Loads a vector that has been stored in binary or HDF5 format with
vec - the newly loaded vector, this needs to have been created with
VecCreate()or some related function before a call to VecLoad().
viewer - binary file viewer, obtained from PetscViewerBinaryOpen() or HDF5 file viewer, obtained from PetscViewerHDF5Open()
Defaults to the standard Seq or MPI Vec, if you want some other type of Vec call VecSetFromOptions() before calling this.
The input file must contain the full global vector, as written by the routine VecView().
If the type or size of vec is not set before a call to VecLoad, PETSc sets the type and the local and global sizes. If type and/or sizes are already set, then the same are used.
If using the binary viewer and the blocksize of the vector is greater than one then you must provide a unique prefix to the vector with PetscObjectSetOptionsPrefix((PetscObject)vec,”uniqueprefix”); BEFORE calling VecView() on the vector to be stored and then set that same unique prefix on the vector that you pass to VecLoad(). The blocksize information is stored in an ASCII file with the same name as the binary file plus a “.info” appended to the filename. If you copy the binary file, make sure you copy the associated .info file with it.
If using HDF5, you must assign the Vec the same name as was used in the Vec that was stored in the file using PetscObjectSetName(). Otherwise you will get the error message: “Cannot H5DOpen2() with Vec name NAMEOFOBJECT”.
If the HDF5 file contains a two dimensional array the first dimension is treated as the block size in loading the vector. Hence, for example, using Matlab notation h5create(‘vector.dat’,’/Test_Vec’,[27 1]); will load a vector of size 27 and block size 27 thus resulting in all 27 entries being on the first process of vectors communicator and the rest of the processes having zero entries
Notes for advanced users when using the binary viewer#
Most users should not need to know the details of the binary storage format, since VecLoad() and VecView() completely hide these details. But for anyone who’s interested, the standard binary vector storage format is
In addition, PETSc automatically uses byte swapping to work on all machines; the files are written ALWAYS using big-endian ordering. On small-endian machines the numbers are converted to the small-endian format when they are read in from the file. See PetscBinaryRead() and PetscBinaryWrite() to see how this may be done.