Redcraft/Redcraft.Utility/Source/Public/Memory/Allocator.h

310 lines
12 KiB
C++

#pragma once
#include "CoreTypes.h"
#include "Memory/Memory.h"
#include "Templates/Noncopyable.h"
#include "TypeTraits/TypeTraits.h"
#include "Miscellaneous/AssertionMacros.h"
NAMESPACE_REDCRAFT_BEGIN
NAMESPACE_MODULE_BEGIN(Redcraft)
NAMESPACE_MODULE_BEGIN(Utility)
struct FAllocatorInterface;
template <typename A, typename T = int>
concept CAllocator = !CSameAs<A, FAllocatorInterface>
&& requires (typename A::template ForElementType<T>& Allocator, T* InPtr, size_t Num, size_t NumAllocated)
{
{ Allocator.Allocate(Num) } -> CSameAs<T*>;
{ Allocator.Deallocate(InPtr) } -> CSameAs<void>;
{ AsConst(Allocator).IsTransferable(InPtr) } -> CBooleanTestable;
{ AsConst(Allocator).CalculateSlackGrow(Num, NumAllocated) } -> CSameAs<size_t>;
{ AsConst(Allocator).CalculateSlackShrink(Num, NumAllocated) } -> CSameAs<size_t>;
{ AsConst(Allocator).CalculateSlackReserve(Num) } -> CSameAs<size_t>;
};
template <typename A, typename T = int>
concept CMultipleAllocator = CAllocator<A, T> && A::bSupportsMultipleAllocation;
/**
* This is the allocator interface, the allocator does not use virtual, this contains the default of
* the allocator interface functions. Unlike std::allocator, IAllocator should be bound to only a object,
* such as a container, because there may be side effects between multiple allocations, for example,
* inline storage cannot be allocated multiple times in TInlineAllocator.
*/
struct FAllocatorInterface
{
/**
* If this flag is false, it is possible to allocate an address that has already been allocated.
* Should be allocated according to the results given by the CalculateSlackReserve() family,
* without needing to allocate memory of the same size as the allocated memory,
* this is to support special allocators such as TInlineAllocator.
*/
static constexpr bool bSupportsMultipleAllocation = true;
template <CObject T>
class ForElementType /*: private FSingleton*/
{
public:
ForElementType() = default;
ForElementType(const ForElementType&) = delete;
ForElementType(ForElementType&&) = delete;
ForElementType& operator=(const ForElementType&) = delete;
ForElementType& operator=(ForElementType&&) = delete;
/** Allocates uninitialized storage. If 'InNum' is zero, return nullptr. */
NODISCARD FORCEINLINE T* Allocate(size_t InNum) = delete;
/** Deallocates storage. */
FORCEINLINE void Deallocate(T* InPtr) = delete;
/** @return true if allocation can be deallocated by another allocator, otherwise false. */
NODISCARD FORCEINLINE bool IsTransferable(T* InPtr) const { return true; }
/** Calculates the amount of slack to allocate for an array that has just grown to a given number of elements. */
NODISCARD FORCEINLINE size_t CalculateSlackGrow(size_t Num, size_t NumAllocated) const = delete;
/** Calculates the amount of slack to allocate for an array that has just shrunk to a given number of elements. */
NODISCARD FORCEINLINE size_t CalculateSlackShrink(size_t Num, size_t NumAllocated) const = delete;
/** Calculates the amount of slack to allocate for an array that has just grown or shrunk to a given number of elements. */
NODISCARD FORCEINLINE size_t CalculateSlackReserve(size_t Num) const = delete;
};
};
#define ALLOCATOR_WRAPPER_BEGIN(Allocator, Type, Name) \
\
struct PREPROCESSOR_JOIN(F, Name) /*: private FSingleton*/
#define ALLOCATOR_WRAPPER_END(Allocator, Type, Name) ; \
\
template <typename A, bool = CEmpty<A> && !CFinal<A>> \
struct PREPROCESSOR_JOIN(T, Name); \
\
template <typename A> \
struct PREPROCESSOR_JOIN(T, Name)<A, true> : public PREPROCESSOR_JOIN(F, Name), private A \
{ \
NODISCARD FORCEINLINE A& operator*() { return *this; } \
NODISCARD FORCEINLINE const A& operator*() const { return *this; } \
NODISCARD FORCEINLINE A* operator->() { return this; } \
NODISCARD FORCEINLINE const A* operator->() const { return this; } \
}; \
\
template <typename A> \
struct PREPROCESSOR_JOIN(T, Name)<A, false> : public PREPROCESSOR_JOIN(F, Name) \
{ \
NODISCARD FORCEINLINE A& operator*() { return AllocatorInstance; } \
NODISCARD FORCEINLINE const A& operator*() const { return AllocatorInstance; } \
NODISCARD FORCEINLINE A* operator->() { return &AllocatorInstance; } \
NODISCARD FORCEINLINE const A* operator->() const { return &AllocatorInstance; } \
\
private: \
\
A AllocatorInstance; \
\
}; \
\
PREPROCESSOR_JOIN(T, Name)<typename Allocator::template ForElementType<Type>> Name;
/** This is heap allocator that calls Memory::Malloc() directly for memory allocation. */
struct FHeapAllocator
{
static constexpr bool bSupportsMultipleAllocation = true;
template <CObject T>
class ForElementType
{
public:
ForElementType() = default;
ForElementType(const ForElementType&) = delete;
ForElementType(ForElementType&&) = delete;
ForElementType& operator=(const ForElementType&) = delete;
ForElementType& operator=(ForElementType&&) = delete;
NODISCARD FORCEINLINE T* Allocate(size_t InNum)
{
return InNum != 0 ? static_cast<T*>(Memory::Malloc(Memory::QuantizeSize(InNum * sizeof(T)), alignof(T))) : nullptr;
}
FORCEINLINE void Deallocate(T* InPtr)
{
Memory::Free(InPtr);
}
NODISCARD FORCEINLINE bool IsTransferable(T* InPtr) const { return true; }
NODISCARD FORCEINLINE size_t CalculateSlackGrow(size_t Num, size_t NumAllocated) const
{
const size_t FirstGrow = 4;
const size_t ConstantGrow = 16;
size_t Result;
check(Num > NumAllocated);
Result = (NumAllocated != 0) ? (Num + 3 * Num / 8 + ConstantGrow) : (Num > FirstGrow ? Num : FirstGrow);
Result = Memory::QuantizeSize(Result * sizeof(T), alignof(T)) / sizeof(T);
return Result;
}
NODISCARD FORCEINLINE size_t CalculateSlackShrink(size_t Num, size_t NumAllocated) const
{
size_t Result;
check(Num < NumAllocated);
const bool bTooManySlackBytes = (NumAllocated - Num) * sizeof(T) >= 16 * 1024;
const bool bTooManySlackElements = 3 * Num < 2 * NumAllocated;
const bool bNeedToShrink = (bTooManySlackBytes || bTooManySlackElements) && (NumAllocated - Num > 64 || Num == 0);
if (bNeedToShrink)
{
Result = Num != 0 ? Memory::QuantizeSize(Num * sizeof(T), alignof(T)) / sizeof(T) : 0;
}
else
{
Result = NumAllocated;
}
return Result;
}
NODISCARD FORCEINLINE size_t CalculateSlackReserve(size_t Num) const
{
return Num != 0 ? Memory::QuantizeSize(Num * sizeof(T), alignof(T)) / sizeof(T) : 0;
}
};
};
/**
* The inline allocator allocates up to a specified number of elements in the same allocation as the container.
* Any allocation needed beyond that causes all data to be moved into an indirect allocation.
*/
template <size_t NumInline, CAllocator SecondaryAllocator = FHeapAllocator>
struct TInlineAllocator
{
static constexpr bool bSupportsMultipleAllocation = false;
template <CObject T>
class ForElementType
{
public:
ForElementType() = default;
ForElementType(const ForElementType&) = delete;
ForElementType(ForElementType&&) = delete;
ForElementType& operator=(const ForElementType&) = delete;
ForElementType& operator=(ForElementType&&) = delete;
NODISCARD FORCEINLINE T* Allocate(size_t InNum)
{
if (InNum == 0) return nullptr;
check(InNum >= NumInline);
if (InNum == NumInline) return Impl.GetInline();
return Impl->Allocate(InNum);
}
FORCEINLINE void Deallocate(T* InPtr)
{
if (InPtr == Impl.GetInline()) return;
Impl->Deallocate(InPtr);
}
NODISCARD FORCEINLINE bool IsTransferable(T* InPtr) const
{
if (InPtr == Impl.GetInline()) return false;
return Impl->IsTransferable(InPtr);
}
NODISCARD FORCEINLINE size_t CalculateSlackGrow(size_t Num, size_t NumAllocated) const
{
check(Num > NumAllocated);
check(NumAllocated >= NumInline);
if (Num <= NumInline) return NumInline;
return Impl->CalculateSlackGrow(Num, NumAllocated <= NumInline ? 0 : NumAllocated);
}
NODISCARD FORCEINLINE size_t CalculateSlackShrink(size_t Num, size_t NumAllocated) const
{
check(Num < NumAllocated);
check(NumAllocated >= NumInline);
if (Num <= NumInline) return NumInline;
return Impl->CalculateSlackShrink(Num, NumAllocated);
}
NODISCARD FORCEINLINE size_t CalculateSlackReserve(size_t Num) const
{
if (Num <= NumInline) return NumInline;
return Impl->CalculateSlackReserve(Num);
}
private:
ALLOCATOR_WRAPPER_BEGIN(SecondaryAllocator, T, Impl)
{
TAlignedStorage<sizeof(T), alignof(T)> InlineStorage[NumInline];
NODISCARD FORCEINLINE T* GetInline() { return reinterpret_cast< T*>(&InlineStorage); }
NODISCARD FORCEINLINE const T* GetInline() const { return reinterpret_cast<const T*>(&InlineStorage); }
}
ALLOCATOR_WRAPPER_END(SecondaryAllocator, T, Impl)
};
};
/** This is a null allocator for which all operations are illegal. */
struct FNullAllocator
{
static constexpr bool bSupportsMultipleAllocation = true;
template <CObject T>
class ForElementType
{
public:
ForElementType() = default;
ForElementType(const ForElementType&) = delete;
ForElementType(ForElementType&&) = delete;
ForElementType& operator=(const ForElementType&) = delete;
ForElementType& operator=(ForElementType&&) = delete;
NODISCARD FORCEINLINE T* Allocate(size_t InNum) { check_no_entry(); return nullptr; }
FORCEINLINE void Deallocate(T* InPtr) { check_no_entry(); }
NODISCARD FORCEINLINE bool IsTransferable(T* InPtr) const { check_no_entry(); return false; }
NODISCARD FORCEINLINE size_t CalculateSlackGrow(size_t Num, size_t NumAllocated) const { check_no_entry(); return 0; }
NODISCARD FORCEINLINE size_t CalculateSlackShrink(size_t Num, size_t NumAllocated) const { check_no_entry(); return 0; }
NODISCARD FORCEINLINE size_t CalculateSlackReserve(size_t Num) const { check_no_entry(); return 0; }
};
};
template <size_t Num>
using TFixedAllocator = TInlineAllocator<Num, FNullAllocator>;
NAMESPACE_MODULE_END(Utility)
NAMESPACE_MODULE_END(Redcraft)
NAMESPACE_REDCRAFT_END