Chunk in HDF5 can be seen as the re-arrangement of data in the memory. So still there is only one .h5 file but the way data is stored in disk is chunked as sort of "hypercubic" style.
Here is a great article about this. http://geology.beer/2015/02/10/hdf-for-large-arrays/ 1. For Fortran, HDF5 is too easy to implement, even for MPI-IO. (refer to your own code)
2. For C++, serial I/O is easy. Correct way to use HDF5-C/C++ is to use a C function of HDF5 together mixed with C++. For C++ wrapper, things are complicated and not easy to deal with. In you main.cpp: .... extern "C" { #include "hdf5write.h" } .... In you hdf5write.c everything is C. Never use passing by reference. Useful commands:
|
AuthorShaowu Pan Archives
December 2017
Categories
All
|