CellChemistry Class Reference

Work memory and kernels that advance the cells of a state as adiabatic constant-volume reactors at fixed (rho, e). More...

#include <cell_chemistry.h>

Classes
struct	Statistics

Public Member Functions
void	init (const Mixture &gas, const chemistry::Mechanism &mechanism, const chemistry::KineticsTable<> &kinetics, const CellChemistryOptions &options, uint32_t n_cells)
void	balance_across_ranks (double threshold)
	Balances the integration across ranks from now on (collective: every rank of the run, every call): see ChemistryBalance.
Statistics	advance (const StateView &U, const SpeciesView &rhoY, const Kokkos::View< rtype * > &T_seed, const Kokkos::View< rtype * > &chem_h, const Kokkos::View< rtype * > &chem_cost, uint32_t n, double dt, const Kokkos::View< rtype * > &time_scale={}, const ChemistryForcing *forcing=nullptr)
	Advance cells [0, n) over dt. Partial densities change; rho, rho u, rho E and the temperature seed do not.
void	heat_release (const StateView &U, const SpeciesView &rhoY, const Kokkos::View< rtype * > &T_seed, const Kokkos::View< rtype * > &hrr, const Kokkos::View< rtype **, Kokkos::LayoutRight > &production, uint32_t n)
	Heat release rate and mass production rates W_k omega_k of cells [0, n).
uint32_t	lanes () const
	Vector lanes per cell thread (1: one thread per cell).
uint32_t	threads () const
	Threads of a cell's team.
uint32_t	wide_threads () const
	Threads of the teams of the cells whose last call was expensive (GPUs), 0 if the same.
bool	binned () const
	Whether cells integrated one per thread are ordered by their last cost (GPUs).
size_t	shared_bytes () const
	Bytes of team scratch memory per cell holding its work memory, 0 if global memory.
uint32_t	sparse_entries () const
	Entries of L + U of the sparse LU, 0 for the dense one.

Detailed Description

Work memory and kernels that advance the cells of a state as adiabatic constant-volume reactors at fixed (rho, e).

A kernel flags the cells that need chemistry (T >= T_frozen and a mass fraction that would change by more than 1e-2 atol over dt at the current rates), a scan compacts them into a queue, and the queued cells are integrated in batches that bound the work memory, after the ranks have balanced the queues if asked to (ChemistryBalance). Each cell's result depends only on its own state, so neither the order of the queue, the batches nor the rank that integrates it changes it: per mechanism and build, every cell takes the same code path (one thread per cell, or a fixed number of lanes).

Member Function Documentation

advance()

CellChemistry::Statistics CellChemistry::advance	(	const StateView &	U,
		const SpeciesView &	rhoY,
		const Kokkos::View< rtype * > &	T_seed,
		const Kokkos::View< rtype * > &	chem_h,
		const Kokkos::View< rtype * > &	chem_cost,
		uint32_t	n,
		double	dt,
		const Kokkos::View< rtype * > &	time_scale = {},
		const ChemistryForcing *	forcing = nullptr
	)

Advance cells [0, n) over dt. Partial densities change; rho, rho u, rho E and the temperature seed do not.

Parameters

chem_h	Last sub-step of each cell, in and out.
chem_cost	Sub-steps of each cell, incremented.
time_scale	Optional per-cell multiplier of dt: the rates of an autonomous constant-volume reactor scaled by s give exactly its state after s dt (combustion models, docs/design/les.md 6.4).
forcing	Optional (options.forced): integrate dU/dt = s R(U) + forcing->rate instead, so that rho, rho u and rho E change by dt times their rates and every cell moves, reacting or not. A cell reacts if its chemistry is active at U or at U + dt rate; its internal energy per volume is taken linear in time between the two.

balance_across_ranks()

void CellChemistry::balance_across_ranks

(

double

threshold

)

Balances the integration across ranks from now on (collective: every rank of the run, every call): see ChemistryBalance.

binned()

bool CellChemistry::binned ( ) const

inline

Whether cells integrated one per thread are ordered by their last cost (GPUs).

heat_release()

void CellChemistry::heat_release	(	const StateView &	U,
		const SpeciesView &	rhoY,
		const Kokkos::View< rtype * > &	T_seed,
		const Kokkos::View< rtype * > &	hrr,
		const Kokkos::View< rtype **, Kokkos::LayoutRight > &	production,
		uint32_t	n
	)

Heat release rate and mass production rates W_k omega_k of cells [0, n).

init()

void CellChemistry::init	(	const Mixture &	gas,
		const chemistry::Mechanism &	mechanism,
		const chemistry::KineticsTable<> &	kinetics,
		const CellChemistryOptions &	options,
		uint32_t	n_cells
	)

Parameters

n_cells

Most cells advanced in one call (sizes the chunks).

lanes()

uint32_t CellChemistry::lanes ( ) const

inline

Vector lanes per cell thread (1: one thread per cell).

shared_bytes()

size_t CellChemistry::shared_bytes ( ) const

inline

Bytes of team scratch memory per cell holding its work memory, 0 if global memory.

sparse_entries()

uint32_t CellChemistry::sparse_entries ( ) const

inline

Entries of L + U of the sparse LU, 0 for the dense one.

threads()

uint32_t CellChemistry::threads ( ) const

inline

Threads of a cell's team.

wide_threads()

uint32_t CellChemistry::wide_threads ( ) const

inline

Threads of the teams of the cells whose last call was expensive (GPUs), 0 if the same.

---

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

• /home/runner/work/mallard/mallard/mallard/src/solver/cell_chemistry.h
• /home/runner/work/mallard/mallard/mallard/src/solver/cell_chemistry.cpp