/**
* Copies this data and the RingBuffer data from host to device.
*/
void hostToDevice()
{
__startTransaction(__getTransactionEvent());
ringBuffer->hostToDevice();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
GridBuffer<VALUE, DIM1, BORDERVALUE>::hostToDevice();
__setTransactionEvent(__endTransaction() + ev1);
}
/**
* Copies data and additional pointers from host to device.
*/
void hostToDevice()
{
__startTransaction(__getTransactionEvent());
ringDataSizes->hostToDevice();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
ringData->hostToDevice();
__setTransactionEvent(__endTransaction() + ev1);
}
/**
* Copies data and additional pointers from device to host.
*/
void deviceToHost()
{
__startTransaction(__getTransactionEvent());
ringDataSizes->deviceToHost();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
ringData->deviceToHost();
__setTransactionEvent(__endTransaction() + ev1);
}
/**
* Copies this data and the RingBuffer data from device to host.
*/
void deviceToHost()
{
__startTransaction(__getTransactionEvent());
ringBuffer->deviceToHost();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
GridBuffer<VALUE, DIM1, BORDERVALUE>::deviceToHost();
EventTask ev2 = __endTransaction();
__setTransactionEvent(ev1 + ev2);
}
virtual void init()
{
m_state = Init;
__startTransaction(m_initDependency);
m_buffer.bashField(m_exchange);
m_initDependency = __endTransaction();
m_state = WaitForBash;
}
/**
* Resets all internal buffers.
*/
void reset()
{
__startTransaction(__getTransactionEvent());
frames->reset(false);
frames->initialFillBuffer();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
superCells->getDeviceBuffer().setValue(SuperCell<vint_t > ());
superCells->getHostBuffer().setValue(SuperCell<vint_t > ());
/*nextFrames->getDeviceBuffer().setValue(INV_IDX);//!\todo: is this needed? On device we set any new frame values to INVALID_INDEX
prevFrames->getDeviceBuffer().setValue(INV_IDX);//!\todo: is this needed? On device we set any new frame values to INVALID_INDEX
nextFrames->getHostBuffer().setValue(INV_IDX);//!\todo: is this needed? On device we set any new frame values to INVALID_INDEX
prevFrames->getHostBuffer().setValue(INV_IDX);//!\todo: is this needed? On device we set any new frame values to INVALID_INDEX
*/
__setTransactionEvent(__endTransaction() + ev1);
}
virtual void init()
{
state = Init;
__startTransaction(initDependency);
buffer.bashField(exchange);
initDependency = __endTransaction();
state = WaitForBash;
}
void setCurrentSize(const size_t size)
{
// do host and device setCurrentSize parallel
EventTask split = __getTransactionEvent();
__startTransaction(split);
stackIndexer.getHostBuffer().setCurrentSize(size);
stack.getHostBuffer().setCurrentSize(size);
EventTask e1 = __endTransaction();
__startTransaction(split);
stackIndexer.getDeviceBuffer().setCurrentSize(size);
EventTask e2 = __endTransaction();
__startTransaction(split);
stack.getDeviceBuffer().setCurrentSize(size);
EventTask e3 = __endTransaction();
__setTransactionEvent(e1 + e2 + e3);
}
virtual void init()
{
__startTransaction();
this->request = Environment<DIM>::get().EnvironmentController()
.getCommunicator().startSend(
exchange->getExchangeType(),
(char*) exchange->getHostBuffer().getPointer(),
exchange->getHostBuffer().getCurrentSize() * sizeof (TYPE),
exchange->getCommunicationTag());
__endTransaction();
}
/**
* Starts copying data from device to host.
*/
void deviceToHost()
{
__startTransaction(__getTransactionEvent());
frames->deviceToHost();
EventTask ev1 = __endTransaction();
__startTransaction(__getTransactionEvent());
superCells->deviceToHost();
EventTask ev2 = __endTransaction();
__startTransaction(__getTransactionEvent());
nextFrames->deviceToHost();
EventTask ev3 = __endTransaction();
__startTransaction(__getTransactionEvent());
prevFrames->deviceToHost();
EventTask ev4 = __endTransaction();
__setTransactionEvent(ev1 + ev2 + ev3 + ev4);
}
bool executeIntern()
{
switch (state)
{
case Init:
break;
case WaitForReceive:
if (NULL == Environment<>::get().Manager().getITaskIfNotFinished(lastReceiveEvent.getTaskId()))
{
state = InitInsert;
//bash is finished
__startTransaction();
lastSize = parBase.getParticlesBuffer().getReceiveExchangeStack(exchange).getHostParticlesCurrentSize();
parBase.insertParticles(exchange);
// std::cout<<"brecv = "<<parBase.getParticlesBuffer().getReceiveExchangeStack(exchange).getHostCurrentSize()<<std::endl;
tmpEvent = __endTransaction();
state = WaitForInsert;
}
break;
case InitInsert:
break;
case WaitForInsert:
if (NULL == Environment<>::get().Manager().getITaskIfNotFinished(tmpEvent.getTaskId()))
{
state=Wait;
assert(lastSize <= maxSize);
//check for next bash round
if (lastSize == maxSize)
{
std::cerr << "recv max size " << maxSize << " particles" << std::endl;
init(); //call init and run a full send cycle
}
else
{
state = Finished;
return true;
}
}
break;
case Wait:
break;
case Finished:
return true;
default:
return false;
}
return false;
}
void shiftParticles()
{
StrideMapping<AREA, DIM3, MappingDesc> mapper(this->cellDescription);
ParticlesBoxType pBox = particlesBuffer->getDeviceParticleBox();
__startTransaction(__getTransactionEvent());
do
{
__cudaKernel(kernelShiftParticles)
(mapper.getGridDim(), TileSize)
(pBox, mapper);
}
while (mapper.next());
__setTransactionEvent(__endTransaction());
}
HINLINE void operator()(
T_StorageTuple& tuple,
const uint32_t currentStep,
const T_Event eventInt,
T_Event& updateEvent,
T_Event& commEvent
) const
{
typedef typename HasFlag<FrameType, particlePusher<> >::type hasPusher;
if (hasPusher::value)
{
PMACC_AUTO(speciesPtr, tuple[SpeciesName()]);
__startTransaction(eventInt);
speciesPtr->update(currentStep);
commEvent += speciesPtr->asyncCommunication(__getTransactionEvent());
updateEvent += __endTransaction();
}
}
bool executeIntern()
{
switch (state)
{
case InitDone:
break;
case DeviceToHostFinished:
state = SendDone;
__startTransaction();
Environment<>::get().Factory().createTaskSendMPI(exchange, this);
__endTransaction(); //we need no blocking because we get a singnal if transaction is finished
break;
case SendDone:
break;
case Finish:
return true;
default:
return false;
}
return false;
}
virtual void init()
{
__startTransaction();
state = InitDone;
if (exchange->hasDeviceDoubleBuffer())
{
Environment<>::get().Factory().createTaskCopyDeviceToDevice(exchange->getDeviceBuffer(),
exchange->getDeviceDoubleBuffer()
);
copyEvent = Environment<>::get().Factory().createTaskCopyDeviceToHost(exchange->getDeviceDoubleBuffer(),
exchange->getHostBuffer(),
this);
}
else
{
copyEvent = Environment<>::get().Factory().createTaskCopyDeviceToHost(exchange->getDeviceBuffer(),
exchange->getHostBuffer(),
this);
}
__endTransaction(); //we need no blocking because we get a singnal if transaction is finished
}
bool executeIntern()
{
switch (state)
{
case WaitForReceived:
break;
case RunCopy:
state = WaitForFinish;
__startTransaction();
exchange->getHostBuffer().setCurrentSize(newBufferSize);
if (exchange->hasDeviceDoubleBuffer())
{
Environment<>::get().Factory().createTaskCopyHostToDevice(exchange->getHostBuffer(),
exchange->getDeviceDoubleBuffer());
Environment<>::get().Factory().createTaskCopyDeviceToDevice(exchange->getDeviceDoubleBuffer(),
exchange->getDeviceBuffer(),
this);
}
else
{
Environment<>::get().Factory().createTaskCopyHostToDevice(exchange->getHostBuffer(),
exchange->getDeviceBuffer(),
this);
}
__endTransaction();
break;
case WaitForFinish:
break;
case Finish:
return true;
default:
return false;
}
return false;
}
void event(id_t, EventType type, IEventData* data)
{
switch (type)
{
case RECVFINISHED:
if (data != NULL)
{
__startTransaction(); //no blocking
EventDataReceive *rdata = static_cast<EventDataReceive*> (data);
// std::cout<<" data rec "<<rdata->getReceivedCount()/sizeof(TYPE)<<std::endl;
newBufferSize = rdata->getReceivedCount() / sizeof (TYPE);
__endTransaction();
state = RunCopy;
executeIntern();
}
break;
case COPYHOST2DEVICE:
case COPYDEVICE2DEVICE:
state = Finish;
break;
default:
return;
}
}
