Axel Huebl

Results 1239 comments of Axel Huebl

Meshes: @PrometheusPi has a case in the [PIConGPU radiation plugin](https://github.com/ComputationalRadiationPhysics/picongpu/wiki/Plugin%3A-Radiation) where he wants to store [complex amplitude data](https://github.com/ComputationalRadiationPhysics/picongpu/pull/1053#issuecomment-169282390) as 2D matrix in a [rectilinear grid](https://en.wikipedia.org/wiki/Regular_grid). Also here, an array for...

Excellent, I will put you both on the review of that change then. I would actually remove `dataOrder` completely and describe attributes such as `axisLabel` more precisely. Making it optional...

Careful: [HDF5 Fortran](https://www.hdfgroup.org/hdf5-quest.html#pcmpdf90) support for compound data types is pretty bad. Link moved: [archive](https://web.archive.org/web/20190913180832/https://support.hdfgroup.org/HDF5/hdf5-quest.html#pcmpdf90)

Side note: ADIOS2 has first-class complex support. > To maximize portability, I would vote for the standard mandating complex storage as real and imaginary datasets or amplitude and phase datasets....

Yes, we should also stay by default what [C11](https://en.cppreference.com/w/c/numeric/complex) and [C++](https://en.cppreference.com/w/cpp/numeric/complex) do: a simple POD datatype of two `float`, `double` and `long double` of the order real+complex. We can easily...

@ChristopherMayes did writing a data set and/or attribute with complex numbers in `h5py` and then identifying it as complex and reading it via Fortran - and vice versa - work...

> How would this new standard definition effect ℂ3 datasets as used in PIConGPU? You could then store them as regular record with record components as ℂ. Note to the...

> So how would you word this in the standard? What happens if someone is using a format that does not support complex nor compound types? I would propose to...

Yes, there ways to express complex numbers in HDF5 which are broadly enough supported to be used (aka `h5py`). I'll show you an example implementation with the HDF5 C api...

> How about something like: 1) If a file format supports complex that is to be used. 2) For HDF5, be compatible with h5py and store complex as a struct...