DiFfRG::HDF5Output Class Reference

A class to output data to a CSV file. More...

#include <hdf5_output.hh>

Classes
struct	ScalarInfo

Public Member Functions
	HDF5Output (const std::string top_folder, const std::string output_name, const std::string &configuration_json="{}")
	Construct a new Csv Output object.
	HDF5Output (const std::string top_folder, const std::string output_name, const ConfigTree &config)
	~HDF5Output ()
template<typename T >
void	scalar (const std::string &name, const T value)
	Add a value to the output.
void	scalar (const std::string &name, const char *value)
template<typename COORD > requires is_coordinates<COORD>
void	map (const std::string &name, const COORD &coordinates)
template<typename COORD , typename T > requires is_coordinates<COORD>
void	map (const std::string &name, const COORD &coordinates, T *data)
template<typename INTERP > requires is_interpolator<INTERP>
void	map (const std::string &name, const INTERP &_interpolator)
template<typename INTERP > requires is_interpolator<INTERP>
void	map (const std::string &name, const std::string &coord_name, const INTERP &_interpolator)
void	stage_fe_frame (unsigned int series_number, double time, const std::string &group_name, const std::string &fe_output_name, unsigned int dim, std::size_t n_nodes, std::vector< double > node_data, std::vector< std::pair< std::string, std::vector< double > > > data_sets)
	Stage the node coordinates and nodal values of one finite-element frame.
DiFfRG::hdf5::File &	get_file ()
	Open the file for setup work performed before any frame is written.
void	flush (const double time)
void	run_staged_frame (double time)
	Execute and clear whatever is staged, without the per-frame scalar bookkeeping.
void	set_writer (HDF5FrameWriter *writer)
	Route staged frames through a writer thread instead of writing them inline.
void	drain ()
void	mark_finished (bool crashed)
	Record how the run ended: sets the root finished attribute to 1, and crashed to whether it ended by throwing.
void	open_file ()
void	close_file ()
FrameTimings	take_frame_timings ()

Static Public Member Functions
static void	write_series_record (DiFfRG::hdf5::Group &group, const int series_number)

Private Member Functions
template<typename T >
void	stage_map (const std::string &name, const std::string &coord_name, const DiFfRG::hdf5::Dims &dims, std::vector< T > payload)
	Stage one map series: the group, its series record, its data and its coordinate link.
template<typename COORD >
void	check_coordinates (const std::string &coord_name, const COORD &coordinates)

Private Attributes
const std::string	top_folder
const std::string	output_name
bool	opened
std::map< std::string, ScalarInfo >	scalar_infos
std::set< std::string >	created_map_groups
HDF5Frame	staged
HDF5FrameWriter *	writer = nullptr
std::list< std::string >	written_scalars
std::vector< std::pair< std::string, size_t > >	written_map_series
std::map< std::string, size_t >	map_series_numbers
std::map< std::string, std::string >	coord_identifiers
std::list< std::string >	initial_scalars
DiFfRG::hdf5::File	h5_file
std::filesystem::path	path
FrameTimings	frame_timings

Detailed Description

A class to output data to a CSV file.

In every time step, the user can add values to the output, which will be written to the file when flush is called. Every timestep has to contain the same values, but the order in which they are added can change.

Constructor & Destructor Documentation

HDF5Output() [1/2]

DiFfRG::HDF5Output::HDF5Output	(	const std::string	top_folder,
		const std::string	output_name,
		const std::string &	configuration_json = "{}" )

Construct a new Csv Output object.

Parameters

top_folder	The top folder to store the output in.
output_name	The name of the output file.
config	The ConfigTree object containing the parameters.

HDF5Output() [2/2]

DiFfRG::HDF5Output::HDF5Output ( const std::string top_folder, const std::string output_name, const ConfigTree & config )

inline

~HDF5Output()

DiFfRG::HDF5Output::~HDF5Output

(

)

Member Function Documentation

check_coordinates()

template<typename COORD >

void DiFfRG::HDF5Output::check_coordinates ( const std::string & coord_name, const COORD & coordinates )

inlineprivate

close_file()

void DiFfRG::HDF5Output::close_file

(

)

drain()

void DiFfRG::HDF5Output::drain

(

)

Wait for this sink's queued frames to reach disk and surface any writer error.

flush()

void DiFfRG::HDF5Output::flush

(

const double

time

)

get_file()

DiFfRG::hdf5::File & DiFfRG::HDF5Output::get_file

(

)

Open the file for setup work performed before any frame is written.

Not for per-frame writes – those go through scalar()/map()/stage_fe_frame() and are executed together when the frame is flushed. Callers must close_file() when done, so the lock is not held into the run.

map() [1/4]

template<typename COORD >
requires is_coordinates<COORD>

void DiFfRG::HDF5Output::map ( const std::string & name, const COORD & coordinates )

inline

map() [2/4]

template<typename COORD , typename T >
requires is_coordinates<COORD>

void DiFfRG::HDF5Output::map ( const std::string & name, const COORD & coordinates, T * data )

inline

map() [3/4]

template<typename INTERP >
requires is_interpolator<INTERP>

void DiFfRG::HDF5Output::map ( const std::string & name, const INTERP & _interpolator )

inline

map() [4/4]

template<typename INTERP >
requires is_interpolator<INTERP>

void DiFfRG::HDF5Output::map ( const std::string & name, const std::string & coord_name, const INTERP & _interpolator )

inline

mark_finished()

void DiFfRG::HDF5Output::mark_finished

(

bool

crashed

)

Record how the run ended: sets the root finished attribute to 1, and crashed to whether it ended by throwing.

Call once, after the last frame. Files start at finished = 0, so a run that never gets here – SIGKILL, OOM, a node eviction – stays marked unfinished. A run that threw still reaches this and is marked finished, with crashed = 1: its output was flushed and closed properly, it just stopped early.

open_file()

void DiFfRG::HDF5Output::open_file

(

)

run_staged_frame()

void DiFfRG::HDF5Output::run_staged_frame

(

double

time

)

Execute and clear whatever is staged, without the per-frame scalar bookkeeping.

flush() is the normal entry point. This exists for the standalone FEOutput case, where nobody flushes the HDF5Output at all and the staged field data would otherwise be discarded at destruction.

scalar() [1/2]

void DiFfRG::HDF5Output::scalar ( const std::string & name, const char * value )

inline

scalar() [2/2]

template<typename T >

void DiFfRG::HDF5Output::scalar ( const std::string & name, const T value )

inline

Add a value to the output.

Validates and copies here, on the calling thread, so that a rejected value still fails at the call site with the context that produced it – but performs no HDF5 call. The write itself is staged and happens when the frame is flushed.

Parameters

name	The name of the value.
value	The value to add.

set_writer()

void DiFfRG::HDF5Output::set_writer ( HDF5FrameWriter * writer )

inline

Route staged frames through a writer thread instead of writing them inline.

Not owned. Every HDF5Output of one run must share the same writer, because HDF5 is not thread safe here and two writers would call into it concurrently. Passing nullptr (the default) keeps everything on the calling thread.

stage_fe_frame()

void DiFfRG::HDF5Output::stage_fe_frame	(	unsigned int	series_number,
		double	time,
		const std::string &	group_name,
		const std::string &	fe_output_name,
		unsigned int	dim,
		std::size_t	n_nodes,
		std::vector< double >	node_data,
		std::vector< std::pair< std::string, std::vector< double > > >	data_sets )

Stage the node coordinates and nodal values of one finite-element frame.

Called by FEOutput, which owns the DataOut and the filtering, but not the file. The arrays are moved in; nothing here touches HDF5.

stage_map()

template<typename T >

void DiFfRG::HDF5Output::stage_map ( const std::string & name, const std::string & coord_name, const DiFfRG::hdf5::Dims & dims, std::vector< T > payload )

inlineprivate

Stage one map series: the group, its series record, its data and its coordinate link.

Shared by the raw-pointer and interpolator overloads, which differ only in how they produce the dense payload and how the coordinate dataset gets its name.

take_frame_timings()

FrameTimings DiFfRG::HDF5Output::take_frame_timings ( )

inline

Timing buckets accumulated since the last call, and reset by it.

write_series_record()

static void DiFfRG::HDF5Output::write_series_record ( DiFfRG::hdf5::Group & group, const int series_number )

inlinestatic

Append the series number, as text, to a per-series-group dataset. Runs on the writer.

Member Data Documentation

coord_identifiers

std::map<std::string, std::string> DiFfRG::HDF5Output::coord_identifiers

private

created_map_groups

std::set<std::string> DiFfRG::HDF5Output::created_map_groups

private

Map names whose parent group has been created (or staged for creation).

frame_timings

FrameTimings DiFfRG::HDF5Output::frame_timings

private

h5_file

DiFfRG::hdf5::File DiFfRG::HDF5Output::h5_file

private

Only ever open for setup work (see get_file()); frames use their own context.

initial_scalars

std::list<std::string> DiFfRG::HDF5Output::initial_scalars

private

map_series_numbers

std::map<std::string, size_t> DiFfRG::HDF5Output::map_series_numbers

private

opened

bool DiFfRG::HDF5Output::opened

private

output_name

const std::string DiFfRG::HDF5Output::output_name

private

path

std::filesystem::path DiFfRG::HDF5Output::path

private

scalar_infos

std::map<std::string, ScalarInfo> DiFfRG::HDF5Output::scalar_infos

private

staged

HDF5Frame DiFfRG::HDF5Output::staged

private

Writes accumulated for the frame in progress; executed by flush().

top_folder

const std::string DiFfRG::HDF5Output::top_folder

private

writer

HDF5FrameWriter* DiFfRG::HDF5Output::writer = nullptr

private

Not owned; shared with every other HDF5Output of the same run.

written_map_series

std::vector<std::pair<std::string, size_t> > DiFfRG::HDF5Output::written_map_series

private

(name, series) of every map written this frame. Carries the series number rather than re-deriving it as map_series_numbers[name] - 1 in flush(): a name written twice in one frame would otherwise stamp attributes onto the same group twice and fail in H5Acreate2, pointing at the flush instead of at the duplicate write.

written_scalars

std::list<std::string> DiFfRG::HDF5Output::written_scalars

private

---

The documentation for this class was generated from the following file:

• /home/runner/work/DiFfRG_current/DiFfRG_current/DiFfRG/include/DiFfRG/discretization/data/hdf5_output.hh