIZE len; }; #define md_mem_to_byteslice(data, len) (MD_ByteSlice){ (MD_U8*)(data), (MD_SSIZE)(len) } // Fixed size arena typedef struct MD_FArena MD_FArena; struct MD_FArena { MD_ByteSlice slice; MD_SSIZE used; }; #define md_farena_from_byteslice(slice) (MD_FArena) { slice, 0 } #define md_farena_from_memory(data, len) (MD_FArena) { md_mem_to_byteslice(data, len), 0 } MD_API void* farena_allocator_proc(void* allocator_data, MD_AllocatorMode mode, MD_SSIZE size, MD_SSIZE alignment, void* old_memory, MD_SSIZE old_size, MD_U64 flags); #define farena_allocator(arena) (MD_AllocatorInfo){ farena_allocator_proc, & arena } // Inlines inline MD_AllocatorType md_allocator_type(MD_AllocatorInfo a) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return (MD_AllocatorType) a.proc(a.data, MD_AllocatorMode_QueryType, 0, 0, md_nullptr, 0, MD_DEFAULT_ALLOCATOR_FLAGS); } inline MD_AllocatorQueryFlags md_allocator_query_support(MD_AllocatorInfo a) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return (MD_AllocatorType) a.proc(a.data, MD_AllocatorMode_QuerySupport, 0, 0, md_nullptr, 0, MD_DEFAULT_ALLOCATOR_FLAGS); } inline void* md_alloc_align( MD_AllocatorInfo a, MD_SSIZE size, MD_SSIZE alignment ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return a.proc( a.data, MD_AllocatorMode_Alloc, size, alignment, md_nullptr, 0, MD_DEFAULT_ALLOCATOR_FLAGS ); } inline void* md_alloc( MD_AllocatorInfo a, MD_SSIZE size ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return md_alloc_align( a, size, MD_DEFAULT_MEMORY_ALIGNMENT ); } inline void md_alloc_free( MD_AllocatorInfo a, void* ptr ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } if ( ptr != md_nullptr ) { a.proc( a.data, MD_AllocatorMode_Free, 0, 0, ptr, 0, MD_DEFAULT_ALLOCATOR_FLAGS ); } } inline void md_free_all( MD_AllocatorInfo a ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } a.proc( a.data, MD_AllocatorMode_FreeAll, 0, 0, md_nullptr, 0, MD_DEFAULT_ALLOCATOR_FLAGS ); } inline void* md_resize( MD_AllocatorInfo a, void* ptr, MD_SSIZE old_size, MD_SSIZE new_size ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return md_resize_align( a, ptr, old_size, new_size, MD_DEFAULT_ALLOCATOR_FLAGS ); } inline void* md_resize_align( MD_AllocatorInfo a, void* ptr, MD_SSIZE old_size, MD_SSIZE new_size, MD_SSIZE alignment ) { if (a.proc == md_nullptr) { a = md_default_allocator(); } return a.proc( a.data, MD_AllocatorMode_Resize, new_size, alignment, ptr, old_size, MD_DEFAULT_ALLOCATOR_FLAGS ); }