Odin/src/common_memory.cpp

#if defined(GB_SYSTEM_LINUX)
#include <malloc.h>
#endif

template <typename U, typename V>
gb_internal gb_inline U bit_cast(V &v) { return reinterpret_cast<U &>(v); }

template <typename U, typename V>
gb_internal gb_inline U const &bit_cast(V const &v) { return reinterpret_cast<U const &>(v); }


gb_internal gb_inline i64 align_formula(i64 size, i64 align) {
	i64 result = size + align-1;
	return result - (i64)((u64)result%(u64)align);
}
gb_internal gb_inline isize align_formula_isize(isize size, isize align) {
	isize result = size + align-1;
	return result - (isize)((usize)result%(usize)align);
}
gb_internal gb_inline void *align_formula_ptr(void *ptr, isize align) {
	uintptr result = (cast(uintptr)ptr) + align-1;
	return (void *)(result - result%align);
}


gb_global BlockingMutex global_memory_block_mutex;

gb_internal void platform_virtual_memory_init(void);

gb_internal void virtual_memory_init(void) {
	platform_virtual_memory_init();
}


gb_internal Thread *get_current_thread(void);


struct MemoryBlock {
	MemoryBlock *prev;
	u8 *         base; 
	isize        size;
	isize        used;
};

struct Arena {
	MemoryBlock * curr_block;
	isize         minimum_block_size;
	// BlockingMutex mutex;
	isize         temp_count;
	Thread *      parent_thread;
};

enum { DEFAULT_MINIMUM_BLOCK_SIZE = 8ll*1024ll*1024ll };

gb_global isize DEFAULT_PAGE_SIZE = 4096;

gb_internal MemoryBlock *virtual_memory_alloc(isize size);
gb_internal void virtual_memory_dealloc(MemoryBlock *block);
gb_internal void *arena_alloc(Arena *arena, isize min_size, isize alignment);
gb_internal void arena_free_all(Arena *arena);

gb_internal isize arena_align_forward_offset(Arena *arena, isize alignment) {
	isize alignment_offset = 0;
	isize ptr = cast(isize)(arena->curr_block->base + arena->curr_block->used);
	isize mask = alignment-1;
	if (ptr & mask) {
		alignment_offset = alignment - (ptr & mask);
	}
	return alignment_offset;
}

gb_internal void thread_init_arenas(Thread *t) {
	t->permanent_arena = gb_alloc_item(heap_allocator(), Arena);
	t->temporary_arena = gb_alloc_item(heap_allocator(), Arena);

	t->permanent_arena->parent_thread = t;
	t->temporary_arena->parent_thread = t;

	t->permanent_arena->minimum_block_size = DEFAULT_MINIMUM_BLOCK_SIZE;
	t->temporary_arena->minimum_block_size = DEFAULT_MINIMUM_BLOCK_SIZE;
}

gb_internal void *arena_alloc(Arena *arena, isize min_size, isize alignment) {
	GB_ASSERT(gb_is_power_of_two(alignment));
	GB_ASSERT(arena->parent_thread == get_current_thread());

	isize size = 0;
	if (arena->curr_block != nullptr) {
		size = min_size + arena_align_forward_offset(arena, alignment);
	}

	if (arena->curr_block == nullptr || (arena->curr_block->used + size) > arena->curr_block->size) {
		size = align_formula_isize(min_size, alignment);
		arena->minimum_block_size = gb_max(DEFAULT_MINIMUM_BLOCK_SIZE, arena->minimum_block_size);
		
		isize block_size = gb_max(size, arena->minimum_block_size);
		
		MemoryBlock *new_block = virtual_memory_alloc(block_size);
		new_block->prev = arena->curr_block;
		arena->curr_block = new_block;
	}
	
	MemoryBlock *curr_block = arena->curr_block;
	GB_ASSERT((curr_block->used + size) <= curr_block->size);
	
	u8 *ptr = curr_block->base + curr_block->used;
	ptr += arena_align_forward_offset(arena, alignment);
	
	curr_block->used += size;
	GB_ASSERT(curr_block->used <= curr_block->size);

	// NOTE(bill): memory will be zeroed by default due to virtual memory 
	return ptr;	
}

gb_internal void arena_free_all(Arena *arena) {
	while (arena->curr_block != nullptr) {
		MemoryBlock *free_block = arena->curr_block;
		arena->curr_block = free_block->prev;
		virtual_memory_dealloc(free_block);
	}
}


struct PlatformMemoryBlock {
	MemoryBlock block; // IMPORTANT NOTE: must be at the start
	isize total_size;
	PlatformMemoryBlock *prev, *next;
};

gb_global std::atomic<isize> global_platform_memory_total_usage;
gb_global PlatformMemoryBlock global_platform_memory_block_sentinel;

gb_internal PlatformMemoryBlock *platform_virtual_memory_alloc(isize total_size);
gb_internal void platform_virtual_memory_free(PlatformMemoryBlock *block);
gb_internal void platform_virtual_memory_protect(void *memory, isize size);

#if defined(GB_SYSTEM_WINDOWS)
	gb_internal void platform_virtual_memory_init(void) {
		global_platform_memory_block_sentinel.prev = &global_platform_memory_block_sentinel;	
		global_platform_memory_block_sentinel.next = &global_platform_memory_block_sentinel;
		
		SYSTEM_INFO sys_info = {};
		GetSystemInfo(&sys_info);
		DEFAULT_PAGE_SIZE = gb_max(DEFAULT_PAGE_SIZE, cast(isize)sys_info.dwPageSize);
		GB_ASSERT(gb_is_power_of_two(DEFAULT_PAGE_SIZE));
	}

	gb_internal PlatformMemoryBlock *platform_virtual_memory_alloc(isize total_size) {
		PlatformMemoryBlock *pmblock = (PlatformMemoryBlock *)VirtualAlloc(0, total_size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
		if (pmblock == nullptr) {
			gb_printf_err("Out of Virtual memory, oh no...\n");
			gb_printf_err("Requested: %lld bytes\n", cast(long long)total_size);
			gb_printf_err("Total Usage: %lld bytes\n", cast(long long)global_platform_memory_total_usage);
			GB_ASSERT_MSG(pmblock != nullptr, "Out of Virtual Memory, oh no...");
		}
		global_platform_memory_total_usage.fetch_add(total_size);
		return pmblock;
	}
	gb_internal void platform_virtual_memory_free(PlatformMemoryBlock *block) {
		global_platform_memory_total_usage.fetch_sub(block->total_size);
		GB_ASSERT(VirtualFree(block, 0, MEM_RELEASE));
	}
	gb_internal void platform_virtual_memory_protect(void *memory, isize size) {
		DWORD old_protect = 0;
		BOOL is_protected = VirtualProtect(memory, size, PAGE_NOACCESS, &old_protect);
		GB_ASSERT(is_protected);
	}
#else
	#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
	#define MAP_ANONYMOUS MAP_ANON
	#endif

	gb_internal void platform_virtual_memory_init(void) {
		global_platform_memory_block_sentinel.prev = &global_platform_memory_block_sentinel;	
		global_platform_memory_block_sentinel.next = &global_platform_memory_block_sentinel;
		
		DEFAULT_PAGE_SIZE = gb_max(DEFAULT_PAGE_SIZE, cast(isize)sysconf(_SC_PAGE_SIZE));
		GB_ASSERT(gb_is_power_of_two(DEFAULT_PAGE_SIZE));
	}
	
	gb_internal PlatformMemoryBlock *platform_virtual_memory_alloc(isize total_size) {
		PlatformMemoryBlock *pmblock = (PlatformMemoryBlock *)mmap(nullptr, total_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
		if (pmblock == nullptr) {
			gb_printf_err("Out of Virtual memory, oh no...\n");
			gb_printf_err("Requested: %lld bytes\n", cast(long long)total_size);
			gb_printf_err("Total Usage: %lld bytes\n", cast(long long)global_platform_memory_total_usage);
			GB_ASSERT_MSG(pmblock != nullptr, "Out of Virtual Memory, oh no...");
		}
		global_platform_memory_total_usage.fetch_add(total_size);
		return pmblock;
	}
	gb_internal void platform_virtual_memory_free(PlatformMemoryBlock *block) {
		isize size = block->total_size;
		global_platform_memory_total_usage.fetch_sub(size);
		munmap(block, size);
	}
	gb_internal void platform_virtual_memory_protect(void *memory, isize size) {
		int err = mprotect(memory, size, PROT_NONE);
		GB_ASSERT(err == 0);
	}
#endif

gb_internal MemoryBlock *virtual_memory_alloc(isize size) {
	isize const page_size = DEFAULT_PAGE_SIZE; 
	
	isize total_size     = size + gb_size_of(PlatformMemoryBlock);
	isize base_offset    = gb_size_of(PlatformMemoryBlock);
	isize protect_offset = 0;
	
	bool do_protection = false;
	{ // overflow protection
		isize rounded_size = align_formula_isize(size, page_size);
		total_size     = rounded_size + 2*page_size;
		base_offset    = page_size + rounded_size - size;
		protect_offset = page_size + rounded_size;
		do_protection  = true;
	}
	
	PlatformMemoryBlock *pmblock = platform_virtual_memory_alloc(total_size);
	GB_ASSERT_MSG(pmblock != nullptr, "Out of Virtual Memory, oh no...");
	
	pmblock->block.base = cast(u8 *)pmblock + base_offset;
	// Should be zeroed
	GB_ASSERT(pmblock->block.used == 0);
	GB_ASSERT(pmblock->block.prev == nullptr);
	
	if (do_protection) {
		platform_virtual_memory_protect(cast(u8 *)pmblock + protect_offset, page_size);
	}
	
	pmblock->block.size = size;
	pmblock->total_size = total_size;

	PlatformMemoryBlock *sentinel = &global_platform_memory_block_sentinel;
	mutex_lock(&global_memory_block_mutex);
	pmblock->next = sentinel;
	pmblock->prev = sentinel->prev;
	pmblock->prev->next = pmblock;
	pmblock->next->prev = pmblock;
	mutex_unlock(&global_memory_block_mutex);
	
	return &pmblock->block;
}

gb_internal void virtual_memory_dealloc(MemoryBlock *block_to_free) {
	PlatformMemoryBlock *block = cast(PlatformMemoryBlock *)block_to_free;
	if (block != nullptr) {
		mutex_lock(&global_memory_block_mutex);
		block->prev->next = block->next;
		block->next->prev = block->prev;
		mutex_unlock(&global_memory_block_mutex);
			
		platform_virtual_memory_free(block);
	}
}

struct ArenaTemp {
	Arena *      arena;
	MemoryBlock *block;
	isize        used;
};

ArenaTemp arena_temp_begin(Arena *arena) {
	GB_ASSERT(arena);
	GB_ASSERT(arena->parent_thread == get_current_thread());

	ArenaTemp temp = {};
	temp.arena = arena;
	temp.block = arena->curr_block;
	if (arena->curr_block != nullptr) {
		temp.used = arena->curr_block->used;
	}
	arena->temp_count += 1;
	return temp;
}

void arena_temp_end(ArenaTemp const &temp) {
	GB_ASSERT(temp.arena);
	Arena *arena = temp.arena;
	GB_ASSERT(arena->parent_thread == get_current_thread());

	if (temp.block) {
		bool memory_block_found = false;
		for (MemoryBlock *block = arena->curr_block; block != nullptr; block = block->prev) {
			if (block == temp.block) {
				memory_block_found = true;
				break;
			}
		}
		GB_ASSERT_MSG(memory_block_found, "memory block stored within ArenaTemp not owned by Arena");

		while (arena->curr_block != temp.block) {
			MemoryBlock *free_block = arena->curr_block;
			if (free_block != nullptr) {
				arena->curr_block = free_block->prev;
				virtual_memory_dealloc(free_block);
			}
		}

		MemoryBlock *block = arena->curr_block;
		if (block) {
			GB_ASSERT_MSG(block->used >= temp.used, "out of order use of arena_temp_end");
			isize amount_to_zero = gb_min(block->used - temp.used, block->size - block->used);
			gb_zero_size(block->base + temp.used, amount_to_zero);
			block->used = temp.used;
		}
	}

	GB_ASSERT_MSG(arena->temp_count > 0, "double-use of arena_temp_end");
	arena->temp_count -= 1;
}

void arena_temp_ignore(ArenaTemp const &temp) {
	GB_ASSERT(temp.arena);
	Arena *arena = temp.arena;
	GB_ASSERT(arena->parent_thread == get_current_thread());

	GB_ASSERT_MSG(arena->temp_count > 0, "double-use of arena_temp_end");
	arena->temp_count -= 1;
}



struct ArenaTempGuard {
	ArenaTempGuard(Arena *arena) {
		this->temp = arena_temp_begin(arena);
	}
	~ArenaTempGuard() {
		arena_temp_end(this->temp);
	}
	ArenaTemp temp;
};


gb_internal GB_ALLOCATOR_PROC(arena_allocator_proc);

gb_internal gbAllocator arena_allocator(Arena *arena) {
	gbAllocator a;
	a.proc = arena_allocator_proc;
	a.data = arena;
	return a;
}


gb_internal GB_ALLOCATOR_PROC(arena_allocator_proc) {
	void *ptr = nullptr;
	Arena *arena = cast(Arena *)allocator_data;
	GB_ASSERT_NOT_NULL(arena);

	switch (type) {
	case gbAllocation_Alloc:
		ptr = arena_alloc(arena, size, alignment);
		break;
	case gbAllocation_Free:
		break;
	case gbAllocation_Resize:
		if (size == 0) {
			ptr = nullptr;
		} else if (size <= old_size) {
			ptr = old_memory;
		} else {
			ptr = arena_alloc(arena, size, alignment);
			gb_memmove(ptr, old_memory, old_size);
		}
		break;
	case gbAllocation_FreeAll:
		GB_PANIC("use arena_free_all directly");
		arena_free_all(arena);
		break;
	}

	return ptr;
}


enum ThreadArenaKind : uintptr {
	ThreadArena_Permanent,
	ThreadArena_Temporary,
};

gb_global Arena default_permanent_arena = {nullptr, DEFAULT_MINIMUM_BLOCK_SIZE};
gb_global Arena default_temporary_arena = {nullptr, DEFAULT_MINIMUM_BLOCK_SIZE};


gb_internal Arena *get_arena(ThreadArenaKind kind) {
	Thread *t = get_current_thread();
	switch (kind) {
	case ThreadArena_Permanent: return t ? t->permanent_arena : &default_permanent_arena;
	case ThreadArena_Temporary: return t ? t->temporary_arena : &default_temporary_arena;
	}
	GB_PANIC("INVALID ARENA KIND");
	return nullptr;
}



gb_internal GB_ALLOCATOR_PROC(thread_arena_allocator_proc) {
	void *ptr = nullptr;
	ThreadArenaKind kind = cast(ThreadArenaKind)cast(uintptr)allocator_data;
	Arena *arena = get_arena(kind);

	switch (type) {
	case gbAllocation_Alloc:
		ptr = arena_alloc(arena, size, alignment);
		break;
	case gbAllocation_Free:
		break;
	case gbAllocation_Resize:
		if (size == 0) {
			ptr = nullptr;
		} else if (size <= old_size) {
			ptr = old_memory;
		} else {
			ptr = arena_alloc(arena, size, alignment);
			gb_memmove(ptr, old_memory, old_size);
		}
		break;
	case gbAllocation_FreeAll:
		GB_PANIC("use arena_free_all directly");
		arena_free_all(arena);
		break;
	}

	return ptr;
}



gb_internal gbAllocator permanent_allocator() {
	return {thread_arena_allocator_proc, cast(void *)cast(uintptr)ThreadArena_Permanent};
}

gb_internal gbAllocator temporary_allocator() {
	return {thread_arena_allocator_proc, cast(void *)cast(uintptr)ThreadArena_Permanent};
}


#define TEMP_ARENA_GUARD(arena) ArenaTempGuard GB_DEFER_3(_arena_guard_){arena}


// #define TEMPORARY_ALLOCATOR_GUARD()
#define TEMPORARY_ALLOCATOR_GUARD() TEMP_ARENA_GUARD(get_arena(ThreadArena_Temporary))
#define PERMANENT_ALLOCATOR_GUARD()



gb_internal bool IS_ODIN_DEBUG(void);

gb_internal GB_ALLOCATOR_PROC(heap_allocator_proc);


gb_internal gbAllocator heap_allocator(void) {
	gbAllocator a;
	a.proc = heap_allocator_proc;
	a.data = nullptr;
	return a;
}

gb_internal std::atomic<isize> total_heap_memory_allocated;


gb_internal GB_ALLOCATOR_PROC(heap_allocator_proc) {
	void *ptr = nullptr;
	gb_unused(allocator_data);
	gb_unused(old_size);


// TODO(bill): Throughly test!
	switch (type) {
#if defined(GB_COMPILER_MSVC)
	case gbAllocation_Alloc:
		if (size == 0) {
			return NULL;
		} else {
			isize aligned_size = align_formula_isize(size, alignment);
			// TODO(bill): Make sure this is aligned correctly
			ptr = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, aligned_size);
		}
		break;
	case gbAllocation_Free:
		if (old_memory != nullptr) {
			HeapFree(GetProcessHeap(), 0, old_memory);
		}
		break;
	case gbAllocation_Resize:
		if (old_memory != nullptr && size > 0) {
			isize aligned_size = align_formula_isize(size, alignment);
			ptr = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, old_memory, aligned_size);
		} else if (old_memory != nullptr) {
			HeapFree(GetProcessHeap(), 0, old_memory);
		} else if (size != 0) {
			isize aligned_size = align_formula_isize(size, alignment);
			// TODO(bill): Make sure this is aligned correctly
			ptr = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, aligned_size);
		}
		break;
#elif defined(GB_SYSTEM_LINUX)
	// TODO(bill): *nix version that's decent
	case gbAllocation_Alloc: {
		isize total_size = (size + alignment - 1) & ~(alignment - 1);
		total_heap_memory_allocated.fetch_add(total_size);
		ptr = aligned_alloc(alignment, total_size);
		gb_zero_size(ptr, size);
	} break;

	case gbAllocation_Free:
		if (old_memory != nullptr) {
			total_heap_memory_allocated.fetch_sub(malloc_usable_size(old_memory));
			free(old_memory);
		}
		break;

	case gbAllocation_Resize: {
		if (size == 0) {
			if (old_memory != nullptr) {
				total_heap_memory_allocated.fetch_sub(malloc_usable_size(old_memory));
				free(old_memory);
			}
			break;
		}

		alignment = gb_max(alignment, gb_align_of(max_align_t));

		if (old_memory == nullptr) {
			isize total_size = (size + alignment - 1) & ~(alignment - 1);
			total_heap_memory_allocated.fetch_add(total_size);
			ptr = aligned_alloc(alignment, total_size);
			gb_zero_size(ptr, size);
			break;
		}
		if (size <= old_size) {
			ptr = old_memory;
			break;
		}

		size_t actual_old_size = malloc_usable_size(old_memory);
		if (size <= actual_old_size) {
			ptr = old_memory;
			break;
		}

		isize total_size = (size + alignment - 1) & ~(alignment - 1);
		total_heap_memory_allocated.fetch_add(total_size);
		ptr = aligned_alloc(alignment, total_size);
		gb_memmove(ptr, old_memory, old_size);
		free(old_memory);
		gb_zero_size(cast(u8 *)ptr + old_size, gb_max(size-old_size, 0));
	} break;
#else
	// TODO(bill): *nix version that's decent
	case gbAllocation_Alloc: {
		int err = 0;
		alignment = gb_max(alignment, gb_align_of(max_align_t));
		
		err = posix_memalign(&ptr, alignment, size);
		GB_ASSERT_MSG(err == 0, "posix_memalign err: %d", err);
		gb_zero_size(ptr, size);
	} break;

	case gbAllocation_Free:
		if (old_memory != nullptr) {
			free(old_memory);
		}
		break;

	case gbAllocation_Resize: {
		int err = 0;
		if (size == 0) {
			free(old_memory);
			break;
		}
		
		alignment = gb_max(alignment, gb_align_of(max_align_t));
		
		if (old_memory == nullptr) {
			err = posix_memalign(&ptr, alignment, size);
			GB_ASSERT_MSG(err == 0, "posix_memalign err: %d", err);
			GB_ASSERT(ptr != nullptr);
			gb_zero_size(ptr, size);
			break;
		}
		if (size <= old_size) {
			ptr = old_memory;
			break;
		}

		err = posix_memalign(&ptr, alignment, size);
		GB_ASSERT_MSG(err == 0, "posix_memalign err: %d", err);
		GB_ASSERT(ptr != nullptr);
		gb_memmove(ptr, old_memory, old_size);
		free(old_memory);
		gb_zero_size(cast(u8 *)ptr + old_size, gb_max(size-old_size, 0));
	} break;
#endif

	case gbAllocation_FreeAll:
		break;
	}

	return ptr;
}


template <typename T>
gb_internal isize resize_array_raw(T **array, gbAllocator const &a, isize old_count, isize new_count, isize custom_alignment=1) {
	GB_ASSERT(new_count >= 0);
	if (new_count == 0) {
		gb_free(a, *array);
		*array = nullptr;
		return 0;
	}
	if (new_count < old_count) {
		return old_count;
	}
	isize old_size = old_count * gb_size_of(T);
	isize new_size = new_count * gb_size_of(T);
	isize alignment = gb_max(gb_align_of(T), custom_alignment);
	auto new_data = cast(T *)gb_resize_align(a, *array, old_size, new_size, alignment);
	GB_ASSERT(new_data != nullptr);
	*array = new_data;
	return new_count;
}