projections_subsampling does not work in some cases
When loading data from HDF5 file through ChunkReader
, the logic ultimately boils down to the following (thanks to nabu.io.utils.get_compacted_dataslices
):
from tomoscan.io import HDF5File
f = HDF5File("/scratch/paleo/bamboo/bambou_hercules_0001/bambou_hercules_0001.nx", "r")
dptr = f["entry0000/instrument/detector/data"]
data = dptr[81:2281, : , :]
The above example works fine. However, when we want to do a 2-subsampling:
from tomoscan.io import HDF5File
f = HDF5File("/scratch/paleo/bamboo/bambou_hercules_0001/bambou_hercules_0001.nx", "r")
dptr = f["entry0000/instrument/detector/data"]
data = dptr[81:2281:2, : , :] # note the "2" here
then nothing happens (neither in dstat
nor in memory usage), and the program is stuck.
(Tested with h5py 3.2.1 and libhdf5 1.12.0).
The culprit is likely to be HDF5 virtual datasets not supporting such access pattern.