EntityComponentMetaSystem/src/EC/Manager.hpp

2174 lines
83 KiB
C++

// This work derives from Vittorio Romeo's code used for cppcon 2015 licensed
// under the Academic Free License.
// His code is available here: https://github.com/SuperV1234/cppcon2015
#ifndef EC_MANAGER_HPP
#define EC_MANAGER_HPP
#include <chrono>
#define EC_INIT_ENTITIES_SIZE 256
#define EC_GROW_SIZE_AMOUNT 256
#include <array>
#include <atomic>
#include <cstddef>
#include <vector>
#include <deque>
#include <tuple>
#include <utility>
#include <functional>
#include <map>
#include <unordered_map>
#include <set>
#include <unordered_set>
#include <algorithm>
#include <thread>
#include <mutex>
#include <type_traits>
#ifndef NDEBUG
#include <iostream>
#endif
#include "Meta/Combine.hpp"
#include "Meta/Matching.hpp"
#include "Meta/ForEachWithIndex.hpp"
#include "Meta/ForEachDoubleTuple.hpp"
#include "Meta/IndexOf.hpp"
#include "Bitset.hpp"
#include "ThreadPool.hpp"
namespace EC
{
/*!
\brief Manages an EntityComponent system.
EC::Manager must be created with a list of all used Components and all
used tags.
Note that all components must have a default constructor.
An optional third template parameter may be given, which is the size of
the number of threads in the internal ThreadPool, and should be at
least 2. If ThreadCount is 1 or less, then the ThreadPool will not be
created and it will never be used, even if the "true" parameter is given
for functions that enable its usage.
Note that when calling one of the "forMatching" functions that make use
of the internal ThreadPool, it is allowed to call addEntity() or
deleteEntity() as the functions cache which entities are alive before
running (allowing for addEntity()), and the functions defer deletions
during concurrent execution (allowing for deleteEntity()).
Example:
\code{.cpp}
EC::Manager<TypeList<C0, C1, C2>, TypeList<T0, T1>> manager;
\endcode
*/
template <typename ComponentsList,
typename TagsList,
unsigned int ThreadCount = 4>
struct Manager
{
public:
using Components = ComponentsList;
using Tags = TagsList;
using Combined = EC::Meta::Combine<ComponentsList, TagsList>;
using BitsetType = EC::Bitset<ComponentsList, TagsList>;
private:
using ComponentsTuple = EC::Meta::Morph<ComponentsList, std::tuple<> >;
static_assert(std::is_default_constructible<ComponentsTuple>::value,
"All components must be default constructible");
template <typename... Types>
struct Storage
{
using type = std::tuple<std::deque<Types>..., std::deque<char> >;
};
using ComponentsStorage =
typename EC::Meta::Morph<ComponentsList, Storage<> >::type;
// Entity: isAlive, ComponentsTags Info
using EntitiesTupleType = std::tuple<bool, BitsetType>;
using EntitiesType = std::deque<EntitiesTupleType>;
EntitiesType entities;
ComponentsStorage componentsStorage;
std::size_t currentCapacity = 0;
std::size_t currentSize = 0;
std::unordered_set<std::size_t> deletedSet;
std::unique_ptr<ThreadPool<ThreadCount> > threadPool;
std::atomic_uint deferringDeletions;
std::vector<std::size_t> deferredDeletions;
std::mutex deferredDeletionsMutex;
std::vector<std::size_t> idStack;
std::size_t idStackCounter;
std::mutex idStackMutex;
public:
// section for "temporary" structures {{{
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructZero {
std::array<std::size_t, 2> range;
Manager *manager;
EntitiesType *entities;
BitsetType signature;
void *userData;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
template <typename Function>
struct TPFnDataStructOne {
std::array<std::size_t, 2> range;
Manager *manager;
EntitiesType *entities;
BitsetType *signature;
void *userData;
Function *fn;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructTwo {
std::array<std::size_t, 2> range;
Manager *manager;
EntitiesType *entities;
void *userData;
const std::vector<std::size_t> *matching;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructThree {
std::array<std::size_t, 2> range;
Manager *manager;
std::vector<std::vector<std::size_t> > *matchingV;
const std::vector<BitsetType*> *bitsets;
EntitiesType *entities;
std::mutex *mutex;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructFour {
std::array<std::size_t, 2> range;
Manager *manager;
std::vector<std::vector<std::size_t> >*
multiMatchingEntities;
BitsetType *signatures;
std::mutex *mutex;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructFive {
std::array<std::size_t, 2> range;
std::size_t index;
Manager *manager;
void *userData;
std::vector<std::vector<std::size_t> >*
multiMatchingEntities;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
struct TPFnDataStructSix {
std::array<std::size_t, 2> range;
Manager *manager;
std::vector<std::vector<std::size_t> > *
multiMatchingEntities;
BitsetType *bitsets;
std::mutex *mutex;
std::unordered_set<std::size_t> dead;
};
/// Temporary struct used internally by ThreadPool
template <typename Iterable>
struct TPFnDataStructSeven {
std::array<std::size_t, 2> range;
Manager *manager;
EntitiesType *entities;
Iterable *iterable;
void *userData;
std::unordered_set<std::size_t> dead;
};
// end section for "temporary" structures }}}
/*!
\brief Initializes the manager with a default capacity.
The default capacity is set with macro EC_INIT_ENTITIES_SIZE,
and will grow by amounts of EC_GROW_SIZE_AMOUNT when needed.
*/
Manager() :
threadPool{},
idStackCounter(0)
{
resize(EC_INIT_ENTITIES_SIZE);
if(ThreadCount >= 2) {
threadPool = std::make_unique<ThreadPool<ThreadCount> >();
}
deferringDeletions.store(0);
}
~Manager() {
if (threadPool) {
while(!threadPool->isNotRunning()) {
std::this_thread::sleep_for(std::chrono::microseconds(30));
}
}
}
private:
void resize(std::size_t newCapacity)
{
if(currentCapacity >= newCapacity)
{
return;
}
EC::Meta::forEach<ComponentsList>([this, newCapacity] (auto t) {
std::get<std::deque<decltype(t)> >(
this->componentsStorage).resize(newCapacity);
});
entities.resize(newCapacity);
for(std::size_t i = currentCapacity; i < newCapacity; ++i)
{
entities[i] = std::make_tuple(false, BitsetType{});
}
currentCapacity = newCapacity;
}
public:
/*!
\brief Adds an entity to the system, returning the ID of the entity.
Note: The ID of an entity is guaranteed to not change.
*/
std::size_t addEntity()
{
if(deletedSet.empty())
{
if(currentSize == currentCapacity)
{
resize(currentCapacity + EC_GROW_SIZE_AMOUNT);
}
std::get<bool>(entities[currentSize]) = true;
return currentSize++;
}
else
{
std::size_t id;
{
auto iter = deletedSet.begin();
id = *iter;
deletedSet.erase(iter);
}
std::get<bool>(entities[id]) = true;
return id;
}
}
private:
void deleteEntityImpl(std::size_t id) {
if(hasEntity(id)) {
std::get<bool>(entities.at(id)) = false;
std::get<BitsetType>(entities.at(id)).reset();
deletedSet.insert(id);
}
}
public:
/*!
\brief Marks an entity for deletion.
A deleted Entity's id is stored to be reclaimed later when
addEntity is called. Thus calling addEntity may return an id of
a previously deleted Entity.
*/
void deleteEntity(std::size_t index)
{
if(deferringDeletions.load() != 0) {
std::lock_guard<std::mutex> lock(deferredDeletionsMutex);
deferredDeletions.push_back(index);
} else {
deleteEntityImpl(index);
}
}
private:
void handleDeferredDeletions() {
if(deferringDeletions.fetch_sub(1) == 1) {
std::lock_guard<std::mutex> lock(deferredDeletionsMutex);
for(std::size_t id : deferredDeletions) {
deleteEntityImpl(id);
}
deferredDeletions.clear();
}
}
public:
/*!
\brief Checks if the Entity with the given ID is in the system.
Note that deleted Entities are still considered in the system.
Consider using isAlive().
*/
bool hasEntity(const std::size_t& index) const
{
return index < currentSize;
}
/*!
\brief Checks if the Entity is not marked as deleted.
Note that invalid Entities (Entities where calls to hasEntity()
returns false) will return false.
*/
bool isAlive(const std::size_t& index) const
{
return hasEntity(index) && std::get<bool>(entities.at(index));
}
/*!
\brief Returns the current size or number of entities in the system.
Note this function will only count entities where isAlive() returns
true.
*/
std::size_t getCurrentSize() const
{
return currentSize - deletedSet.size();
}
/*
\brief Returns the current capacity or number of entities the system
can hold.
Note that when capacity is exceeded, the capacity is increased by
EC_GROW_SIZE_AMOUNT.
*/
std::size_t getCurrentCapacity() const
{
return currentCapacity;
}
/*!
\brief Returns a const reference to an Entity's info.
An Entity's info is a std::tuple with a bool, and a
bitset.
\n The bool determines if the Entity is alive.
\n The bitset shows what Components and Tags belong to the Entity.
*/
const EntitiesTupleType& getEntityInfo(const std::size_t& index) const
{
return entities.at(index);
}
/*!
\brief Returns a pointer to a component belonging to the given
Entity.
This function will return a pointer to a Component regardless of
whether or not the Entity actually owns the Component. If the Entity
doesn't own the Component, changes to the Component will not affect
any Entity. It is recommended to use hasComponent() to determine if
the Entity actually owns that Component.
If the given Component is unknown to the Manager, then this function
will return a nullptr.
*/
template <typename Component>
Component* getEntityData(const std::size_t& index)
{
constexpr auto componentIndex = EC::Meta::IndexOf<
Component, Components>::value;
if(componentIndex < Components::size)
{
// Cast required due to compiler thinking that an invalid
// Component is needed even though the enclosing if statement
// prevents this from ever happening.
return (Component*) &std::get<componentIndex>(
componentsStorage).at(index);
}
else
{
return nullptr;
}
}
/*!
\brief Returns a pointer to a component belonging to the given
Entity.
Note that this function is the same as getEntityData().
This function will return a pointer to a Component regardless of
whether or not the Entity actually owns the Component. If the Entity
doesn't own the Component, changes to the Component will not affect
any Entity. It is recommended to use hasComponent() to determine if
the Entity actually owns that Component.
If the given Component is unknown to the Manager, then this function
will return a nullptr.
*/
template <typename Component>
Component* getEntityComponent(const std::size_t& index)
{
return getEntityData<Component>(index);
}
/*!
\brief Returns a const pointer to a component belonging to the
given Entity.
This function will return a const pointer to a Component
regardless of whether or not the Entity actually owns the Component.
If the Entity doesn't own the Component, changes to the Component
will not affect any Entity. It is recommended to use hasComponent()
to determine if the Entity actually owns that Component.
If the given Component is unknown to the Manager, then this function
will return a nullptr.
*/
template <typename Component>
const Component* getEntityData(const std::size_t& index) const
{
constexpr auto componentIndex = EC::Meta::IndexOf<
Component, Components>::value;
if(componentIndex < Components::size)
{
// Cast required due to compiler thinking that an invalid
// Component is needed even though the enclosing if statement
// prevents this from ever happening.
return (Component*) &std::get<componentIndex>(
componentsStorage).at(index);
}
else
{
return nullptr;
}
}
/*!
\brief Returns a const pointer to a component belonging to the
given Entity.
Note that this function is the same as getEntityData() (const).
This function will return a const pointer to a Component
regardless of whether or not the Entity actually owns the Component.
If the Entity doesn't own the Component, changes to the Component
will not affect any Entity. It is recommended to use hasComponent()
to determine if the Entity actually owns that Component.
If the given Component is unknown to the Manager, then this function
will return a nullptr.
*/
template <typename Component>
const Component* getEntityComponent(const std::size_t& index) const
{
return getEntityData<Component>(index);
}
/*!
\brief Checks whether or not the given Entity has the given
Component.
Example:
\code{.cpp}
manager.hasComponent<C0>(entityID);
\endcode
*/
template <typename Component>
bool hasComponent(const std::size_t& index) const
{
return std::get<BitsetType>(
entities.at(index)).template getComponentBit<Component>();
}
/*!
\brief Checks whether or not the given Entity has the given Tag.
Example:
\code{.cpp}
manager.hasTag<T0>(entityID);
\endcode
*/
template <typename Tag>
bool hasTag(const std::size_t& index) const
{
return std::get<BitsetType>(
entities.at(index)).template getTagBit<Tag>();
}
/*!
\brief Adds a component to the given Entity.
Additional parameters given to this function will construct the
Component with those parameters.
Note that if the Entity already has the same component, then it
will be overwritten by the newly created Component with the given
arguments.
If the Entity is not alive or the given Component is not known to
the Manager, then nothing will change.
Example:
\code{.cpp}
struct C0
{
// constructor is compatible as a default constructor
C0(int a = 0, char b = 'b') :
a(a), b(b)
{}
int a;
char b;
}
manager.addComponent<C0>(entityID, 10, 'd');
\endcode
*/
template <typename Component, typename... Args>
void addComponent(const std::size_t& entityID, Args&&... args)
{
if(!EC::Meta::Contains<Component, Components>::value
|| !isAlive(entityID))
{
return;
}
Component component(std::forward<Args>(args)...);
std::get<BitsetType>(
entities[entityID]
).template getComponentBit<Component>() = true;
constexpr auto index =
EC::Meta::IndexOf<Component, Components>::value;
// Cast required due to compiler thinking that deque<char> at
// index = Components::size is being used, even if the previous
// if statement will prevent this from ever happening.
(*((std::deque<Component>*)(&std::get<index>(
componentsStorage
))))[entityID] = std::move(component);
}
/*!
\brief Removes the given Component from the given Entity.
If the Entity does not have the Component given, nothing will
change.
Example:
\code{.cpp}
manager.removeComponent<C0>(entityID);
\endcode
*/
template <typename Component>
void removeComponent(const std::size_t& entityID)
{
if(!EC::Meta::Contains<Component, Components>::value
|| !isAlive(entityID))
{
return;
}
std::get<BitsetType>(
entities[entityID]
).template getComponentBit<Component>() = false;
}
/*!
\brief Adds the given Tag to the given Entity.
Example:
\code{.cpp}
manager.addTag<T0>(entityID);
\endcode
*/
template <typename Tag>
void addTag(const std::size_t& entityID)
{
if(!EC::Meta::Contains<Tag, Tags>::value
|| !isAlive(entityID))
{
return;
}
std::get<BitsetType>(
entities[entityID]
).template getTagBit<Tag>() = true;
}
/*!
\brief Removes the given Tag from the given Entity.
If the Entity does not have the Tag given, nothing will change.
Example:
\code{.cpp}
manager.removeTag<T0>(entityID);
\endcode
*/
template <typename Tag>
void removeTag(const std::size_t& entityID)
{
if(!EC::Meta::Contains<Tag, Tags>::value
|| !isAlive(entityID))
{
return;
}
std::get<BitsetType>(
entities[entityID]
).template getTagBit<Tag>() = false;
}
/*!
\brief Resets the Manager, removing all entities.
Some data may persist but will be overwritten when new entities
are added. Thus, do not depend on data to persist after a call to
reset().
*/
void reset()
{
clearForMatchingFunctions();
currentSize = 0;
currentCapacity = 0;
deletedSet.clear();
resize(EC_INIT_ENTITIES_SIZE);
std::lock_guard<std::mutex> lock(deferredDeletionsMutex);
deferringDeletions.store(0);
deferredDeletions.clear();
}
private:
template <typename... Types>
struct ForMatchingSignatureHelper
{
template <typename CType, typename Function>
static void call(
const std::size_t& entityID,
CType& ctype,
Function&& function,
void* userData = nullptr)
{
function(
entityID,
userData,
ctype.template getEntityData<Types>(entityID)...
);
}
template <typename CType, typename Function>
static void callPtr(
const std::size_t& entityID,
CType& ctype,
Function* function,
void* userData = nullptr)
{
(*function)(
entityID,
userData,
ctype.template getEntityData<Types>(entityID)...
);
}
template <typename CType, typename Function>
void callInstance(
const std::size_t& entityID,
CType& ctype,
Function&& function,
void* userData = nullptr) const
{
ForMatchingSignatureHelper<Types...>::call(
entityID,
ctype,
std::forward<Function>(function),
userData);
}
template <typename CType, typename Function>
void callInstancePtr(
const std::size_t& entityID,
CType& ctype,
Function* function,
void* userData = nullptr) const
{
ForMatchingSignatureHelper<Types...>::callPtr(
entityID,
ctype,
function,
userData);
}
};
public:
/*!
\brief Calls the given function on all Entities matching the given
Signature.
The function object given to this function must accept std::size_t
as its first parameter, void* as its second parameter, and Component
pointers for the rest of the parameters. Tags specified in the
Signature are only used as filters and will not be given as a
parameter to the function.
The second parameter is default nullptr and will be passed to the
function call as the second parameter as a means of providing
context (useful when the function is not a lambda function).
The third parameter is default false (not multi-threaded).
Otherwise, if true, then the thread pool will be used to call the
given function in parallel across all entities. Note that
multi-threading is based on splitting the task of calling the
function across sections of entities. Thus if there are only a small
amount of entities in the manager, then using multiple threads may
not have as great of a speed-up.
Example:
\code{.cpp}
Context c; // some class/struct with data
manager.forMatchingSignature<TypeList<C0, C1, T0>>([]
(std::size_t ID,
void* context,
C0* component0, C1* component1)
{
// Lambda function contents here
},
&c, // "Context" object passed to the function
true // enable use of internal ThreadPool
);
\endcode
Note, the ID given to the function is not permanent. An entity's ID
may change when cleanup() is called.
*/
template <typename Signature, typename Function>
void forMatchingSignature(Function&& function,
void* userData = nullptr,
const bool useThreadPool = false)
{
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
using SignatureComponents =
typename EC::Meta::Matching<Signature, ComponentsList>::type;
using Helper =
EC::Meta::Morph<
SignatureComponents,
ForMatchingSignatureHelper<> >;
BitsetType signatureBitset =
BitsetType::template generateBitset<Signature>();
if(!useThreadPool || !threadPool)
{
for(std::size_t i = 0; i < currentSize; ++i)
{
if(!std::get<bool>(entities[i]))
{
continue;
}
if((signatureBitset & std::get<BitsetType>(entities[i]))
== signatureBitset)
{
Helper::call(i, *this,
std::forward<Function>(function), userData);
}
}
}
else
{
std::array<TPFnDataStructZero*, ThreadCount * 2> fnDataAr;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i] = new TPFnDataStructZero{};
fnDataAr[i]->range = {begin, end};
fnDataAr[i]->manager = this;
fnDataAr[i]->entities = &entities;
fnDataAr[i]->signature = signatureBitset;
fnDataAr[i]->userData = userData;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i]->dead.insert(j);
}
}
threadPool->queueFn([&function] (void *ud) {
auto *data = static_cast<TPFnDataStructZero*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
if(((data->signature)
& std::get<BitsetType>(
data->entities->at(i)))
== data->signature) {
Helper::call(i,
*data->manager,
std::forward<Function>(function),
data->userData);
}
}
delete data;
}, fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
/*!
\brief Calls the given function on all Entities matching the given
Signature.
The function pointer given to this function must accept std::size_t
as its first parameter, void* as its second parameter, and
Component pointers for the rest of the parameters. Tags specified in
the Signature are only used as filters and will not be given as a
parameter to the function.
The second parameter is default nullptr and will be passed to the
function call as the second parameter as a means of providing
context (useful when the function is not a lambda function).
The third parameter is default false (not multi-threaded).
Otherwise, if true, then the thread pool will be used to call the
given function in parallel across all entities. Note that
multi-threading is based on splitting the task of calling the
function across sections of entities. Thus if there are only a small
amount of entities in the manager, then using multiple threads may
not have as great of a speed-up.
Example:
\code{.cpp}
Context c; // some class/struct with data
auto function = []
(std::size_t ID,
void* context,
C0* component0, C1* component1)
{
// Lambda function contents here
};
manager.forMatchingSignaturePtr<TypeList<C0, C1, T0>>(
&function, // ptr
&c, // "Context" object passed to the function
true // enable use of ThreadPool
);
\endcode
Note, the ID given to the function is not permanent. An entity's ID
may change when cleanup() is called.
*/
template <typename Signature, typename Function>
void forMatchingSignaturePtr(Function* function,
void* userData = nullptr,
const bool useThreadPool = false)
{
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
using SignatureComponents =
typename EC::Meta::Matching<Signature, ComponentsList>::type;
using Helper =
EC::Meta::Morph<
SignatureComponents,
ForMatchingSignatureHelper<> >;
BitsetType signatureBitset =
BitsetType::template generateBitset<Signature>();
if(!useThreadPool || !threadPool)
{
for(std::size_t i = 0; i < currentSize; ++i)
{
if(!std::get<bool>(entities[i]))
{
continue;
}
if((signatureBitset & std::get<BitsetType>(entities[i]))
== signatureBitset)
{
Helper::callPtr(i, *this, function, userData);
}
}
}
else
{
std::array<TPFnDataStructOne<Function>, ThreadCount * 2> fnDataAr;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].entities = &entities;
fnDataAr[i].signature = &signatureBitset;
fnDataAr[i].userData = userData;
fnDataAr[i].fn = function;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([] (void *ud) {
auto *data = static_cast<TPFnDataStructOne<Function>*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
if(((*data->signature)
& std::get<BitsetType>(
data->entities->at(i)))
== *data->signature) {
Helper::callPtr(i,
*data->manager,
data->fn,
data->userData);
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
private:
std::map<std::size_t, std::tuple<
BitsetType,
void*,
std::function<void(
std::size_t,
std::vector<std::size_t>,
void*)> > >
forMatchingFunctions;
std::size_t functionIndex = 0;
public:
/*!
\brief Stores a function in the manager to be called later.
As an alternative to calling functions directly with
forMatchingSignature(), functions can be stored in the manager to
be called later with callForMatchingFunctions() and
callForMatchingFunction, and removed with
clearForMatchingFunctions() and removeForMatchingFunction().
The syntax for the Function is the same as with
forMatchingSignature().
Note that functions will be called in the same order they are
inserted if called by callForMatchingFunctions() unless the
internal functionIndex counter has wrapped around (is a
std::size_t). Calling clearForMatchingFunctions() will reset this
counter to zero.
Note that the context pointer provided here (default nullptr) will
be provided to the stored function when called.
Example:
\code{.cpp}
manager.addForMatchingFunction<TypeList<C0, C1, T0>>([]
(std::size_t ID,
void* context,
C0* component0, C1* component1)
{
// Lambda function contents here
});
// call all stored functions
manager.callForMatchingFunctions();
// remove all stored functions
manager.clearForMatchingFunctions();
\endcode
\return The index of the function, used for deletion with
removeForMatchingFunction() or filtering with
keepSomeMatchingFunctions() or removeSomeMatchingFunctions(),
or calling with callForMatchingFunction().
*/
template <typename Signature, typename Function>
std::size_t addForMatchingFunction(
Function&& function,
void* userData = nullptr)
{
deferringDeletions.fetch_add(1);
while(forMatchingFunctions.find(functionIndex)
!= forMatchingFunctions.end())
{
++functionIndex;
}
using SignatureComponents =
typename EC::Meta::Matching<Signature, ComponentsList>::type;
using Helper =
EC::Meta::Morph<
SignatureComponents,
ForMatchingSignatureHelper<> >;
Helper helper;
BitsetType signatureBitset =
BitsetType::template generateBitset<Signature>();
forMatchingFunctions.emplace(std::make_pair(
functionIndex,
std::make_tuple(
signatureBitset,
userData,
[function, helper, this]
(const bool useThreadPool,
std::vector<std::size_t> matching,
void* userData)
{
if(!useThreadPool || !threadPool)
{
for(auto eid : matching)
{
if(isAlive(eid))
{
helper.callInstancePtr(
eid, *this, &function, userData);
}
}
}
else
{
std::array<TPFnDataStructTwo, ThreadCount * 2> fnDataAr;
std::size_t s = matching.size() / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = matching.size();
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].entities = &entities;
fnDataAr[i].userData = userData;
fnDataAr[i].matching = &matching;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(matching.at(j))) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([&function, helper] (void* ud) {
auto *data = static_cast<TPFnDataStructTwo*>(ud);
for(std::size_t i = data->range[0];
i < data->range[1];
++i) {
if(data->dead.find(i) == data->dead.end()) {
helper.callInstancePtr(
data->matching->at(i),
*data->manager,
&function,
data->userData);
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
})));
handleDeferredDeletions();
return functionIndex++;
}
private:
std::vector<std::vector<std::size_t> > getMatchingEntities(
std::vector<BitsetType*> bitsets, const bool useThreadPool = false)
{
std::vector<std::vector<std::size_t> > matchingV(bitsets.size());
if(!useThreadPool || !threadPool)
{
for(std::size_t i = 0; i < currentSize; ++i)
{
if(!isAlive(i))
{
continue;
}
for(std::size_t j = 0; j < bitsets.size(); ++j)
{
if(((*bitsets[j]) & std::get<BitsetType>(entities[i]))
== (*bitsets[j]))
{
matchingV[j].push_back(i);
}
}
}
}
else
{
std::array<TPFnDataStructThree, ThreadCount * 2> fnDataAr;
std::size_t s = currentSize / (ThreadCount * 2);
std::mutex mutex;
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].matchingV = &matchingV;
fnDataAr[i].bitsets = &bitsets;
fnDataAr[i].entities = &entities;
fnDataAr[i].mutex = &mutex;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([] (void *ud) {
auto *data = static_cast<TPFnDataStructThree*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
for(std::size_t j = 0; j < data->bitsets->size();
++j) {
if((*data->bitsets->at(j)
& std::get<BitsetType>(
data->entities->at(i)))
== *data->bitsets->at(j)) {
std::lock_guard<std::mutex> lock(
*data->mutex);
data->matchingV->at(j).push_back(i);
}
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
return matchingV;
}
public:
/*!
\brief Call all stored functions.
The first (and only) parameter can be optionally used to enable the
use of the internal ThreadPool to call all stored functions in
parallel. Using the value false (which is the default) will not use
the ThreadPool and run all stored functions sequentially on the main
thread. Note that multi-threading is based on splitting the task of
calling the functions across sections of entities. Thus if there are
only a small amount of entities in the manager, then using multiple
threads may not have as great of a speed-up.
Example:
\code{.cpp}
manager.addForMatchingFunction<TypeList<C0, C1, T0>>([]
(std::size_t ID,
void* context,
C0* component0, C1* component1) {
// Lambda function contents here
});
// call all stored functions
manager.callForMatchingFunctions();
// call all stored functions with ThreadPool enabled
manager.callForMatchingFunctions(true);
// remove all stored functions
manager.clearForMatchingFunctions();
\endcode
*/
void callForMatchingFunctions(const bool useThreadPool = false)
{
deferringDeletions.fetch_add(1);
std::vector<BitsetType*> bitsets;
for(auto iter = forMatchingFunctions.begin();
iter != forMatchingFunctions.end();
++iter)
{
bitsets.push_back(&std::get<BitsetType>(iter->second));
}
std::vector<std::vector<std::size_t> > matching =
getMatchingEntities(bitsets, useThreadPool);
std::size_t i = 0;
for(auto iter = forMatchingFunctions.begin();
iter != forMatchingFunctions.end();
++iter)
{
std::get<2>(iter->second)(
useThreadPool, matching[i++], std::get<1>(iter->second));
}
handleDeferredDeletions();
}
/*!
\brief Call a specific stored function.
The second parameter can be optionally used to enable the use of the
internal ThreadPool to call the stored function in parallel. Using
the value false (which is the default) will not use the ThreadPool
and run the stored function sequentially on the main thread. Note
that multi-threading is based on splitting the task of calling the
functions across sections of entities. Thus if there are only a
small amount of entities in the manager, then using multiple threads
may not have as great of a speed-up.
Example:
\code{.cpp}
std::size_t id =
manager.addForMatchingFunction<TypeList<C0, C1, T0>>(
[] (std::size_t ID, void* context, C0* c0, C1* c1) {
// Lambda function contents here
});
// call the previously added function
manager.callForMatchingFunction(id);
// call the previously added function with ThreadPool enabled
manager.callForMatchingFunction(id, true);
\endcode
\return False if a function with the given id does not exist.
*/
bool callForMatchingFunction(std::size_t id,
const bool useThreadPool = false)
{
auto iter = forMatchingFunctions.find(id);
if(iter == forMatchingFunctions.end())
{
return false;
}
deferringDeletions.fetch_add(1);
std::vector<std::vector<std::size_t> > matching =
getMatchingEntities(std::vector<BitsetType*>{
&std::get<BitsetType>(iter->second)}, useThreadPool);
std::get<2>(iter->second)(
useThreadPool, matching[0], std::get<1>(iter->second));
handleDeferredDeletions();
return true;
}
/*!
\brief Remove all stored functions.
Also resets the index counter of stored functions to 0.
Example:
\code{.cpp}
manager.addForMatchingFunction<TypeList<C0, C1, T0>>([]
(std::size_t ID,
void* context,
C0* component0, C1* component1)
{
// Lambda function contents here
});
// call all stored functions
manager.callForMatchingFunctions();
// remove all stored functions
manager.clearForMatchingFunctions();
\endcode
*/
void clearForMatchingFunctions()
{
forMatchingFunctions.clear();
functionIndex = 0;
}
/*!
\brief Removes a function that has the given id.
\return True if a function was erased.
*/
bool removeForMatchingFunction(std::size_t id)
{
return forMatchingFunctions.erase(id) == 1;
}
/*!
\brief Removes all functions that do not have the index specified
in argument "list".
The given List must be iterable.
This is the only requirement, so a set could also be given.
\return The number of functions deleted.
*/
template <typename List>
std::size_t keepSomeMatchingFunctions(List list)
{
std::size_t deletedCount = 0;
for(auto iter = forMatchingFunctions.begin();
iter != forMatchingFunctions.end();)
{
if(std::find(list.begin(), list.end(), iter->first)
== list.end())
{
iter = forMatchingFunctions.erase(iter);
++deletedCount;
}
else
{
++iter;
}
}
return deletedCount;
}
/*!
\brief Removes all functions that do not have the index specified
in argument "list".
This function allows for passing an initializer list.
\return The number of functions deleted.
*/
std::size_t keepSomeMatchingFunctions(
std::initializer_list<std::size_t> list)
{
return keepSomeMatchingFunctions<decltype(list)>(list);
}
/*!
\brief Removes all functions that do have the index specified
in argument "list".
The given List must be iterable.
This is the only requirement, so a set could also be given.
\return The number of functions deleted.
*/
template <typename List>
std::size_t removeSomeMatchingFunctions(List list)
{
std::size_t deletedCount = 0;
for(auto listIter = list.begin();
listIter != list.end();
++listIter)
{
deletedCount += forMatchingFunctions.erase(*listIter);
}
return deletedCount;
}
/*!
\brief Removes all functions that do have the index specified
in argument "list".
This function allows for passing an initializer list.
\return The number of functions deleted.
*/
std::size_t removeSomeMatchingFunctions(
std::initializer_list<std::size_t> list)
{
return removeSomeMatchingFunctions<decltype(list)>(list);
}
/*!
\brief Sets the context pointer of a stored function
\return True if id is valid and context was updated
*/
bool changeForMatchingFunctionContext(std::size_t id, void* userData)
{
auto f = forMatchingFunctions.find(id);
if(f != forMatchingFunctions.end())
{
std::get<1>(f->second) = userData;
return true;
}
return false;
}
/*!
\brief Call multiple functions with mulitple signatures on all
living entities.
(Living entities as in entities that have not been marked for
deletion.)
This function requires the first template parameter to be a
EC::Meta::TypeList of signatures. Note that a signature is a
EC::Meta::TypeList of components and tags, meaning that SigList
is a TypeList of TypeLists.
The second template parameter can be inferred from the function
parameter which should be a tuple of functions. The function
at any index in the tuple should match with a signature of the
same index in the SigList. Behavior is undefined if there are
less functions than signatures.
See the Unit Test of this function in src/test/ECTest.cpp for
usage examples.
The second parameter (default nullptr) will be provided to every
function call as a void* (context).
The third parameter is default false (not multi-threaded).
Otherwise, if true, then the thread pool will be used to call the
given function in parallel across all entities. Note that
multi-threading is based on splitting the task of calling the
function across sections of entities. Thus if there are only a small
amount of entities in the manager, then using multiple threads may
not have as great of a speed-up.
This function was created for the use case where there are many
entities in the system which can cause multiple calls to
forMatchingSignature to be slow due to the overhead of iterating
through the entire list of entities on each invocation.
This function instead iterates through all entities once,
storing matching entities in a vector of vectors (for each
signature and function pair) and then calling functions with
the matching list of entities.
Note that multi-threaded or not, functions will be called in order
of signatures. The first function signature pair will be called
first, then the second, third, and so on.
If this function is called with more than 1 thread specified, then
the order of entities called is not guaranteed. Otherwise entities
will be called in consecutive order by their ID.
*/
template <typename SigList, typename FTuple>
void forMatchingSignatures(
FTuple fTuple,
void* userData = nullptr,
const bool useThreadPool = false)
{
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
std::vector<std::vector<std::size_t> >
multiMatchingEntities(SigList::size);
BitsetType signatureBitsets[SigList::size];
// generate bitsets for each signature
EC::Meta::forEachWithIndex<SigList>(
[&signatureBitsets] (auto signature, const auto index) {
signatureBitsets[index] =
BitsetType::template generateBitset
<decltype(signature)>();
});
// find and store entities matching signatures
if(!useThreadPool || !threadPool)
{
for(std::size_t eid = 0; eid < currentSize; ++eid)
{
if(!isAlive(eid))
{
continue;
}
for(std::size_t i = 0; i < SigList::size; ++i)
{
if((signatureBitsets[i]
& std::get<BitsetType>(entities[eid]))
== signatureBitsets[i])
{
multiMatchingEntities[i].push_back(eid);
}
}
}
}
else
{
std::array<TPFnDataStructFour, ThreadCount * 2> fnDataAr;
std::mutex mutex;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].multiMatchingEntities = &multiMatchingEntities;
fnDataAr[i].signatures = signatureBitsets;
fnDataAr[i].mutex = &mutex;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([] (void *ud) {
auto *data = static_cast<TPFnDataStructFour*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
for(std::size_t j = 0; j < SigList::size; ++j) {
if((data->signatures[j]
& std::get<BitsetType>(
data->manager->entities[i]))
== data->signatures[j]) {
std::lock_guard<std::mutex> lock(
*data->mutex);
data->multiMatchingEntities->at(j)
.push_back(i);
}
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// call functions on matching entities
EC::Meta::forEachDoubleTuple(
EC::Meta::Morph<SigList, std::tuple<> >{},
fTuple,
[this, &multiMatchingEntities, useThreadPool, &userData]
(auto sig, auto func, auto index)
{
using SignatureComponents =
typename EC::Meta::Matching<
decltype(sig), ComponentsList>::type;
using Helper =
EC::Meta::Morph<
SignatureComponents,
ForMatchingSignatureHelper<> >;
if(!useThreadPool || !threadPool) {
for(const auto& id : multiMatchingEntities[index]) {
if(isAlive(id)) {
Helper::call(id, *this, func, userData);
}
}
} else {
std::array<TPFnDataStructFive, ThreadCount * 2>
fnDataAr;
std::size_t s = multiMatchingEntities[index].size()
/ (ThreadCount * 2);
for(unsigned int i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = multiMatchingEntities[index].size();
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].index = index;
fnDataAr[i].manager = this;
fnDataAr[i].userData = userData;
fnDataAr[i].multiMatchingEntities =
&multiMatchingEntities;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(multiMatchingEntities.at(index).at(j))) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([&func] (void *ud) {
auto *data = static_cast<TPFnDataStructFive*>(ud);
for(std::size_t i = data->range[0];
i < data->range[1]; ++i) {
if(data->dead.find(i) == data->dead.end()) {
Helper::call(
data->multiMatchingEntities
->at(data->index).at(i),
*data->manager,
func,
data->userData);
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
}
);
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
/*!
\brief Call multiple functions with mulitple signatures on all
living entities.
(Living entities as in entities that have not been marked for
deletion.)
Note that this function requires the tuple of functions to hold
pointers to functions, not just functions.
This function requires the first template parameter to be a
EC::Meta::TypeList of signatures. Note that a signature is a
EC::Meta::TypeList of components and tags, meaning that SigList
is a TypeList of TypeLists.
The second template parameter can be inferred from the function
parameter which should be a tuple of functions. The function
at any index in the tuple should match with a signature of the
same index in the SigList. Behavior is undefined if there are
less functions than signatures.
See the Unit Test of this function in src/test/ECTest.cpp for
usage examples.
The second parameter (default nullptr) will be provided to every
function call as a void* (context).
The third parameter is default false (not multi-threaded).
Otherwise, if true, then the thread pool will be used to call the
given function in parallel across all entities. Note that
multi-threading is based on splitting the task of calling the
function across sections of entities. Thus if there are only a small
amount of entities in the manager, then using multiple threads may
not have as great of a speed-up.
This function was created for the use case where there are many
entities in the system which can cause multiple calls to
forMatchingSignature to be slow due to the overhead of iterating
through the entire list of entities on each invocation.
This function instead iterates through all entities once,
storing matching entities in a vector of vectors (for each
signature and function pair) and then calling functions with
the matching list of entities.
Note that multi-threaded or not, functions will be called in order
of signatures. The first function signature pair will be called
first, then the second, third, and so on.
If this function is called with more than 1 thread specified, then
the order of entities called is not guaranteed. Otherwise entities
will be called in consecutive order by their ID.
*/
template <typename SigList, typename FTuple>
void forMatchingSignaturesPtr(FTuple fTuple,
void* userData = nullptr,
const bool useThreadPool = false)
{
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
std::vector<std::vector<std::size_t> > multiMatchingEntities(
SigList::size);
BitsetType signatureBitsets[SigList::size];
// generate bitsets for each signature
EC::Meta::forEachWithIndex<SigList>(
[&signatureBitsets] (auto signature, const auto index) {
signatureBitsets[index] =
BitsetType::template generateBitset
<decltype(signature)>();
});
// find and store entities matching signatures
if(!useThreadPool || !threadPool)
{
for(std::size_t eid = 0; eid < currentSize; ++eid)
{
if(!isAlive(eid))
{
continue;
}
for(std::size_t i = 0; i < SigList::size; ++i)
{
if((signatureBitsets[i]
& std::get<BitsetType>(entities[eid]))
== signatureBitsets[i])
{
multiMatchingEntities[i].push_back(eid);
}
}
}
}
else
{
std::array<TPFnDataStructSix, ThreadCount * 2> fnDataAr;
std::mutex mutex;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].multiMatchingEntities = &multiMatchingEntities;
fnDataAr[i].bitsets = signatureBitsets;
fnDataAr[i].mutex = &mutex;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([] (void *ud) {
auto *data = static_cast<TPFnDataStructSix*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
for(std::size_t j = 0; j < SigList::size; ++j) {
if((data->bitsets[j]
& std::get<BitsetType>(
data->manager->entities[i]))
== data->bitsets[j]) {
std::lock_guard<std::mutex> lock(
*data->mutex);
data->multiMatchingEntities->at(j)
.push_back(i);
}
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// call functions on matching entities
EC::Meta::forEachDoubleTuple(
EC::Meta::Morph<SigList, std::tuple<> >{},
fTuple,
[this, &multiMatchingEntities, useThreadPool, &userData]
(auto sig, auto func, auto index)
{
using SignatureComponents =
typename EC::Meta::Matching<
decltype(sig), ComponentsList>::type;
using Helper =
EC::Meta::Morph<
SignatureComponents,
ForMatchingSignatureHelper<> >;
if(!useThreadPool || !threadPool)
{
for(const auto& id : multiMatchingEntities[index])
{
if(isAlive(id))
{
Helper::callPtr(id, *this, func, userData);
}
}
}
else
{
std::array<TPFnDataStructFive, ThreadCount * 2>
fnDataAr;
std::size_t s = multiMatchingEntities[index].size()
/ (ThreadCount * 2);
for(unsigned int i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = multiMatchingEntities[index].size();
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].index = index;
fnDataAr[i].manager = this;
fnDataAr[i].userData = userData;
fnDataAr[i].multiMatchingEntities =
&multiMatchingEntities;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(multiMatchingEntities.at(index).at(j))) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([&func] (void *ud) {
auto *data = static_cast<TPFnDataStructFive*>(ud);
for(std::size_t i = data->range[0];
i < data->range[1]; ++i) {
if(data->dead.find(i) == data->dead.end()) {
Helper::callPtr(
data->multiMatchingEntities
->at(data->index).at(i),
*data->manager,
func,
data->userData);
}
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
}
);
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
typedef void ForMatchingFn(std::size_t,
Manager<ComponentsList, TagsList>*,
void*);
/*!
\brief A simple version of forMatchingSignature()
This function behaves like forMatchingSignature(), but instead of
providing a function with each requested component as a parameter,
the function receives a pointer to the manager itself, with which to
query component/tag data.
The third parameter can be optionally used to enable the use of the
internal ThreadPool to call the function in parallel. Using the
value false (which is the default) will not use the ThreadPool and
run the function sequentially on all entities on the main thread.
Note that multi-threading is based on splitting the task of calling
the functions across sections of entities. Thus if there are only a
small amount of entities in the manager, then using multiple threads
may not have as great of a speed-up.
*/
template <typename Signature>
void forMatchingSimple(ForMatchingFn fn,
void *userData = nullptr,
const bool useThreadPool = false) {
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
const BitsetType signatureBitset =
BitsetType::template generateBitset<Signature>();
if(!useThreadPool || !threadPool) {
for(std::size_t i = 0; i < currentSize; ++i) {
if(!std::get<bool>(entities[i])) {
continue;
} else if((signatureBitset
& std::get<BitsetType>(entities[i]))
== signatureBitset) {
fn(i, this, userData);
}
}
} else {
std::array<TPFnDataStructZero*, ThreadCount * 2> fnDataAr;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i] = new TPFnDataStructZero{};
fnDataAr[i]->range = {begin, end};
fnDataAr[i]->manager = this;
fnDataAr[i]->entities = &entities;
fnDataAr[i]->signature = signatureBitset;
fnDataAr[i]->userData = userData;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i]->dead.insert(j);
}
}
threadPool->queueFn([&fn] (void *ud) {
auto *data = static_cast<TPFnDataStructZero*>(ud);
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
} else if((data->signature
& std::get<BitsetType>(
data->entities->at(i)))
== data->signature) {
fn(i, data->manager, data->userData);
}
}
delete data;
}, fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
/*!
\brief Similar to forMatchingSimple(), but with a collection of Component/Tag indices
This function works like forMatchingSimple(), but instead of
providing template types that filter out non-matching entities, an
iterable of indices must be provided which correlate to matching
Component/Tag indices. The function given must match the previously
defined typedef of type ForMatchingFn.
The fourth parameter can be optionally used to enable the use of the
internal ThreadPool to call the function in parallel. Using the
value false (which is the default) will not use the ThreadPool and
run the function sequentially on all entities on the main thread.
Note that multi-threading is based on splitting the task of calling
the functions across sections of entities. Thus if there are only a
small amount of entities in the manager, then using multiple threads
may not have as great of a speed-up.
*/
template <typename Iterable>
void forMatchingIterable(Iterable iterable,
ForMatchingFn fn,
void* userData = nullptr,
const bool useThreadPool = false) {
std::size_t current_id;
{
// push to idStack "call stack"
std::lock_guard<std::mutex> lock(idStackMutex);
current_id = idStackCounter++;
idStack.push_back(current_id);
}
deferringDeletions.fetch_add(1);
if(!useThreadPool || !threadPool) {
bool isValid;
for(std::size_t i = 0; i < currentSize; ++i) {
if(!std::get<bool>(entities[i])) {
continue;
}
isValid = true;
for(const auto& integralValue : iterable) {
if(!std::get<BitsetType>(entities[i]).getCombinedBit(
integralValue)) {
isValid = false;
break;
}
}
if(!isValid) { continue; }
fn(i, this, userData);
}
} else {
std::array<TPFnDataStructSeven<Iterable>, ThreadCount * 2>
fnDataAr;
std::size_t s = currentSize / (ThreadCount * 2);
for(std::size_t i = 0; i < ThreadCount * 2; ++i) {
std::size_t begin = s * i;
std::size_t end;
if(i == ThreadCount * 2 - 1) {
end = currentSize;
} else {
end = s * (i + 1);
}
if(begin == end) {
continue;
}
fnDataAr[i].range = {begin, end};
fnDataAr[i].manager = this;
fnDataAr[i].entities = &entities;
fnDataAr[i].iterable = &iterable;
fnDataAr[i].userData = userData;
for(std::size_t j = begin; j < end; ++j) {
if(!isAlive(j)) {
fnDataAr[i].dead.insert(j);
}
}
threadPool->queueFn([&fn] (void *ud) {
auto *data = static_cast<TPFnDataStructSeven<Iterable>*>(ud);
bool isValid;
for(std::size_t i = data->range[0]; i < data->range[1];
++i) {
if(data->dead.find(i) != data->dead.end()) {
continue;
}
isValid = true;
for(const auto& integralValue : *data->iterable) {
if(!std::get<BitsetType>(data->entities->at(i))
.getCombinedBit(integralValue)) {
isValid = false;
break;
}
}
if(!isValid) { continue; }
fn(i, data->manager, data->userData);
}
}, &fnDataAr[i]);
}
threadPool->easyStartAndWait();
}
// pop from idStack "call stack"
do {
{
std::lock_guard<std::mutex> lock(idStackMutex);
if (idStack.back() == current_id) {
idStack.pop_back();
break;
}
}
std::this_thread::sleep_for(std::chrono::microseconds(15));
} while (true);
handleDeferredDeletions();
}
};
}
#endif