Fix NDims during writeDomain

[ax3l]

Fixes the bug for writing fields introduced with Pull #80

fieldTmp (1 component but in a 3D domain living) should have screwed up the other fields a little due to that mismatch...

[ax3l]

Info: still failing, we should consider a possible splash2txt since 422ab7c bug, too

[ax3l]

Actually it looks like I mismatched x and z somewhere image

[ax3l]

Ok, the last bug resides in splash2txt:436

excontainer.container = dc.readDomain( options.step, iter->c_str( ), offset, domain_size, NULL );
reads one element too less per dimension for the slice.

Examples with a total Domain of 128x512x256 writing excontainer.container.getSize( ):

$ splash2txt -s 300 --offset 64 --slice xy -d fields_Density_e -o dens_e_300_new.txt h5
127
511
64

$ splash2txt -s 300 --offset 64 --slice yz -d fields_Density_e -o dens_e_300_new.txt h5                                                            
64
511
255

[ax3l]

I think the problem resides in size[i] = std::max(...) in DataContainer::add :

...
const Dimensions &entryOffset = entry->getOffset();
const Dimensions entryEnd = entry->getEnd();

for (uint32_t i = 0; i < 3; ++i)
{
    offset[i] = std::min(entryOffset[i], offset[i]);
    size[i] = std::max(entryEnd[i] - offset[i], size[i]);
}
...

But Domain::getEnd() returns the last element (e.g. size = 256 -> it returns 255).

That's a little uncommon but fine, but we should change DataContainer::add in that case to:

...
    size[i] = std::max(entryEnd[i] + 1 - offset[i], size[i]);
...

Edit:
Btw: DataContainer::getEnd shows to the first element behind the regular elements in this Container:

Dimensions getEnd() const
{
    return offset + size;
}

That's more STL like but different to the Domain.hpp::getEnd behaviour - what do we like to agree on?

[ax3l]

libSplash fix: see proposed pull request ComputationalRadiationPhysics/libSplash#39
PIConGPU: ready to merge 👯