void getLineSliceFields()
{
typedef typename MappingDesc::SuperCellSize SuperCellSize;
const float3_X tmpFloat3(float3_X(float_X(0.0), float_X(0.0), float_X(0.0)));
sliceDataField->getDeviceBuffer().setValue(tmpFloat3);
dim3 block(SuperCellSize::toRT().toDim3());
const SubGrid<simDim>& subGrid = Environment<simDim>::get().SubGrid();
// global cell id offset (without guardings!)
// returns the global id offset of the "first" border cell on a GPU
const DataSpace<simDim> globalCellIdOffset(subGrid.getLocalDomain().offset);
// global number of cells for whole simulation: local cells on GPU * GPUs
// (assumed same size on each gpu :-/ -> todo: provide interface!)
//! \todo create a function for: global number of cells for whole simulation
//!
const DataSpace<simDim> localNrOfCells(subGrid.getLocalDomain().size);
const DataSpace<simDim> globalNrOfCells (subGrid.getGlobalDomain().size);
__picKernelArea(kernelLineSliceFields, *cellDescription, AREA)
(block)
(fieldE->getDeviceDataBox(),
fieldB->getDeviceDataBox(),
sliceDataField->getDeviceBuffer().getBasePointer(),
globalCellIdOffset,
globalNrOfCells
);
sliceDataField->deviceToHost();
//return sliceDataField->getHostBuffer().getDataBox()[0];
}
void update_afterCurrent(uint32_t currentStep)
{
FieldManipulator::absorbBorder(currentStep,this->cellDescription, this->fieldE->getDeviceDataBox());
if (laserProfile::INIT_TIME > float_X(0.0))
fieldE->laserManipulation(currentStep);
EventTask eRfieldE = fieldE->asyncCommunication(__getTransactionEvent());
updateBHalf < CORE> ();
__setTransactionEvent(eRfieldE);
updateBHalf < BORDER > ();
FieldManipulator::absorbBorder(currentStep,this->cellDescription, fieldB->getDeviceDataBox());
EventTask eRfieldB = fieldB->asyncCommunication(__getTransactionEvent());
__setTransactionEvent(eRfieldB);
}
void update_beforeCurrent(uint32_t currentStep)
{
updateBHalf < CORE+BORDER >();
EventTask eRfieldB = fieldB->asyncCommunication(__getTransactionEvent());
updateE<CORE>();
__setTransactionEvent(eRfieldB);
updateE<BORDER>();
}
void createImage(uint32_t currentStep, VirtualWindow window)
{
DataConnector &dc = DataConnector::getInstance();
// Data does not need to be synchronized as visualization is
// done at the device.
FieldB *fieldB = &(dc.getData<FieldB > (FieldB::getName(), true));
FieldE* fieldE = &(dc.getData<FieldE > (FieldE::getName(), true));
FieldJ* fieldJ = &(dc.getData<FieldJ > (FieldJ::getName(), true));
ParticlesType* particles = &(dc.getData<ParticlesType > (particleTag, true));
PMACC_AUTO(simBox, SubGrid<simDim>::getInstance().getSimulationBox());
uint32_t globalOffset = 0;
#if(SIMDIM==DIM3)
globalOffset = SubGrid<simDim>::getInstance().getSimulationBox().getGlobalOffset()[sliceDim];
#endif
typedef MappingDesc::SuperCellSize SuperCellSize;
assert(cellDescription != NULL);
//create image fields
__picKernelArea((kernelPaintFields), *cellDescription, CORE + BORDER)
(SuperCellSize::getDataSpace())
(fieldE->getDeviceDataBox(),
fieldB->getDeviceDataBox(),
fieldJ->getDeviceDataBox(),
img->getDeviceBuffer().getDataBox(),
transpose,
sliceOffset,
globalOffset, sliceDim
);
// find maximum for img.x()/y and z and return it as float3_X
int elements = img->getGridLayout().getDataSpace().productOfComponents();
//Add one dimension access to 2d DataBox
typedef DataBoxDim1Access<typename GridBuffer<float3_X, DIM2 >::DataBoxType> D1Box;
D1Box d1access(img->getDeviceBuffer().getDataBox(), img->getGridLayout().getDataSpace());
#if (EM_FIELD_SCALE_CHANNEL1 == -1 || EM_FIELD_SCALE_CHANNEL2 == -1 || EM_FIELD_SCALE_CHANNEL3 == -1)
//reduce with functor max
float3_X max = reduce(nvidia::functors::Max(),
d1access,
elements);
//reduce with functor min
//float3_X min = reduce(nvidia::functors::Min(),
// d1access,
// elements);
#if (EM_FIELD_SCALE_CHANNEL1 != -1 )
max.x() = float_X(1.0);
#endif
#if (EM_FIELD_SCALE_CHANNEL2 != -1 )
max.y() = float_X(1.0);
#endif
#if (EM_FIELD_SCALE_CHANNEL3 != -1 )
max.z() = float_X(1.0);
#endif
//We don't know the superCellSize at compile time
// (because of the runtime dimension selection in any analyser),
// thus we must use a one dimension kernel and no mapper
__cudaKernel(vis_kernels::divideAnyCell)(ceil((double) elements / 256), 256)(d1access, elements, max);
#endif
// convert channels to RGB
__cudaKernel(vis_kernels::channelsToRGB)(ceil((double) elements / 256), 256)(d1access, elements);
// add density color channel
DataSpace<simDim> blockSize(MappingDesc::SuperCellSize::getDataSpace());
DataSpace<DIM2> blockSize2D(blockSize[transpose.x()], blockSize[transpose.y()]);
//create image particles
__picKernelArea((kernelPaintParticles3D), *cellDescription, CORE + BORDER)
(SuperCellSize::getDataSpace(), blockSize2D.productOfComponents() * sizeof (int))
(particles->getDeviceParticlesBox(),
img->getDeviceBuffer().getDataBox(),
transpose,
sliceOffset,
globalOffset, sliceDim
);
// send the RGB image back to host
img->deviceToHost();
header.update(*cellDescription, window, transpose, currentStep);
__getTransactionEvent().waitForFinished(); //wait for copy picture
DataSpace<DIM2> size = img->getGridLayout().getDataSpace();
PMACC_AUTO(hostBox, img->getHostBuffer().getDataBox());
if (picongpu::white_box_per_GPU)
{
hostBox[0 ][0 ] = float3_X(1.0, 1.0, 1.0);
hostBox[size.y() - 1 ][0 ] = float3_X(1.0, 1.0, 1.0);
hostBox[0 ][size.x() - 1] = float3_X(1.0, 1.0, 1.0);
hostBox[size.y() - 1 ][size.x() - 1] = float3_X(1.0, 1.0, 1.0);
}
PMACC_AUTO(resultBox, gather(hostBox, header));
if (isMaster)
{
output(resultBox.shift(header.window.offset), header.window.size, header);
}
}