Use hierarchical groups in libSplash

[f-schmitt]

Use hierarchical groups to better distinguish particles, fields, ghost cells, indices

Current idea:
fileset 1 (simulation data):

/data/timestamp/...

fields/FieldB/x
fields/FieldB/y

particles/electrons/momentum
particles/electrons/positions
particles/ions/

fileset 2 (simulation restart data):

as above + 
fields/FieldB/_ghosts/top/x
fields/FieldB/_ghosts/bottom/x

[f-schmitt]

Will be integrated with or after @psychocoderHPC changes

[f-schmitt]

The group level shouldn't get too deep so it does not affect performance.
Especially h5* tools such as h5list seem to have problems with very deep hierarchies.
I propose the following internal hierarchy below /data/timestamp/:

fields/FieldB/x
fields/FieldB/y
fields/FieldB/ghosts/x
fields/FieldB/ghosts/y

particles/electrons/momentum
particles/electrons/positions
particles/electrons/ghosts/momentum

particles/ions/

[ax3l]

👍 that would be perfect.

We always have to keep in mind, that a user just wants to skip the ghosts at all during post-processing (or he/she will ask for them explicitly).
The ghost group will have "holes" in the domain, will that cause huge problems?

[f-schmitt]

I am pretty sure this will trigger some problems, yes. This would be even weird for writing the data, since you would try to combine data from top and bottom gpus (?) into one dataset. Maybe we should just split this once more time in top and bottom _ghosts ?

[psychocoderHPC]

Do we need ghosts if we implement #114? Than we write the sliding window data and the restart data in a own file.
I am not sure if a physicist need the ghosts outside of the sliding window. But for our implementation all would be more easy without explicit writing out ghosts.

[ax3l]

@psychocoderHPC is probably right.

Else, holes could be interpreted as zero-value data transparently, but maybe that is harder to achieve than dividing restart hdf5 and output hdf5 in #114 .

[f-schmitt]

+1, let's write to files so the poor physicists are not confused :)
I updated the issue description.

[ax3l]

yeah, let us write to files :P

[f-schmitt]

so it will look like this

[f-schmitt]

I agree and we can drop ghosts and write them to a seperate file later on. For now, let's just move over the existing code.

[psychocoderHPC]

Cool 👍

[ax3l]

Nice! 👍