/**
* destructor
*/
virtual ~MappedBufferIntern()
{
__startOperation(ITask::TASK_CUDA);
__startOperation(ITask::TASK_HOST);
if (pointer && ownPointer)
{
#if( PMACC_CUDA_ENABLED == 1 )
/* cupla 0.1.0 does not support the function cudaHostAlloc to create mapped memory.
* Therefore we need to call the native CUDA function cudaFreeHost to free memory.
* Due to the renaming of cuda functions with cupla via macros we need to remove
* the renaming to get access to the native cuda function.
* @todo this is a workaround please fix me. We need to investigate if
* it is possible to have mapped/unified memory in alpaka.
*
* corresponding alpaka issues:
* https://github.com/ComputationalRadiationPhysics/alpaka/issues/296
* https://github.com/ComputationalRadiationPhysics/alpaka/issues/612
*/
# undef cudaFreeHost
CUDA_CHECK((cuplaError_t)cudaFreeHost(pointer));
// re-introduce the cupla macro
# define cudaFreeHost(...) cuplaFreeHost(__VA_ARGS__)
#else
__deleteArray(pointer);
#endif
}
}
void reset(bool preserveData = true)
{
__startOperation(ITask::TASK_HOST);
this->setCurrentSize(this->getDataSpace().productOfComponents());
if (!preserveData)
memset(pointer, 0, this->getDataSpace().productOfComponents() * sizeof (TYPE));
}
void reset(bool preserveData = true)
{
__startOperation(ITask::TASK_HOST);
setCurrentSize(this->getDataSpace().getElementCount());
if (!preserveData)
memset(pointer, 0, this->getDataSpace().getElementCount() * sizeof (TYPE));
}
DataBoxType getDataBox()
{
__startOperation(ITask::TASK_CUDA);
TYPE* dPointer;
cudaHostGetDevicePointer(&dPointer, pointer, 0);
return DataBoxType(PitchedBox<TYPE, DIM > (dPointer, DataSpace<DIM > (),
this->data_space, this->data_space[0] * sizeof (TYPE)));
}
void setValue(const TYPE& value)
{
__startOperation(ITask::TASK_HOST);
size_t current_size = this->getCurrentSize();
for (size_t i = 0; i < current_size; i++)
{
pointer[i] = value;
}
}
const cudaPitchedPtr getCudaPitched() const
{
__startOperation(ITask::TASK_CUDA);
TYPE* dPointer;
cudaHostGetDevicePointer(&dPointer, pointer, 0);
/* on 1D memory we have no size for y, therefore we set y to 1 to
* get a valid cudaPitchedPtr
*/
int size_y=1;
if(DIM>DIM1)
size_y= this->data_space[1];
return make_cudaPitchedPtr(dPointer,
this->data_space.x() * sizeof (TYPE),
this->data_space.x(),
size_y
);
}
/**
* Returns the current size pointer.
*
* @return pointer to current size
*/
virtual size_t* getCurrentSizePointer()
{
__startOperation(ITask::TASK_HOST);
return this->current_size;
}
/*! sets the current size (count of elements)
* @param newsize new current size
*/
virtual void setCurrentSize(const size_t newsize)
{
__startOperation(ITask::TASK_HOST);
assert(static_cast<size_t>(newsize) <= static_cast<size_t>(data_space.productOfComponents()));
*current_size = newsize;
}
/*! returns the current size (count of elements)
* @return current size
*/
virtual size_t getCurrentSize()
{
__startOperation(ITask::TASK_HOST);
return *current_size;
}
DataBoxType getHostDataBox()
{
__startOperation(ITask::TASK_HOST);
return DataBoxType(PitchedBox<TYPE, DIM > (pointer, DataSpace<DIM > (),
this->data_space, this->data_space[0] * sizeof (TYPE)));
}
/*! Get device pointer of memory
*
* This pointer is shifted by the offset, if this buffer points to other
* existing buffer
*
* @return device pointer to memory
*/
TYPE* getPointer()
{
__startOperation(ITask::TASK_HOST);
return (TYPE*) this->getCudaPitched().ptr;
}
TYPE* getPointer()
{
__startOperation(ITask::TASK_HOST);
return pointer;
}
/*! sets the current size (count of elements)
* @param newsize new current size
*/
virtual void setCurrentSize(size_t newsize)
{
__startOperation(ITask::TASK_HOST);
assert(static_cast<size_t>(newsize) <= static_cast<size_t>(data_space.getElementCount()));
*current_size = newsize;
}
