output_vtkhdf.h

VTKHDF File Format

VTKHDF File Format

These routines are compatible with the VTKHDF Model and Format which can be read using Paraview or Visit.

Data is composed of large datasets stored using the Hierarchical Data Format HDF5. As the name implies, data is organized following a hierarchical structure. HDF5 files can be read without any prior knowledge of the stored data. The type, rank, dimension and other properties of each array are stored inside the file in the form of meta-data. Additional features include support for a large number of objects, file compression, a parallel I/O implementation through the MPI-IO or MPI POSIX drivers. Using this format requires the HDF5 library, which is usually installed in most computing centers or may be installed locally through a repository. Linking is automatic but requires the environment variables HDF5_INCDIR and HDF5_LIBDIR, which are usually set when you load the module hdf5. You might also add something like this to your Makefile

    HDFLIBS=-I$(HDF5_INCDIR) -L$(HDF5_LIBDIR) -lhdf5 -lhdf5_hl
    LIBS+=$(HDFLIBS)

To use HDF5 we need to “declare” the library in the qcc system. This should not be the case, but HDF5 is a messy library that just won’t pass through qcc

    echo "@include <hdf5.h>" > $BASILISK/ast/std/hdf5.h
    echo "@include <hdf5_hl.h>" > $BASILISK/ast/std/hdf5_hl.h

We also need to declare the datatypes at the end of $BASILISK/ast/defaults.h

    echo "typedef hid_t, hsize_t, herr_t, H5L_info_t;" >> $BASILISK/ast/defaults.h

An explanation can be found here

#pragma autolink -lhdf5
#include <hdf5.h>

preamble: define some useful macros

#define shortcut_slice(n, _alpha)                                \
  double alpha = (_alpha - n.x * x - n.y * y - n.z * z) / Delta; \
  if (fabs(alpha) > 0.87)                                        \
    continue;

#include "output_vtkhdf_helpers.h"

output_vtkhdf(): Exports full 2D (or 3D) fields.

This function writes one VTKHDF file which can be read using Paraview. The VTKHDF file format is a file format relying on HDF5. It is meant to provide good I/O performance as well as robust and flexible parallel I/O capabilities. The file stores scalar and vector fields defined at the center points are stored at the cell center of an unstructured grid with type VTK_QUAD in 2D, or VTK_HEXAHEDRON in 3D.

A quick description of the file format is copied from here. VTK HDF files start with a group called VTKHDF with two attributes: Version, an array of two integers and Type, a string showing the VTK dataset type stored in the file — in our case UnstructuredGrid. The data type for each HDF dataset is part of the dataset and it is determined at write time. In the diagram below, showing the HDF file structure for UnstructuredGrid, the rounded blue rectangles are HDF groups and the gray rectangles are HDF datasets. Each rectangle shows the name of the group or dataset in bold font and the attributes underneath with regular font. In our case, scalar and vector fields are stored as CellData. The unstructured grid is split into partitions, with a partition for each MPI rank.


Image data VTKHDF File Format	Linear cell types found in VTK

The arguments and their default values are:

slist : Pointer to an array of scalar data.
vlist : Pointer to an array of vector data.
name : Output file name generally uses the .vtkhdf extension.
compression_level : Level of compression to use when writing data to the HDF5 file (default=9).

Example Usage

scalar * slist = {a,b};
vector * vlist = {c,d};
output_vtkhdf(slist, vlist, "domain.vtkhdf");

see, also example.

trace void output_vtkhdf(scalar *slist, vector *vlist, char *name = "domain.vtkhdf", int compression_level = 9){
  hid_t acc_tpl1;    // File access template
  hid_t file_id;     // HDF5 file ID
  hid_t group_id;    // HDF5 group ID
  hid_t subgroup_id; // HDF5 subgroup ID
  hsize_t count[2];  // Hyperslab selection parameters
  hsize_t offset[2]; // Offset for hyperslab
  hsize_t dims[1] = {2};
  
  // Define a scalar mask for periodic conditions
  scalar per_mask[];
  foreach () {
    per_mask[] = 1.;
  }

  // VTK cell types: VTK_QUAD (in 2D) or VTK_HEXAHEDRON (in 3D)
  int type, noffset;
  #if dimension == 2
    type = 9;
    noffset = 4;
  #elif dimension == 3
    type = 12;
    noffset = 8;
  #endif

  // Obtain the number of points and cells and get a marker to reconstruct the topology
  int number_of_points = 0, number_of_points_loc = 0;
  int number_of_cells = 0,  number_of_cells_loc = 0;
  
  count_points_and_cells(&number_of_points, &number_of_cells, &number_of_points_loc, &number_of_cells_loc, per_mask);
  
  int number_of_ids = number_of_cells*noffset;
  int number_of_ids_loc = number_of_cells_loc*noffset;

  // Define chunk size for parallel I/O
  hsize_t chunk_size = number_of_cells / npe() / 8;

  // Calculate offsets for parallel I/O
  int offset_points[npe()], offset_cells[npe()], offset_ids[npe()], offset_offset[npe()];
  calculate_offsets2(offset_offset, number_of_cells_loc+1,  offset);
  calculate_offsets2(offset_ids,    number_of_ids_loc,      offset);
  calculate_offsets2(offset_cells,  number_of_cells_loc,    offset);
  calculate_offsets2(offset_points, number_of_points_loc,   offset);

  // Initialize marker for topology reconstruction
  vertex scalar marker[];
  initialize_marker(marker, offset, 0);
  
  // Setup file access template with parallel I/O access
  acc_tpl1 = H5Pcreate(H5P_FILE_ACCESS);
  H5Pset_fapl_mpio(acc_tpl1, MPI_COMM_WORLD, MPI_INFO_NULL);

  // Create a new HDF5 file collectively
  file_id = H5Fcreate(name, H5F_ACC_TRUNC, H5P_DEFAULT, acc_tpl1);

  // Release file-access template
  H5Pclose(acc_tpl1);
  
  // Create group 
  group_id = H5Gcreate(file_id, "VTKHDF", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);

  // Create "Version", "Type" (and other) attributes
  dims[0] = 2;
  int version_data[2] = {2, 1};
  create_attribute(group_id, "Version", version_data, dims);
  create_attribute_type(group_id, "Type", "UnstructuredGrid", 16);
  create_attribute_type(group_id, "Description", "Simulation perfomed using Basilisk (http://basilisk.fr/)", 57);
  
  // Write "NumberOfConnectivityIds", "NumberOfPoints", "NumberOfCells"
  dims[0] = npe();
  write_simple_dataset(group_id, "NumberOfConnectivityIds", offset_ids, dims);
  write_simple_dataset(group_id, "NumberOfPoints", offset_points, dims);
  write_simple_dataset(group_id, "NumberOfCells", offset_cells, dims);

  // Populate and write the points dataset
  double *points_dset;
  populate_points_dset(&points_dset, number_of_points_loc, offset_points, count, offset);
  create_chunked_dataset(group_id, count, offset, "Points", number_of_points, number_of_points_loc, 3, points_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  free(points_dset);

  // Populate and write the types dataset
  char * types_dset;
  populate_types_dset(&types_dset, type, number_of_cells_loc, offset_cells, count, offset);
  create_chunked_dataset(group_id, count, offset, "Types", number_of_cells, number_of_cells_loc, 1, types_dset, H5T_STD_U8LE, chunk_size, compression_level);
  free(types_dset);  

  // Populate and write the connectivity dataset
  long *topo_dset;
  populate_topo_dset(&topo_dset, number_of_cells_loc, offset_ids, count, offset, per_mask, marker);
  create_chunked_dataset(group_id, count, offset, "Connectivity", number_of_ids, number_of_ids_loc, 1, topo_dset, H5T_NATIVE_LONG, chunk_size, compression_level);
  free(topo_dset);  

  // Populate and write the offsets dataset
  long *offsets_dset;
  populate_offsets_dset(&offsets_dset, noffset, number_of_cells_loc+1, offset_offset, count, offset);
  create_chunked_dataset(group_id, count, offset, "Offsets", number_of_cells+npe(), number_of_cells_loc+1, 1, offsets_dset, H5T_NATIVE_LONG, chunk_size, compression_level);
  free(offsets_dset);  
  
  // Create subgroup "CellData"
  subgroup_id = H5Gcreate(group_id, "CellData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);

  // Allocate memory and write scalar datasets
  double *scalar_dset = (double *)malloc(number_of_cells_loc * sizeof(double));
  for (scalar s in slist) {
    populate_scalar_dset(s, scalar_dset, number_of_cells_loc, offset_cells, count, offset, per_mask);
    create_chunked_dataset(subgroup_id, count, offset, s.name, number_of_cells, number_of_cells_loc, 1, scalar_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  }
  free(scalar_dset);

  // Allocate memory and write vector datasets
  double *vector_dset = (double *)malloc(number_of_cells_loc * 3 * sizeof(double));
  for (vector v in vlist) {    
    populate_vector_dset(v, vector_dset, number_of_cells_loc, offset_cells, count, offset, per_mask);
    create_chunked_dataset(subgroup_id, count, offset, v.x.name, number_of_cells, number_of_cells_loc, 3, vector_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  }
  free(vector_dset);
  H5Gclose(subgroup_id);

  // Create subgroup "FieldData"
  subgroup_id = H5Gcreate(group_id, "FieldData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
  H5Gclose(subgroup_id);

  // Create subgroup "PointData"
  subgroup_id = H5Gcreate(group_id, "PointData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
  H5Gclose(subgroup_id);
  H5Gclose(group_id);

  // Close HDF5 resources
  H5Fclose(file_id);
}

output_vtkhdf_slice(): Exports a 2D slice of a 3D field along x, y, or z.

This function takes a slice defined by n={n_x, n_y, n_z} and _alpha as as in view.h. Only works for x, y, and/or z planes. Here, _alpha is assumed to intersect a cell face and must be a multiple of L0/1<<MINLEVEL. This also means that scalar and vector fields are written at the corresponding face values.

Naturaly, this routine only works in 3D.

The arguments and their default values are:

list: pointer to the list of scalar fields to be exported.
vlist: pointer to the list of vector fields to be exported.
name: Output file name generally uses the .vtkhdf extension.
n: vector \vec{n} normal to the plane.
alpha: coordinate \alpha intersecting the plane.
compression_level: Level of compression to use when writing data to the HDF5 file (default=9).

Example Usage

scalar * slist = {a,b};
vector * vlist = {c,d};
output_vtkhdf_slice(slist, vlist, "slice_x.vtkhdf", (coord){1,0,0}, 0);
output_vtkhdf_slice(slist, vlist, "slice_y.vtkhdf", (coord){0,1,0}, 0);
output_vtkhdf_slice(slist, vlist, "slice_z.vtkhdf", (coord){0,0,1}, 0);

see, also example.

#if dimension == 3
trace void output_vtkhdf_slice(scalar *slist, vector *vlist, char *name, coord n = {0, 0, 1}, double _alpha = 0, int compression_level = 9){
  hid_t acc_tpl1;    // File access template
  hid_t file_id;     // HDF5 file ID
  hid_t group_id;    // HDF5 group ID
  hid_t subgroup_id; // HDF5 subgroup ID
  hsize_t count[2];  // Hyperslab selection parameters
  hsize_t offset[2]; // Offset for hyperslab
  hsize_t dims[1] = {2};

  // Define a scalar mask to deal with solids and periodic conditions
  scalar per_mask[];
  foreach (){
    per_mask[] = 0.;
    shortcut_slice(n, _alpha);
    if (alpha > 0.){
#if EMBED
      per_mask[] = cs[];
#else
      per_mask[] = 1.;
#endif
    }
  }

  // VTK cell types: VTK_QUAD
  int type = 9;
  int noffset = 4;

  // Obtain the number of points and cells and get a marker to reconstruct the topology
  int number_of_points = 0, number_of_points_loc = 0;
  int number_of_cells = 0,  number_of_cells_loc = 0;

  count_points_and_cells_slice(&number_of_points, &number_of_cells, &number_of_points_loc, &number_of_cells_loc, per_mask, n, _alpha);

  int number_of_ids = number_of_cells*noffset;
  int number_of_ids_loc = number_of_cells_loc*noffset;

  // Define chunk size for parallel I/O
  hsize_t chunk_size = number_of_cells / npe() / 2;

  // Calculate offsets for parallel I/O
  int offset_points[npe()], offset_cells[npe()], offset_ids[npe()], offset_offset[npe()];
  calculate_offsets2(offset_offset, number_of_cells_loc+1,  offset);
  calculate_offsets2(offset_ids,    number_of_ids_loc,      offset);
  calculate_offsets2(offset_cells,  number_of_cells_loc,    offset);
  calculate_offsets2(offset_points, number_of_points_loc,   offset);

  // Initialize marker for topology reconstruction
  vertex scalar marker[];
  initialize_marker_slice(marker, offset, n, _alpha, 0);

  // Setup file access template with parallel I/O access
  acc_tpl1 = H5Pcreate(H5P_FILE_ACCESS);
  H5Pset_fapl_mpio(acc_tpl1, MPI_COMM_WORLD, MPI_INFO_NULL);

  // Create a new HDF5 file collectively
  file_id = H5Fcreate(name, H5F_ACC_TRUNC, H5P_DEFAULT, acc_tpl1);

  // Release file-access template
  H5Pclose(acc_tpl1);

  // Create group 
  group_id = H5Gcreate(file_id, "VTKHDF", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);

  // Create "Version", "Type" (and other) attributes
  dims[0] = 2;
  int version_data[2] = {2, 1};
  create_attribute(group_id, "Version", version_data, dims);
  create_attribute_type(group_id, "Type", "UnstructuredGrid", 16);
  create_attribute_type(group_id, "Description", "Simulation perfomed using Basilisk (http://basilisk.fr/)", 57);
  
  // Write "NumberOfConnectivityIds", "NumberOfPoints", "NumberOfCells"
  dims[0] = npe();
  write_simple_dataset(group_id, "NumberOfConnectivityIds", offset_ids, dims);
  write_simple_dataset(group_id, "NumberOfPoints", offset_points, dims);
  write_simple_dataset(group_id, "NumberOfCells", offset_cells, dims);

  // Populate and write the points dataset
  double *points_dset;
  populate_points_dset_slice(&points_dset, number_of_points_loc, offset_points, count, offset, n, _alpha);
  create_chunked_dataset(group_id, count, offset, "Points", number_of_points, number_of_points_loc, 3, points_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  free(points_dset);

  // Populate and write the types dataset
  char * types_dset;
  populate_types_dset(&types_dset, type, number_of_cells_loc, offset_cells, count, offset);
  create_chunked_dataset(group_id, count, offset, "Types", number_of_cells, number_of_cells_loc, 1, types_dset, H5T_STD_U8LE, chunk_size, compression_level);
  free(types_dset);  

  // Populate and write the connectivity dataset
  long *topo_dset;
  populate_topo_dset_slice(&topo_dset, number_of_cells_loc, offset_ids, count, offset, per_mask, marker, n, _alpha);
  create_chunked_dataset(group_id, count, offset, "Connectivity", number_of_ids, number_of_ids_loc, 1, topo_dset, H5T_NATIVE_LONG, chunk_size, compression_level);
  free(topo_dset);  

  // Populate and write the offsets dataset
  long *offsets_dset;
  populate_offsets_dset(&offsets_dset, noffset, number_of_cells_loc+1, offset_offset, count, offset);
  create_chunked_dataset(group_id, count, offset, "Offsets", number_of_cells+npe(), number_of_cells_loc+1, 1, offsets_dset, H5T_NATIVE_LONG, chunk_size, compression_level);
  free(offsets_dset);  
  
  // Create subgroup "CellData"
  subgroup_id = H5Gcreate(group_id, "CellData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
  // Allocate memory and write scalar datasets
  double *scalar_dset = (double *)malloc(number_of_cells_loc * sizeof(double));
  for (scalar s in slist) {
    populate_scalar_dset_slice(s, scalar_dset, number_of_cells_loc, offset_cells, count, offset, per_mask, n, _alpha);
    create_chunked_dataset(subgroup_id, count, offset, s.name, number_of_cells, number_of_cells_loc, 1, scalar_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  }
  free(scalar_dset);

  // Allocate memory and write vector datasets
  double *vector_dset = (double *)malloc(number_of_cells_loc * 3 * sizeof(double));
  for (vector v in vlist) {    
    populate_vector_dset_slice(v, vector_dset, number_of_cells_loc, offset_cells, count, offset, per_mask, n, _alpha);
    create_chunked_dataset(subgroup_id, count, offset, v.x.name, number_of_cells, number_of_cells_loc, 3, vector_dset, H5T_NATIVE_DOUBLE, chunk_size, compression_level);
  }
  free(vector_dset);
  H5Gclose(subgroup_id);

  // Create subgroup "FieldData"
  subgroup_id = H5Gcreate(group_id, "FieldData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
  H5Gclose(subgroup_id);

  // Create subgroup "PointData"
  subgroup_id = H5Gcreate(group_id, "PointData", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
  H5Gclose(subgroup_id);
  H5Gclose(group_id);
  
  // Close HDF5 resources
  H5Fclose(file_id);
}
#endif

postamble: delete macros

#undef shortcut_slice