I am using c++03, and trying to figure out which container to use. The hardware is MSP430F5324 with total of 6k of RAM (only 5k is available to me), and 64k flash. I am trying to store combination of derived classes, for example:
class OptNode {
public:
virtual ~OptNode();
explicit OptNode(uint16_t option_no) : option(option_no), _next(0) {}
protected:
uint16_t option;
unsigned length;
};
// --- Use heap memory for the option
class OptNodeDynamic : public OptNode {
public:
// --- assignment operators and copy constructor are not defined
OptNodeDynamic(const OptNodeDynamic &cSource);
OptNodeDynamic& operator= (const OptNodeDynamic &cSource);
OptNodeDynamic(uint16_t option_no, uint8_t* option_data, size_t length);
~OptNodeDynamic();
private:
uint8_t* data;
};
// --- Use the c string for the option
class OptNodeCstr : public OptNode {
public:
OptNodeCstr(uint16_t option_no, const char* option_data);
private:
const char* data;
};
// ----- OptNodeDynamic() ------------------------------------------------------
OptNodeDynamic::OptNodeDynamic(uint16_t option_no,
uint8_t* option_data,
size_t length) : OptNode(option_no) {
//
option = option_no;
data = new uint8_t[length];
std::memcpy(data, option_data, length);
}
// ----- ~OptNodeDynamic() -----------------------------------------------------
OptNodeDynamic::~OptNodeDynamic() {
if (data != 0 && length != 0) {
delete data;
}
}
// ----- OptNodeCstr() ---------------------------------------------------------
OptNodeCstr::OptNodeCstr(uint16_t option_no,
const char* option_data)
: OptNode(option_no),
data(option_data) {
//
length = 0;
for (const char* itr_ptr = data; *itr_ptr != '\0'; ++itr_ptr, ++length) {
/* no code */
}
return;
}
The container should allocate only what it needs for each of the objects. I should be able to add objects of type B or C in any order and as many as needed. Is there such container? Or do I instantiate each object using new operator and pass the pointer to a container, such as list.
Edit: I added hardware description and changed class from example to the actual. I am thinking that the system will need from 0 to 8 instances during data upload process, which happens at most once per hour. Other times, I can use that heap memory for something else.
None.
You do not have enough memory on your platform for the C++ runtime library. You should not and cannot use any C++ container.
Frankly I'd stick with good ol' C for this. As for what code you actually write to achieve your goal, well that depends on your goal. :)
There is a significant overhead when allocating individual objects with new (up to several tenths of bytes per call to new).
If you want to save memory, the best thing to do is to pre-allocate three chunks of memory (pools) for storing the instances of A, B and C, and overload the new and delete operators for A,B and C.
You will find examples in the answer of the following question:
Using operator new and operator delete with a custom memory pool/allocator
Preallocating all the objects at init time and then trying to use the pointers throughout the life cycle of the process will be better idea.
Related
I want to create my own game engine so I bought a few books one being Game Engine Architecture Second Edition by Jason Gregory and in it he suggests implementing a few custom allocators. One type of allocator the book talked about was a stack-based allocator, but I got confused when reading it. How do you store data in it? What data type do you use? For example, do you use a void*, void**, an array of char[]? The book says you're meant to allocate one big block of memory using malloc in the begining and free it in the end, and "allocate" memory by incrementing a pointer. If you could help explain this more that would be great because I can't seem to find a tutorial that doesn't use std::allocator. I also thought this might help others interested in a custom allocator so I posted the question here.
This is the header file example they give in the book:
class StackAllocator
{
public:
// Represents the current top of the stack.
// You can only roll back to the marker not to arbitrary locations within the stack
typedef U32 Marker;
explicit StackAllocator(U32 stackSize_bytes);
void* alloc(U32 size_bytes); // Allocates a new block of the given size from stack top
Marker getMarker(); // Returns a Marker to the current stack top
void freeToMarker(Marker marker); // Rolls the stack back to a previous marker
void clear(); // Clears the entire stack(rolls the stack back to zero)
private:
// ...
}
EDIT:
After a while I got this working but I don't know if I'm doing it right
Header File
typedef std::uint32_t U32;
struct Marker {
size_t currentSize;
};
class StackAllocator
{
private:
void* m_buffer; // Buffer of memory
size_t m_currSize = 0;
size_t m_maxSize;
public:
void init(size_t stackSize_bytes); // allocates size of memory
void shutDown();
void* allocUnaligned(U32 size_bytes);
Marker getMarker();
void freeToMarker(Marker marker);
void clear();
};
.cpp File
void StackAllocator::init(size_t stackSize_bytes) {
this->m_buffer = malloc(stackSize_bytes);
this->m_maxSize = stackSize_bytes;
}
void StackAllocator::shutDown() {
this->clear();
free(m_buffer);
m_buffer = nullptr;
}
void* StackAllocator::allocUnaligned(U32 size_bytes) {
assert(m_maxSize - m_currSize >= size_bytes);
m_buffer = static_cast<char*>(m_buffer) + size_bytes;
m_currSize += size_bytes;
return m_buffer;
}
Marker StackAllocator::getMarker() {
Marker marker;
marker.currentSize = m_currSize;
return marker;
}
void StackAllocator::freeToMarker(Marker marker) {
U32 difference = m_currSize - marker.currentSize;
m_currSize -= difference;
m_buffer = static_cast<char*>(m_buffer) - difference;
}
void StackAllocator::clear() {
m_buffer = static_cast<char*>(m_buffer) - m_currSize;
}
Okay for simplicity let's say you're tracking a collection of MyFunClass for your engine. It could be anything, and your linear allocator doesn't necessarily have to track objects of a homogenous type, but often that's how it's done. In general, when using custom allocators you're trying to "shape" your memory allocations to separate static data from dynamic, infrequently accessed vs. frequently accessed, with a view towards optimizing your working set and achieving locality of reference.
Given the code you provided, first, you'd allocate your memory pool. For simplicity, assume you want enough space to pool 1000 objects of type MyFunClass.
StackAllocator sa;
sa.Init( 1000 * sizeof(MyFunClass) );
Then each time you need to "allocate" a new block of memory for a FunClass, you might do it like this:
void* mem = sa.allocUnaligned( sizeof(MyFunClass) );
Of course, this doesn't actually allocate anything. All the allocation already happened in Step 1. It just marks some of your already-allocated memory as in-use.
It also doesn't construct a MyFunClass. Your allocator isn't strongly typed, so the memory it returns can be interpreted however you want: as a stream of bytes; as a backing representation of a C++ class object; etc.
Now, how would you use a buffer allocated in this fashion? One common way is with placement new:
auto myObj = new (mem) MyFunClass();
So now you're constructing your C++ object in the memory space you reserved with the call to allocUnaligned.
(Note that the allocUnaligned bit gives you some insight into why we don't usually write our own custom allocators: because they're hard as heck to get right! We haven't even mentioned alignment issues yet.)
For extra credit, take a look at scope stacks which take the linear allocator approach to the next level.
I have a class that contains several arrays whose sizes can be determined by parameters to its constructor. My problem is that instances of this class have sizes that can't be determined at compile time, and I don't know how to tell a new method at run time how big I need my object to be. Each object will be of a fixed size, but different instances may be different sizes.
There are several ways around the problem:- use a factory- use a placement constructor- allocate arrays in the constructor and store pointers to them in my object.
I am adapting some legacy code from an old application written in C. In the original code, the program figures out how much memory will be needed for the entire object, calls malloc() for that amount, and proceeds to initialize the various fields.
For the C++ version, I'd like to be able to make a (fairly) normal constructor for my object. It will be a descendant of a parent class, and some of the code will be depending on polymorphism to call the right method. Other classes descended from the same parent have sizes known at compile time, and thus present no problem.
I'd like to avoid some of the special considerations necessary when using placement new, and I'd like to be able to delete the objects in a normal way.
I'd like to avoid carrying pointers within the body of my object, partially to avoid ownership problems associated with copying the object, and partially because I would like to re-use as much of the existing C code as possible. If ownership were the only issue, I could probably just use shared pointers and not worry.
Here's a very trimmed-down version of the C code that creates the objects:
typedef struct
{
int controls;
int coords;
} myobject;
myobject* create_obj(int controls, int coords)
{
size_t size = sizeof(myobject) + (controls + coords*2) * sizeof(double);
char* mem = malloc(size);
myobject* p = (myobject *) mem;
p->controls = controls;
p->coords = coords;
return p;
}
The arrays within the object maintain a fixed size of the life of the object. In the code above, memory following the structure of myobject will be used to hold the array elements.
I feel like I may be missing something obvious. Is there some way that I don't know about to write a (fairly) normal constructor in C++ but be able to tell it how much memory the object will require at run time, without resorting to a "placement new" scenario?
How about a pragmatic approach: keep the structure as is (if compatibility with C is important) and wrap it into a c++ class?
typedef struct
{
int controls;
int coords;
} myobject;
myobject* create_obj(int controls, int coords);
void dispose_obj(myobject* obj);
class MyObject
{
public:
MyObject(int controls, int coords) {_data = create_obj(controls, coords);}
~MyObject() {dispose_obj(_data);}
const myobject* data() const
{
return _data;
}
myobject* data()
{
return _data;
}
int controls() const {return _data->controls;}
int coords() const {return _data->coords;}
double* array() { return (double*)(_data+1); }
private:
myobject* _data;
}
While I understand the desire to limit the changes to the existing C code, it would be better to do it correctly now rather than fight with bugs in the future. I suggest the following structure and changes to your code to deal with it (which I suspect would mostly be pulling out code that calculates offsets).
struct spots
{
double x;
double y;
};
struct myobject
{
std::vector<double> m_controls;
std::vector<spots> m_coordinates;
myobject( int controls, int coordinates ) :
m_controls( controls ),
m_coordinates( coordinates )
{ }
};
To maintain the semantics of the original code, where the struct and array are in a single contigious block of memory, you can simply replace malloc(size) with new char[size] instead:
myobject* create_obj(int controls, int coords)
{
size_t size = sizeof(myobject) + (controls + coords*2) * sizeof(double);
char* mem = new char[size];
myobject* p = new(mem) myobject;
p->controls = controls;
p->coords = coords;
return p;
}
You will have to use a type-cast when freeing the memory with delete[], though:
myobject *p = create_obj(...);
...
p->~myobject();
delete[] (char*) p;
In this case, I would suggest wrapping that logic in another function:
void free_obj(myobject *p)
{
p->~myobject();
delete[] (char*) p;
}
myobject *p = create_obj(...);
...
free_obj(p);
That being said, if you are allowed to, it would be better to re-write the code to follow C++ semantics instead, eg:
struct myobject
{
int controls;
int coords;
std::vector<double> values;
myobject(int acontrols, int acoords) :
controls(acontrols),
coords(acoords),
values(acontrols + acoords*2)
{
}
};
And then you can do this:
std::unique_ptr<myobject> p = std::make_unique<myobject>(...); // C++14
...
std::unique_ptr<myobject> p(new myobject(...)); // C++11
...
std::auto_ptr<myobject> p(new myobject(...)); // pre C++11
...
New Answer (given comment from OP):
Allocate a std::vector<byte> of the correct size. The array allocated to back the vector will be contiguous memory. This vector size can be calculated and the vector will manage your memory correctly. You will still need to be very careful about how you manage your access to that byte array obviously, but you can use iterators and the like at least (if you want).
By the way here is a little template thing I use to move along byte blobs with a little more grace (note this has aliasing issues as pointed out by Sergey in the comments below, I'm leaving it here because it seems to be a good example of what not to do... :-) ) :
template<typename T>
T readFromBuf(byte*& ptr) {
T * const p = reinterpret_cast<T*>(ptr);
ptr += sizeof(T);
return *p;
}
Old Answer:
As the comments suggest, you can easily use a std::vector to do what you want. Also I would like to make another suggestion.
size_t size = sizeof(myobject) + (controls + coords*2) * sizeof(double);
The above line of code suggests to me that you have some "hidden structure" in your code. Your myobject struct has two int values from which you are calculating the size of what you actually need. What you actually need is this:
struct ControlCoord {
double control;
std::pair<double, double> coordinate;
};
std::vector<ControlCoord>> controlCoords;
When the comments finally scheded some light on the actual requirements, the solution would be following:
allocate a buffer large enough to hold your object and the array
use placement new in the beginning of the buffer
Here is how:
class myobject {
myobject(int controls, int coords) : controls(controls), coords(coords) {}
~myobject() {};
public:
const int controls;
const int coords;
static myobject* create(int controls, int coords) {
std::unique_ptr<char> buffer = new char[sizeof(myobject) + (controls + coords*2) * sizeof(double)];
myobject obj* = new (buffer.get()) myobject(controls, coords);
buffer.release();
return obj;
}
void dispose() {
~myobject();
char* p = (char*)this;
delete[] p;
}
};
myobject *p = myobject::create(...);
...
p->dispose();
(or suitably wrapped inside deleter for smart pointer)
I have an OOP entity-component system that currently works like this:
// In the component system
struct Component { virtual void update() = 0; }
struct Entity
{
bool alive{true};
vector<unique_ptr<Component>> components;
void update() { for(const auto& c : components) c->update(); }
}
// In the user application
struct MyComp : Component
{
void update() override { ... }
}
To create new entities and components, I use C++'s usual new and delete:
// In the component system
struct Manager
{
vector<unique_ptr<Entity>> entities;
Entity& createEntity()
{
auto result(new Entity);
entities.emplace_back(result);
return *result;
}
template<typename TComp, typename... TArgs>
TComp& createComponent(Entity& mEntity, TArgs... mArgs)
{
auto result(new TComp(forward<TArgs>(mArgs)...));
mEntity.components.emplace_back(result);
return result;
}
void removeDead() { /* remove all entities with 'alive == false' - 'delete' is called here by the 'unique_ptr' */ }
}
// In the user application
{
Manager m;
auto& myEntity(m.createEntity());
auto& myComp(m.createComponent<MyComp>(myEntity));
// Do stuff with myEntity and myComp
m.removeDead();
}
The system works fine, and I like the syntax and flexibility. However, when continuously adding and removing entities and components to the manager, memory allocation/deallocation slows down the application. (I've profiled and determined that the slow down is caused by new and delete).
I've recently read that it's possible to pre-allocate heap memory in C++ - how can that be applied to my situation?
Desired result:
// In the user application
{
Manager m{1000};
// This manager can hold about 1000 entities with components
// (may not be 1000 because of dynamic component size,
// since the user can define it's on components, but it's ok for me)
auto& myEntity(m.createEntity());
auto& myComp(m.createComponent<MyComp>(myEntity));
// Do stuff with myEntity and myComp
m.removeDead();
// No 'delete' is called here! Memory of the 'dead' entities can
// be reused for new entity creation
}
// Manager goes out of scope: 'delete' is called here
There are a few things you can do to make the implementation of your design scale better.
In your current implementation there are two memory allocations per Entity and Component. The first one allocates an object and the second one when the object is put into the vector. The second one happens when the vector runs out of space and allocates a bigger array and moves the old elements into the new array.
In this case the best you can do is to use intrusive lists. That is, each of Entity and Component become also list nodes. Then, after these have been allocated no extra memory allocations are necessary to put the object into a list. Use a single or double-linked list from Boost.Intrusive, or write your own. This is how Linux kernel keeps track of many different objects.
The next step is to preallocate Entity and Component elements. Preallocating could be something as simple as a global array of these, or something more sophisticated, such as Boost.Pool. There are quite a few ways to build a memory pool of objects.
Once Entity and Component are preallocated and intrusive lists are used you are done.
An example which uses boost components:
#include <boost/intrusive/list.hpp>
#include <boost/pool/pool_alloc.hpp>
#include <new>
namespace bi = boost::intrusive;
// api.h
//
// Object pooling support begin.
//
template<class T>
struct Pool
{
static boost::pool_allocator<T> pool;
};
// Singleton. Although it is defined in the header, the linkers
// make sure there is only one instance of it in the application.
// It is instantiated on demand when Pool<T> is used.
template<class T>
boost::pool_allocator<T> Pool<T>::pool;
template<class Derived>
struct Pooled // use it on the most derived class only, not on intermediate base classes
{
// Automatically use the object pool for plain new/delete.
static void* operator new(size_t) { return Pool<Derived>::pool.allocate(1); }
static void operator delete(void* p) { return Pool<Derived>::pool.deallocate(static_cast<Derived*>(p), 1); }
};
//
// Object pooling support end.
//
// Using bi::list_base_hook<bi::link_mode<bi::auto_unlink> > because it automatically
// unlinks from the list when the object is destroyed. No need to manually
// remove the object from the list when an object is about to be destroyed.
struct Component
: bi::list_base_hook<bi::link_mode<bi::auto_unlink> > // make it an intrusive list node
{
virtual void update() = 0;
virtual ~Component() {}
};
struct Entity
: bi::list_base_hook<bi::link_mode<bi::auto_unlink> > // make it an intrusive list node
, Pooled<Entity> // optional, make it allocated from the pool
{
bool active = false;
bi::list<Component, bi::constant_time_size<false> > components;
~Entity() {
for(auto i = components.begin(), j = components.end(); i != j;)
delete &*i++; // i++ to make sure i stays valid after the object is destroyed
}
void update() {
for(auto& c : components)
c.update();
}
};
struct Manager
{
bi::list<Entity, bi::constant_time_size<false> > entities;
~Manager() {
for(auto i = entities.begin(), j = entities.end(); i != j;)
delete &*i++; // i++ to make sure i stays valid after the object is destroyed
}
Entity& createEntity() {
auto result = new Entity;
entities.push_back(*result);
return *result;
}
template<typename TComp, typename... TArgs>
TComp& createComponent(Entity& mEntity, TArgs... mArgs)
{
auto result = new TComp(std::forward<TArgs>(mArgs)...);
mEntity.components.push_back(*result);
return *result;
}
void removeDead() {
for(auto i = entities.begin(), j = entities.end(); i != j;) {
auto& entity = *i++;
if(!entity.active)
delete &entity;
}
}
};
// user.cc
struct MyComp
: Component
, Pooled<MyComp> // optional, make it allocated from the pool
{
void update() override {}
};
int main() {
Manager m;
auto& myEntity(m.createEntity());
auto& myComp(m.createComponent<MyComp>(myEntity));
m.removeDead();
}
In the above example boost::pool_allocator<T> actually uses new to allocate objects and then it keeps reusing destroyed objects rather than invoking delete on them. You can do better by preallocating all objects, but there are many ways to do so depending on your requirements, so that I use boost::pool_allocator<T> for simplicity to avoid hair splitting here. You can change the implementation of Pooled<T> to something like Pooled<T, N> where N stands for the maximum number of objects, the rest of the code stays the same because it uses plain new/delete which happen to be overridden for objects allocated from a pool.
C++ supports class-specific memory pools for this kind of thing. The general-purpose new/delete pair inevitably trades off among
Time spent searching for a free block of the right size to meet each request
Time spent coalescing free blocks
Time spent maintaining and perhaps reorganizing internal data structures to make the above two operations faster.
The primary way to gain speed is to avoid these tradeoffs entirely with custom allocators that - as you say - pre-allocate a big chunk of memory viewed as a simple array of free objects all of the same size. Initially these are all linked on a free list, where the link pointers occupy the first bytes of each block "overlaid" where the data will eventually go. Allocation is just unchaining a block from the head of the free list - a "pop" operation needing about 2 instructions. Deallocation is a "push:" two more instructions. In many cases, memory hardware can be set to generate a trap when the the pool is empty so there is no per-allocation overhead for detecting
this error condition. (In GC systems the same trick is used to initiate collection with no overhead.)
In your case you'd need two pools: one for Entities and one for Components.
Defining your own pool allocator is not so hard, especially if your application is single threaded. See this document for a tutorial treatment.
Using most of answers and Google as references, I implemented some pre-allocation utilities in my SSVUtils library.
Prealloc.h
Example:
using MemUnit = char;
using MemUnitPtr = MemUnit*;
using MemSize = decltype(sizeof(MemUnit)); // Should always be 1 byte
class MemBuffer
{
Uptr<MemUnit[]> buffer;
MemRange range;
MemBuffer(MemSize mSize) : ...
{
// initialize buffer from mSize
}
};
class PreAllocatorChunk
{
protected:
MemSize chunkSize;
MemBuffer buffer;
std::stack<MemRange> available;
public:
PreAllocatorChunk(MemSize mChunkSize, unsigned int mChunks) : ...
{
// Add "chunks" to to available...
}
template<typename T, typename... TArgs> T* create(TArgs&&... mArgs)
{
// create on first "chunk" using placement new
auto toUse(available.top().begin); available.pop();
return new (toUse) T{std::forward<TArgs>(mArgs)...};
}
};
More pre-allocation utilities are available:
PreAllocatorDynamic: pre-allocates a big buffer, then, when creating an object, splits the buffer in two parts:
[buffer start, buffer start + obj size)
[buffer start + obj size, buffer end)
When an object is destroyed, its occupied memory range is set as "available". If during creation of a new object no big enough "chunk" is found, the pre-allocator tries to unify contiguous memory chunks before throwing a runtime exception. This pre-allocator is sometimes faster than new/delete, but it greatly depends on the size of pre-allocated buffer.
PreAllocatorStatic<T>: inherited from PreAllocatorChunk. Size of a chunk is equal to sizeof(T). Fastest pre-allocator, less flexible. Almost always faster than new/delete.
One of your issues can be solved by allocating enough space in the vectors on their creation
For
vector<unique_ptr<Entity>> entities;
provide enough space in the constructor
Manager::Manager() : entities(10000)
{
//...
}
Thus, you avoid reallocations and copying during later stages.
The second issue is the creation of your unique_ptr<Entity> pointers. Here, as you will always use default constructed objects, you could also use a pre-allocated pool of objects from which you create the pointers. Instead of calling new you would call an own class
class EntityPool
{
public:
EntityPool(unsigned int size = 10000) : pool(size), nextEntity(0)
{
}
Entity* getNext(void)
{
if (nextEntity != pool.size()) // if pool is exhausted create new
{
pool.emplace_back(Entity());
}
return pool[nextEntity++];
}
private:
vector<Entity> pool;
unsigned int nextEntity; // index into the vector to the next Entity
};
struct Manager
{
vector<unique_ptr<Entity>> entities;
Entity& createEntity()
{
entities.emplace_back(entityPoolInstance.getNext());
return *result;
}
//...
Or you could wire in the standard 'placement new'. This allows you to allocate a big block of memory to construct (place) objects into as you choose. This will keep the block on the heap for as long as you need it, and allow you to allocate multiple short lived objects into this block instead of doing the costly allocations and de-allocations that just end up fragmenting the heap. There's a couple of gotcha's involved, but all in all its a REALLY simple solution without having to go down the custom memory manager route.
Here's an excellent treatment of removing some of the pitfalls and describing placement new in detail.
I've used data structures as simple as a stack to keep track of the next free block to allocate into: push the address of a block that's about to be deleted onto the stack. When allocating just pop the next free block off of the stack and use that as the arg to placement new. Super easy and super fast!
I have a program that contains a processing phase that needs to use a bunch of different object instances (all allocated on the heap) from a tree of polymorphic types, all eventually derived from a common base class.
As the instances may cyclically reference each other, and do not have a clear owner, I want allocated them with new, handle them with raw pointers, and leave them in memory for the phase (even if they become unreferenced), and then after the phase of the program that uses these instances, I want to delete them all at once.
How I thought to structure it is as follows:
struct B; // common base class
vector<unique_ptr<B>> memory_pool;
struct B
{
B() { memory_pool.emplace_back(this); }
virtual ~B() {}
};
struct D : B { ... }
int main()
{
...
// phase begins
D* p = new D(...);
...
// phase ends
memory_pool.clear();
// all B instances are deleted, and pointers invalidated
...
}
Apart from being careful that all B instances are allocated with new, and that noone uses any pointers to them after the memory pool is cleared, are there problems with this implementation?
Specifically I am concerned about the fact that the this pointer is used to construct a std::unique_ptr in the base class constructor, before the derived class constructor has completed. Does this result in undefined behaviour? If so is there a workaround?
In case you haven't already, familiarize yourself with Boost.Pool. From the Boost documentation:
What is Pool?
Pool allocation is a memory allocation scheme that is very fast, but
limited in its usage. For more information on pool allocation (also
called simple segregated storage, see concepts concepts and Simple Segregated Storage.
Why should I use Pool?
Using Pools gives you more control over how memory is used in your
program. For example, you could have a situation where you want to
allocate a bunch of small objects at one point, and then reach a point
in your program where none of them are needed any more. Using pool
interfaces, you can choose to run their destructors or just drop them
off into oblivion; the pool interface will guarantee that there are no
system memory leaks.
When should I use Pool?
Pools are generally used when there is a lot of allocation and
deallocation of small objects. Another common usage is the situation
above, where many objects may be dropped out of memory.
In general, use Pools when you need a more efficient way to do unusual
memory control.
Which pool allocator should I use?
pool_allocator is a more general-purpose solution, geared towards
efficiently servicing requests for any number of contiguous chunks.
fast_pool_allocator is also a general-purpose solution but is geared
towards efficiently servicing requests for one chunk at a time; it
will work for contiguous chunks, but not as well as pool_allocator.
If you are seriously concerned about performance, use
fast_pool_allocator when dealing with containers such as std::list,
and use pool_allocator when dealing with containers such as
std::vector.
Memory management is tricky business (threading, caching, alignment, fragmentation, etc. etc.) For serious production code, well-designed and carefully optimized libraries are the way to go, unless your profiler demonstrates a bottleneck.
Your idea is great and millions of applications are already using it. This pattern is most famously known as «autorelease pool». It forms a base for ”smart” memory management in Cocoa and Cocoa Touch Objective-C frameworks. Despite the fact that C++ provides hell of a lot of other alternatives, I still think this idea got a lot of upside. But there are few things where I think your implementation as it stands may fall short.
The first problem that I can think of is thread safety. For example, what happens when objects of the same base are created from different threads? A solution might be to protect the pool access with mutually exclusive locks. Though I think a better way to do this is to make that pool a thread-specific object.
The second problem is invoking an undefined behavior in case where derived class's constructor throws an exception. You see, if that happens, the derived object won't be constructed, but your B's constructor would have already pushed a pointer to this to the vector. Later on, when the vector is cleared, it would try to call a destructor through a virtual table of the object that either doesn't exist or is in fact a different object (because new could reuse that address).
The third thing I don't like is that you have only one global pool, even if it is thread-specific, that just doesn't allow for a more fine grained control over the scope of allocated objects.
Taking the above into account, I would do a couple of improvements:
Have a stack of pools for more fine-grained scope control.
Make that pool stack a thread-specific object.
In case of failures (like exception in derived class constructor), make sure the pool doesn't hold a dangling pointer.
Here is my literally 5 minutes solution, don't judge for quick and dirty:
#include <new>
#include <set>
#include <stack>
#include <cassert>
#include <memory>
#include <stdexcept>
#include <iostream>
#define thread_local __thread // Sorry, my compiler doesn't C++11 thread locals
struct AutoReleaseObject {
AutoReleaseObject();
virtual ~AutoReleaseObject();
};
class AutoReleasePool final {
public:
AutoReleasePool() {
stack_.emplace(this);
}
~AutoReleasePool() noexcept {
std::set<AutoReleaseObject *> obj;
obj.swap(objects_);
for (auto *p : obj) {
delete p;
}
stack_.pop();
}
static AutoReleasePool &instance() {
assert(!stack_.empty());
return *stack_.top();
}
void add(AutoReleaseObject *obj) {
objects_.insert(obj);
}
void del(AutoReleaseObject *obj) {
objects_.erase(obj);
}
AutoReleasePool(const AutoReleasePool &) = delete;
AutoReleasePool &operator = (const AutoReleasePool &) = delete;
private:
// Hopefully, making this private won't allow users to create pool
// not on stack that easily... But it won't make it impossible of course.
void *operator new(size_t size) {
return ::operator new(size);
}
std::set<AutoReleaseObject *> objects_;
struct PrivateTraits {};
AutoReleasePool(const PrivateTraits &) {
}
struct Stack final : std::stack<AutoReleasePool *> {
Stack() {
std::unique_ptr<AutoReleasePool> pool
(new AutoReleasePool(PrivateTraits()));
push(pool.get());
pool.release();
}
~Stack() {
assert(!stack_.empty());
delete stack_.top();
}
};
static thread_local Stack stack_;
};
thread_local AutoReleasePool::Stack AutoReleasePool::stack_;
AutoReleaseObject::AutoReleaseObject()
{
AutoReleasePool::instance().add(this);
}
AutoReleaseObject::~AutoReleaseObject()
{
AutoReleasePool::instance().del(this);
}
// Some usage example...
struct MyObj : AutoReleaseObject {
MyObj() {
std::cout << "MyObj::MyObj(" << this << ")" << std::endl;
}
~MyObj() override {
std::cout << "MyObj::~MyObj(" << this << ")" << std::endl;
}
void bar() {
std::cout << "MyObj::bar(" << this << ")" << std::endl;
}
};
struct MyObjBad final : AutoReleaseObject {
MyObjBad() {
throw std::runtime_error("oops!");
}
~MyObjBad() override {
}
};
void bar()
{
AutoReleasePool local_scope;
for (int i = 0; i < 3; ++i) {
auto o = new MyObj();
o->bar();
}
}
void foo()
{
for (int i = 0; i < 2; ++i) {
auto o = new MyObj();
bar();
o->bar();
}
}
int main()
{
std::cout << "main start..." << std::endl;
foo();
std::cout << "main end..." << std::endl;
}
Hmm, I needed almost exactly the same thing recently (memory pool for one phase of a program that gets cleared all at once), except that I had the additional design constraint that all my objects would be fairly small.
I came up with the following "small-object memory pool" -- perhaps it will be of use to you:
#pragma once
#include "defs.h"
#include <cstdint> // uintptr_t
#include <cstdlib> // std::malloc, std::size_t
#include <type_traits> // std::alignment_of
#include <utility> // std::forward
#include <algorithm> // std::max
#include <cassert> // assert
// Small-object allocator that uses a memory pool.
// Objects constructed in this arena *must not* have delete called on them.
// Allows all memory in the arena to be freed at once (destructors will
// be called).
// Usage:
// SmallObjectArena arena;
// Foo* foo = arena::create<Foo>();
// arena.free(); // Calls ~Foo
class SmallObjectArena
{
private:
typedef void (*Dtor)(void*);
struct Record
{
Dtor dtor;
short endOfPrevRecordOffset; // Bytes between end of previous record and beginning of this one
short objectOffset; // From the end of the previous record
};
struct Block
{
size_t size;
char* rawBlock;
Block* prevBlock;
char* startOfNextRecord;
};
template<typename T> static void DtorWrapper(void* obj) { static_cast<T*>(obj)->~T(); }
public:
explicit SmallObjectArena(std::size_t initialPoolSize = 8192)
: currentBlock(nullptr)
{
assert(initialPoolSize >= sizeof(Block) + std::alignment_of<Block>::value);
assert(initialPoolSize >= 128);
createNewBlock(initialPoolSize);
}
~SmallObjectArena()
{
this->free();
std::free(currentBlock->rawBlock);
}
template<typename T>
inline T* create()
{
return new (alloc<T>()) T();
}
template<typename T, typename A1>
inline T* create(A1&& a1)
{
return new (alloc<T>()) T(std::forward<A1>(a1));
}
template<typename T, typename A1, typename A2>
inline T* create(A1&& a1, A2&& a2)
{
return new (alloc<T>()) T(std::forward<A1>(a1), std::forward<A2>(a2));
}
template<typename T, typename A1, typename A2, typename A3>
inline T* create(A1&& a1, A2&& a2, A3&& a3)
{
return new (alloc<T>()) T(std::forward<A1>(a1), std::forward<A2>(a2), std::forward<A3>(a3));
}
// Calls the destructors of all currently allocated objects
// then frees all allocated memory. Destructors are called in
// the reverse order that the objects were constructed in.
void free()
{
// Destroy all objects in arena, and free all blocks except
// for the initial block.
do {
char* endOfRecord = currentBlock->startOfNextRecord;
while (endOfRecord != reinterpret_cast<char*>(currentBlock) + sizeof(Block)) {
auto startOfRecord = endOfRecord - sizeof(Record);
auto record = reinterpret_cast<Record*>(startOfRecord);
endOfRecord = startOfRecord - record->endOfPrevRecordOffset;
record->dtor(endOfRecord + record->objectOffset);
}
if (currentBlock->prevBlock != nullptr) {
auto memToFree = currentBlock->rawBlock;
currentBlock = currentBlock->prevBlock;
std::free(memToFree);
}
} while (currentBlock->prevBlock != nullptr);
currentBlock->startOfNextRecord = reinterpret_cast<char*>(currentBlock) + sizeof(Block);
}
private:
template<typename T>
static inline char* alignFor(char* ptr)
{
const size_t alignment = std::alignment_of<T>::value;
return ptr + (alignment - (reinterpret_cast<uintptr_t>(ptr) % alignment)) % alignment;
}
template<typename T>
T* alloc()
{
char* objectLocation = alignFor<T>(currentBlock->startOfNextRecord);
char* nextRecordStart = alignFor<Record>(objectLocation + sizeof(T));
if (nextRecordStart + sizeof(Record) > currentBlock->rawBlock + currentBlock->size) {
createNewBlock(2 * std::max(currentBlock->size, sizeof(T) + sizeof(Record) + sizeof(Block) + 128));
objectLocation = alignFor<T>(currentBlock->startOfNextRecord);
nextRecordStart = alignFor<Record>(objectLocation + sizeof(T));
}
auto record = reinterpret_cast<Record*>(nextRecordStart);
record->dtor = &DtorWrapper<T>;
assert(objectLocation - currentBlock->startOfNextRecord < 32768);
record->objectOffset = static_cast<short>(objectLocation - currentBlock->startOfNextRecord);
assert(nextRecordStart - currentBlock->startOfNextRecord < 32768);
record->endOfPrevRecordOffset = static_cast<short>(nextRecordStart - currentBlock->startOfNextRecord);
currentBlock->startOfNextRecord = nextRecordStart + sizeof(Record);
return reinterpret_cast<T*>(objectLocation);
}
void createNewBlock(size_t newBlockSize)
{
auto raw = static_cast<char*>(std::malloc(newBlockSize));
auto blockStart = alignFor<Block>(raw);
auto newBlock = reinterpret_cast<Block*>(blockStart);
newBlock->rawBlock = raw;
newBlock->prevBlock = currentBlock;
newBlock->startOfNextRecord = blockStart + sizeof(Block);
newBlock->size = newBlockSize;
currentBlock = newBlock;
}
private:
Block* currentBlock;
};
To answer your question, you're not invoking undefined behaviour since nobody is using the pointer until the object is fully constructed (the pointer value itself is safe to copy around until then). However, it's a rather intrusive method, as the object(s) themselves need to know about the memory pool. Additionally, if you're constructing a large number of small objects, it would likely be faster to use an actual pool of memory (like my pool does) instead of calling out to new for every object.
Whatever pool-like approach you use, be careful that the objects are never manually deleteed, because that would lead to a double free!
I still think this is an interesting question without a definitive reply, but please let me break it down into the different questions you are actually asking:
1.) Does inserting a pointer to a base class into a vector before initialisation of a subclass prevent or cause issues with retrieving inherited classes from that pointer. [slicing for example.]
Answer: No, so long as you are 100% sure of the relevant type that is being pointed to, this mechanism does not cause these issues however note the following points:
If the derived constructor fails, you are left with an issue later when you are likely to have a dangling pointer at least sitting in the vector, as that address space it [the derived class] thought it was getting would be freed to the operating environment on failure, but the vector still has the address as being of the base class type.
Note that a vector, although kind of useful, is not the best structure for this, and even if it was, there should be some inversion of control involved here to allow the vector object to control initialisation of your objects, so that you have awareness of success/failure.
These points lead to the implied 2nd question:
2.) Is this a good pattern for pooling?
Answer: Not really, for the reasons mentioned above, plus others (Pushing a vector past it's end point basically ends up with a malloc which is unnecessary and will impact performance.) Ideally you want to use a pooling library, or a template class, and even better, separate the allocation/de-allocation policy implementation away from the pool implementation, with a low level solution already being hinted at, which is to allocate adequate pool memory from pool initialisation, and then use this using pointers to void from within the pool address space (See Alex Zywicki's solution above.) Using this pattern, the pool destruction is safe as the pool which will be contiguous memory can be destroyed en masse without any dangling issues, or memory leaks through losing all references to an object (losing all reference to an object whose address is allocated through the pool by the storage manager leaves you with dirty chunk/s, but will not cause a memory leak as it is managed by the pool implementation.
In the early days of C/C++ (before mass proliferation of the STL), this was a well discussed pattern and many implementations and designs can be found out there in good literature: As an example:
Knuth (1973 The art of computer programming: Multiple volumes), and for a more complete list, with more on pooling, see:
http://www.ibm.com/developerworks/library/l-memory/
The 3rd implied question seems to be:
3) Is this a valid scenario to use pooling?
Answer: This is a localised design decision based on what you are comfortable with, but to be honest, your implementation (no controlling structure/aggregate, possibly cyclic sharing of sub sets of objects) suggests to me that you would be better off with a basic linked list of wrapper objects, each of which contains a pointer to your superclass, used only for addressing purposes. Your cyclical structures are built on top of this, and you simply amend/grow shrink the list as required to accommodate all of your first class objects as required, and when finished, you can then easily destroy them in effectively an O(1) operation from within the linked list.
Having said that, I would personally recommend that at this time (when you have a scenario where pooling does have a use and so you are in the right mind-set) to carry out the building of a storage management/pooling set of classes that are paramaterised/typeless now as it will hold you in good stead for the future.
This sounds what I have heard called a Linear Allocator.
I will explain the basics of how I understand how it works.
Allocate a block of memory using ::operator new(size);
Have a void* that is your Pointer to the next free space in memory.
You will have an alloc(size_t size) function that will give you a pointer to the location in the block from step one for you to construct on to using Placement New
Placement new looks like... int* i = new(location)int(); where location is a void* to a block of memory you alloced from the allocator.
when you are done with all of your memory you will call a Flush() function that will dealloc the memory from the pool or at least wipe the data clean.
I programmed one of these recently and i will post my code here for you as well as do my best to explain.
#include <iostream>
class LinearAllocator:public ObjectBase
{
public:
LinearAllocator();
LinearAllocator(Pool* pool,size_t size);
~LinearAllocator();
void* Alloc(Size_t size);
void Flush();
private:
void** m_pBlock;
void* m_pHeadFree;
void* m_pEnd;
};
don't worry about what i'm inheriting from. i have been using this allocator in conjunction with a memory pool. but basically instead of getting the memory from operator new i am getting memory from a memory pool. the internal workings are the same essentially.
Here is the implementation:
LinearAllocator::LinearAllocator():ObjectBase::ObjectBase()
{
m_pBlock = nullptr;
m_pHeadFree = nullptr;
m_pEnd=nullptr;
}
LinearAllocator::LinearAllocator(Pool* pool,size_t size):ObjectBase::ObjectBase(pool)
{
if (pool!=nullptr) {
m_pBlock = ObjectBase::AllocFromPool(size);
m_pHeadFree = * m_pBlock;
m_pEnd = (void*)((unsigned char*)*m_pBlock+size);
}
else{
m_pBlock = nullptr;
m_pHeadFree = nullptr;
m_pEnd=nullptr;
}
}
LinearAllocator::~LinearAllocator()
{
if (m_pBlock!=nullptr) {
ObjectBase::FreeFromPool(m_pBlock);
}
m_pBlock = nullptr;
m_pHeadFree = nullptr;
m_pEnd=nullptr;
}
MemoryBlock* LinearAllocator::Alloc(size_t size)
{
if (m_pBlock!=nullptr) {
void* test = (void*)((unsigned char*)m_pEnd-size);
if (m_pHeadFree<=test) {
void* temp = m_pHeadFree;
m_pHeadFree=(void*)((unsigned char*)m_pHeadFree+size);
return temp;
}else{
return nullptr;
}
}else return nullptr;
}
void LinearAllocator::Flush()
{
if (m_pBlock!=nullptr) {
m_pHeadFree=m_pBlock;
size_t size = (unsigned char*)m_pEnd-(unsigned char*)*m_pBlock;
memset(*m_pBlock,0,size);
}
}
This code is fully functional except for a few lines which will need to be changed because of my inheritance and use of the memory pool. but I bet you can figure out what needs to change and just let me know if you need a hand changing the code. This code has not been tested in any sort of professional manor and is not guaranteed to be thread safe or anything fancy like that. i just whipped it up and thought i could share it with you since you seemed to need help.
I also have a working implementation of a fully generic memory pool if you think it may help you. I can explain how it works if you need.
Once again if you need any help let me know. Good luck.
So I have the following code
class UserDB
{
private:
AccountInfo* _accounts[200] ; // store up to 200 accounts
public:
UserDB();
virtual ~UserDB();
}
UserDB::UserDB(){
//code for initializing it to null
}
UserDB::~UserDB(){
delete [] _accounts;
}
So basically I am trying to find this code to initialize _accounts to null but I cannot find a real answer, all the guides in the internet either say how to initialize an array, an object, or a pointer, but not something that is all three altogether, and even less how to initialize this kind of pointer to null, even whatever they are initializing [in the guides] looks very confusing, so I come once again to ask for help here.
Also, AccountInfo is just any random class.
use std::array or std::vector.
you don't delete[] _accounts because the array is a value -- it is an array of pointers. IOW, its size is not equal to a pointer.
Here's a std::vector approach:
class UserDB {
private:
std::vector<AccountInfo*> _accounts;
public:
UserDB() : _accounts(200, 0) {}
virtual ~UserDB() {}
};
However, you may prefer to use the vector's default initializer so you can use it to determine the number of accounts it holds.
Update in response to comments below:
Although there are reasons to hold an array of AccountInfo* pointers, you may also consider std::vector to hold an array of AccountInfos values:
class UserDB {
private:
std::vector<AccountInfo> _accounts;
public:
UserDB() : _accounts() {}
virtual ~UserDB() {}
void appendAccountInfo(const AccountInfo& info) {
this->_accounts.push_back(info);
}
};
std::vector will handle all your allocation and reallocation needs for you. It's also nice because it's dynamically resizable, and you won't be constrained to a fixed number of AccountInfos.
create a constant instead of the magic number 200 (not really necessary but it makes the code more readable and safer when later changing)
const int numberOfAccounts = 200;
AccountInfo* _accounts[numberOfAccounts]
UserDB::UserDB()
{
for (int i = 0; i < numberOfAccounts; ++i)
{
_accounts[i] = 0;
}
}
now you have you 200 zeroed pointers.
also have a habit of putting private members at the end of the class and public at the start,
especially by bigger classes you want to see the public stuff first, the private stuff you normally
don't want somebody to mess with.
class C
{
public:
protected:
private:
};
I do remember having read that this would work:
UserDB():_accounts(){}
This should initialize the contents to NULL