gridfire::engine::scratch::GraphEngineScratchPad Struct Reference

Scratchpad for storing CppAD automatic differentiation state for GraphEngine. More...

#include <engine_graph_scratchpad.h>

Inheritance diagram for gridfire::engine::scratch::GraphEngineScratchPad:

Collaboration diagram for gridfire::engine::scratch::GraphEngineScratchPad:

Public Types
enum class	ADFunRegistrationResult : uint8_t { SUCCESS , ALREADY_REGISTERED }
	Result codes for ADFun registration operations. More...

Public Member Functions
bool	is_initialized () const override
	Check whether the scratchpad has been initialized.
void	initialize (const GraphEngine &engine)
	Initialize the scratchpad from a GraphEngine.
std::unique_ptr< AbstractScratchPad >	clone () const override
	Create a deep copy of this scratchpad.
Public Member Functions inherited from gridfire::engine::scratch::AbstractScratchPad
virtual	~AbstractScratchPad ()=default
	Virtual destructor for proper cleanup of derived classes.

Public Attributes
std::optional< CppAD::ADFun< double > >	rhsADFun
	CppAD function object for evaluating the ODE right-hand side.
CppAD::sparse_jac_work	jac_work
	Work structure for sparse Jacobian calculations.
std::vector< double >	local_abundance_cache
	Local cache of abundance values for efficient RHS evaluation.
std::unordered_map< size_t, StepDerivatives< double > >	stepDerivativesCache
	Cache of step derivatives indexed by timestep identifier.
std::unordered_map< size_t, CppAD::sparse_rcv< std::vector< size_t >, std::vector< double > > >	jacobianSubsetCache
	Cache of sparse Jacobian subsets indexed by timestep identifier.
std::unordered_map< size_t, CppAD::sparse_jac_work >	jacWorkCache
	Cache of Jacobian work structures indexed by timestep identifier.
std::optional< StepDerivatives< double > >	most_recent_rhs_calculation
	The most recent RHS calculation result for warm-starting.
bool	has_initialized = false
	Flag indicating whether the scratchpad has been initialized.

Static Public Attributes
static constexpr auto	ID = ScratchPadType::GRAPH_ENGINE_SCRATCHPAD
	Unique identifier for this scratchpad type.

Detailed Description

Scratchpad for storing CppAD automatic differentiation state for GraphEngine.

GraphEngineScratchPad provides thread-local storage for all CppAD-related objects needed during ODE integration with the GraphEngine. This includes the ADFun object for evaluating the right-hand side of the ODE and computing Jacobians, as well as various caches to improve performance.

Thread Safety

This class is not thread-safe. CppAD ADFun objects maintain internal state that is modified during Forward and Reverse mode operations. Each thread must have its own scratchpad instance. Use clone() to create independent copies for parallel execution.

Note

When cloning, if the rhsADFun has not been initialized, the clone will also be uninitialized and has_initialized will be false.

Member Enumeration Documentation

ADFunRegistrationResult

enum class gridfire::engine::scratch::GraphEngineScratchPad::ADFunRegistrationResult : uint8_t

strong

Result codes for ADFun registration operations.

Enumerator
SUCCESS	Registration completed successfully.
ALREADY_REGISTERED	ADFun was already registered; no action taken.

Member Function Documentation

clone()

std::unique_ptr< AbstractScratchPad > gridfire::engine::scratch::GraphEngineScratchPad::clone ( ) const

inlinenodiscardoverridevirtual

Create a deep copy of this scratchpad.

Creates an independent copy of all internal state, including the CppAD ADFun object and all caches. The clone can be safely used in a separate thread without affecting the original.

Returns

A unique pointer to the cloned scratchpad.

Note

If rhsADFun is not initialized, the clone will also be uninitialized (has_initialized = false).

Examples

GraphEngineScratchPad scratch;
scratch.initialize(engine);
 
// Create independent copies for parallel workers
std::vector<std::unique_ptr<AbstractScratchPad>> worker_pads;
for (int i = 0; i < num_threads; ++i) {
    worker_pads.push_back(scratch.clone());
}

Implements gridfire::engine::scratch::AbstractScratchPad.

initialize()

void gridfire::engine::scratch::GraphEngineScratchPad::initialize ( const GraphEngine & engine )

inline

Initialize the scratchpad from a GraphEngine.

Copies the authoritative ADFun from the engine and clears all caches. This method is idempotent; calling it multiple times has no effect after the first successful initialization.

Parameters

engine

The GraphEngine to initialize from.

Examples

GraphEngineScratchPad scratch;
GraphEngine engine = create_engine();
scratch.initialize(engine);
 
// Now safe to use for thread-local computations
auto& adfun = scratch.rhsADFun.value();

is_initialized()

bool gridfire::engine::scratch::GraphEngineScratchPad::is_initialized ( ) const

inlinenodiscardoverridevirtual

Check whether the scratchpad has been initialized.

Returns

true if initialized with a valid ADFun, false otherwise.

Implements gridfire::engine::scratch::AbstractScratchPad.

Member Data Documentation

has_initialized

bool gridfire::engine::scratch::GraphEngineScratchPad::has_initialized = false

Flag indicating whether the scratchpad has been initialized.

ID

auto gridfire::engine::scratch::GraphEngineScratchPad::ID = ScratchPadType::GRAPH_ENGINE_SCRATCHPAD

staticconstexpr

Unique identifier for this scratchpad type.

jac_work

CppAD::sparse_jac_work gridfire::engine::scratch::GraphEngineScratchPad::jac_work

Work structure for sparse Jacobian calculations.

Reused across Jacobian evaluations to avoid reallocation.

jacobianSubsetCache

std::unordered_map<size_t, CppAD::sparse_rcv<std::vector<size_t>, std::vector<double> > > gridfire::engine::scratch::GraphEngineScratchPad::jacobianSubsetCache

Cache of sparse Jacobian subsets indexed by timestep identifier.

jacWorkCache

std::unordered_map<size_t, CppAD::sparse_jac_work> gridfire::engine::scratch::GraphEngineScratchPad::jacWorkCache

Cache of Jacobian work structures indexed by timestep identifier.

local_abundance_cache

std::vector<double> gridfire::engine::scratch::GraphEngineScratchPad::local_abundance_cache

Local cache of abundance values for efficient RHS evaluation.

most_recent_rhs_calculation

std::optional<StepDerivatives<double> > gridfire::engine::scratch::GraphEngineScratchPad::most_recent_rhs_calculation

The most recent RHS calculation result for warm-starting.

rhsADFun

std::optional<CppAD::ADFun<double> > gridfire::engine::scratch::GraphEngineScratchPad::rhsADFun

CppAD function object for evaluating the ODE right-hand side.

Contains the computational graph for automatic differentiation.

stepDerivativesCache

std::unordered_map<size_t, StepDerivatives<double> > gridfire::engine::scratch::GraphEngineScratchPad::stepDerivativesCache

Cache of step derivatives indexed by timestep identifier.

---

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

• src/include/gridfire/engine/scratchpads/engine_graph_scratchpad.h