Specifies some necessaries for AMP style endgame implementations, which differ from the fixed precision ones. More...

#include <amp_endgame.hpp>

Inheritance diagram for bertini::endgame::AMPEndgame:

Collaboration diagram for bertini::endgame::AMPEndgame:

Public Types
using	TrackerT = tracking::AMPTracker
	The path-tracker type.
using	EmitterType = AMPEndgame
	The event-emitter type.
using	TrackerType
	The path-tracker type.

Public Member Functions
SuccessCode	RefineSampleImpl (Vec< complex_mp > &result, Vec< complex_mp > const &current_sample, complex_mp const &current_time, NumErrorT tol, unsigned max_iterations) const
	Refine a single sample point, escalating precision if Newton refinement needs it.
SuccessCode	RefineSampleImpl (Vec< complex_dbl > &result, Vec< complex_dbl > const &current_sample, complex_dbl const &current_time, NumErrorT tol, unsigned max_iterations) const
	Hardware-double fast-lane refine for the adaptive-numeric-type (double-first) endgame.
unsigned	NextEscalatedPrecision () const
	Return the next (higher) precision to escalate to, following the tracker's authority and strictly above the current precision.
template<typename TimesTuple>
void	CrossTimesUp (TimesTuple &times, unsigned newprec) const
	Widen a times container's complex_dbl slot into the complex_mp slot at the new precision.
template<typename SampsTuple>
void	CrossSampsUp (SampsTuple &samps, unsigned newprec) const
	Widen a samples container's complex_dbl slot into the complex_mp slot at the new precision.
template<typename VecTuple>
void	CrossVecUp (VecTuple &vec, unsigned newprec) const
	Widen a vector container's complex_dbl slot into the complex_mp slot at the new precision.
Vec< complex_dbl >	DowncastToDouble (Vec< complex_mp > const &v) const
	Convert an mp vector down to the hardware complex_dbl fast lane.
complex_mp	AtActivePrecisionScalar (complex_mp const &x) const
	Copy a complex_mp scalar set to the endgame's active working precision.
Vec< complex_mp >	AtActivePrecisionVec (Vec< complex_mp > const &v) const
	Copy a complex_mp vector set to the endgame's active working precision.
template<typename ComplexT>
Vec< complex_mp >	ToBCT (Vec< ComplexT > const &v, unsigned prec) const
	Widen a vector to the boundary complex type (complex_mp) at the given precision.
	AMPEndgame (TrackerT const &new_tracker)
	Construct the AMP endgame precision policy for a tracker.
virtual	~AMPEndgame ()=default
void	SetTracker (tracking::AMPTracker const &new_tracker)
	Tell the endgame to use the given tracker.
const tracking::AMPTracker &	GetTracker () const
	Getter for the tracker used inside an instance of the endgame.
const System &	GetSystem () const
	Get the system being tracked on, which is referred to by the tracker.
void	ChangePrecision (unsigned p)
	Change the working precision of the underlying tracker.
void	AddObserver (AnyObserver &new_observer) const
	Add an observer (non-owning) to observe this observable.
void	AddObserver (std::shared_ptr< AnyObserver > const &new_observer) const
	Add an observer whose lifetime is co-owned by this observable.
void	RemoveObserver (AnyObserver &observer) const
	Remove an observer from this observable.
virtual bool	ObservableIsA (std::type_index t) const
	Whether this observable may be observed as the given type.

Static Public Member Functions
template<typename... T>
static unsigned	EnsureAtUniformPrecision (T &...args)
	Bring a set of values to a common (highest) precision.
static void	EnsureAtPrecision (double &, unsigned prec)
	No-op precision check for a double (throws if asked for non-double precision).
static void	EnsureAtPrecision (std::complex< double > &, unsigned prec)
	No-op precision check for a std::complex<double> (throws if asked for non-double precision).
static void	EnsureAtPrecision (real_mp &obj, unsigned prec)
	Set a multiprecision real to the given precision.
static void	EnsureAtPrecision (complex_mp &obj, unsigned prec)
	Set a multiprecision complex to the given precision.

Public Attributes
unsigned	current_endgame_precision_ = DoublePrecision()
	The endgame's current working precision (double = fast lane, higher = mpfr slot).
bool	adaptive_numeric_type_active_ = false
	When true, the double-first adaptive-numeric-type lane is armed.

Protected Member Functions
void	NotifyObservers (AnyEvent const &e) const
	Sends an event to observers that subscribed to its exact dynamic type, then to all catch-all (untyped) observers.
void	NotifyObservers (AnyEvent &e) const
	Notify all observers of a (mutable) event.
bool	HasObservers () const
	Whether anyone is currently observing (conservative).

Detailed Description

Specifies some necessaries for AMP style endgame implementations, which differ from the fixed precision ones.

Member Typedef Documentation

◆ EmitterType

using bertini::endgame::AMPEndgame::EmitterType = AMPEndgame

The event-emitter type.

◆ TrackerT

using bertini::endgame::AMPEndgame::TrackerT = tracking::AMPTracker

The path-tracker type.

◆ TrackerType

using bertini::endgame::EndgamePrecPolicyBase< tracking::AMPTracker >::TrackerType

inherited

The path-tracker type.

Constructor & Destructor Documentation

◆ AMPEndgame()

bertini::endgame::AMPEndgame::AMPEndgame ( TrackerT const & new_tracker )

inlineexplicit

Construct the AMP endgame precision policy for a tracker.

◆ ~AMPEndgame()

virtual bertini::endgame::AMPEndgame::~AMPEndgame ( )

virtualdefault

Member Function Documentation

◆ AddObserver() [1/2]

void bertini::Observable::AddObserver ( AnyObserver & new_observer ) const

inlineinherited

Add an observer (non-owning) to observe this observable.

Stores only a raw reference: the caller is responsible for keeping the observer alive while it is attached. This is the right overload for stack-allocated C++ observers added and removed within a scope. For an observer whose lifetime you would rather have tracked, see the shared_ptr overload.

Observers that override SubscribedEventTypes() with a non-empty list are registered in a type-indexed map so NotifyObservers only calls them for events they declared interest in. Observers returning an empty list (the default) are placed in a catch-all list and receive every event.

Safe to call from inside an observer's Observe() (i.e. during a notification): the add is deferred and applied once the current notification loop finishes. This is what lets a "meta-observer" attach new observers in response to events.

Exceptions

IncompatibleObserver if the observer's ObservedKind() is not one this observable can be observed as.

◆ AddObserver() [2/2]

void bertini::Observable::AddObserver ( std::shared_ptr< AnyObserver > const & new_observer ) const

inlineinherited

Add an observer whose lifetime is co-owned by this observable.

The observable keeps a shared_ptr, so the observer stays alive (and keeps receiving events) for as long as it is attached, even after the caller drops their own reference — "attach it and forget it", with no dangling. It is released when RemoveObserver is called or this observable is destroyed. (Stack-allocated C++ observers can't be shared_ptr-owned; use the reference overload for those.)

Exceptions

IncompatibleObserver if incompatible (see the reference overload).

◆ AtActivePrecisionScalar()

complex_mp bertini::endgame::AMPEndgame::AtActivePrecisionScalar ( complex_mp const & x ) const

inline

Copy a complex_mp scalar set to the endgame's active working precision.

◆ AtActivePrecisionVec()

Vec< complex_mp > bertini::endgame::AMPEndgame::AtActivePrecisionVec ( Vec< complex_mp > const & v ) const

inline

Copy a complex_mp vector set to the endgame's active working precision.

◆ ChangePrecision()

void bertini::endgame::EndgamePrecPolicyBase< tracking::AMPTracker >::ChangePrecision ( unsigned p )

inlineinherited

Change the working precision of the underlying tracker.

◆ CrossSampsUp()

template<typename SampsTuple>

void bertini::endgame::AMPEndgame::CrossSampsUp	(	SampsTuple &	samps,
		unsigned	newprec ) const

inline

Widen a samples container's complex_dbl slot into the complex_mp slot at the new precision.

◆ CrossTimesUp()

template<typename TimesTuple>

void bertini::endgame::AMPEndgame::CrossTimesUp	(	TimesTuple &	times,
		unsigned	newprec ) const

inline

Widen a times container's complex_dbl slot into the complex_mp slot at the new precision.

◆ CrossVecUp()

template<typename VecTuple>

void bertini::endgame::AMPEndgame::CrossVecUp	(	VecTuple &	vec,
		unsigned	newprec ) const

inline

Widen a vector container's complex_dbl slot into the complex_mp slot at the new precision.

◆ DowncastToDouble()

Vec< complex_dbl > bertini::endgame::AMPEndgame::DowncastToDouble ( Vec< complex_mp > const & v ) const

inline

Convert an mp vector down to the hardware complex_dbl fast lane.

◆ EnsureAtPrecision() [1/4]

void bertini::endgame::AMPEndgame::EnsureAtPrecision	(	complex_mp &	obj,
		unsigned	prec )

inlinestatic

Set a multiprecision complex to the given precision.

◆ EnsureAtPrecision() [2/4]

void bertini::endgame::AMPEndgame::EnsureAtPrecision	(	double &	,
		unsigned	prec )

inlinestatic

No-op precision check for a double (throws if asked for non-double precision).

◆ EnsureAtPrecision() [3/4]

void bertini::endgame::AMPEndgame::EnsureAtPrecision	(	real_mp &	obj,
		unsigned	prec )

inlinestatic

Set a multiprecision real to the given precision.

◆ EnsureAtPrecision() [4/4]

void bertini::endgame::AMPEndgame::EnsureAtPrecision	(	std::complex< double > &	,
		unsigned	prec )

inlinestatic

No-op precision check for a std::complex<double> (throws if asked for non-double precision).

◆ EnsureAtUniformPrecision()

template<typename... T>

unsigned bertini::endgame::AMPEndgame::EnsureAtUniformPrecision ( T &... args )

inlinestatic

Bring a set of values to a common (highest) precision.

◆ GetSystem()

const System & bertini::endgame::EndgamePrecPolicyBase< tracking::AMPTracker >::GetSystem ( ) const

inlineinherited

Get the system being tracked on, which is referred to by the tracker.

◆ GetTracker()

const tracking::AMPTracker & bertini::endgame::EndgamePrecPolicyBase< tracking::AMPTracker >::GetTracker ( ) const

inlineinherited

Getter for the tracker used inside an instance of the endgame.

◆ HasObservers()

bool bertini::Observable::HasObservers ( ) const

inlineprotectedinherited

Whether anyone is currently observing (conservative).

Lets an emitter skip constructing an event – and any work needed to build its payload, such as a numeric-type conversion – when there is no observer to receive it. Conservative: returns true if any watcher list is non-empty (even if a stale, emptied typed bucket lingers), so it never wrongly suppresses a real observer.

◆ NextEscalatedPrecision()

unsigned bertini::endgame::AMPEndgame::NextEscalatedPrecision ( ) const

inline

Return the next (higher) precision to escalate to, following the tracker's authority and strictly above the current precision.

◆ NotifyObservers() [1/2]

void bertini::Observable::NotifyObservers ( AnyEvent & e ) const

inlineprotectedinherited

Notify all observers of a (mutable) event.

◆ NotifyObservers() [2/2]

void bertini::Observable::NotifyObservers ( AnyEvent const & e ) const

inlineprotectedinherited

Sends an event to observers that subscribed to its exact dynamic type, then to all catch-all (untyped) observers.

Mutations of the observer lists are deferred for the duration of a notification (see AddObserver/RemoveObserver), so iterating the live lists here is safe even when an observer subscribes/unsubscribes mid-dispatch. An observer that returns ObserveResult::Unsubscribe is dropped once the (possibly nested) notification completes. The depth counter keeps this correct under re-entrant emission.

◆ ObservableIsA()

virtual bool bertini::Observable::ObservableIsA ( std::type_index t ) const

inlinevirtualinherited

Whether this observable may be observed as the given type.

Drives the compatibility check in AddObserver. The default accepts the wildcard typeid(void) and this observable's own dynamic type. An observable that emits its events templated on a base type (so a single observer type serves many concrete observables) should also accept that base type — e.g. ZeroDim accepts AnyZeroDim.

Reimplemented in bertini::algorithm::HomotopySolver< TrackerType, EndgameType, SystemType >, and bertini::algorithm::HomotopySolver< A, B, C >.

◆ RefineSampleImpl() [1/2]

SuccessCode bertini::endgame::AMPEndgame::RefineSampleImpl	(	Vec< complex_dbl > &	result,
		Vec< complex_dbl > const &	current_sample,
		complex_dbl const &	current_time,
		NumErrorT	tol,
		unsigned	max_iterations ) const

inline

Hardware-double fast-lane refine for the adaptive-numeric-type (double-first) endgame.

Pure-(i) escalation: this does NOT raise precision in place. The endgame is computing in the complex_dbl slot, with the tracker and system already at double precision; we refine there. If double cannot reach the tolerance the tracker returns HigherPrecisionNecessary / FailedToConverge, and we hand that code straight back to RunImplAMP, which migrates every container to mpfr and retries in the complex_mp slot (where the in-place mp->higher-mp escalation above takes over).

◆ RefineSampleImpl() [2/2]

SuccessCode bertini::endgame::AMPEndgame::RefineSampleImpl	(	Vec< complex_mp > &	result,
		Vec< complex_mp > const &	current_sample,
		complex_mp const &	current_time,
		NumErrorT	tol,
		unsigned	max_iterations ) const

inline

Refine a single sample point, escalating precision if Newton refinement needs it.

◆ RemoveObserver()

void bertini::Observable::RemoveObserver ( AnyObserver & observer ) const

inlineinherited

Remove an observer from this observable.

Like AddObserver, this is safe to call during a notification: the removal is deferred until the current notification loop finishes.

◆ SetTracker()

void bertini::endgame::EndgamePrecPolicyBase< tracking::AMPTracker >::SetTracker ( tracking::AMPTracker const & new_tracker )

inlineinherited

Tell the endgame to use the given tracker.

Takes a reference.

Note: Ensure the tracker you are using doesn not go out of scope!

◆ ToBCT()

template<typename ComplexT>

Vec< complex_mp > bertini::endgame::AMPEndgame::ToBCT	(	Vec< ComplexT > const &	v,
		unsigned	prec ) const

inline

Widen a vector to the boundary complex type (complex_mp) at the given precision.

Member Data Documentation

◆ adaptive_numeric_type_active_

bool bertini::endgame::AMPEndgame::adaptive_numeric_type_active_ = false

mutable

When true, the double-first adaptive-numeric-type lane is armed.

◆ current_endgame_precision_

unsigned bertini::endgame::AMPEndgame::current_endgame_precision_ = DoublePrecision()

mutable

The endgame's current working precision (double = fast lane, higher = mpfr slot).

Public Types

Public Member Functions

Static Public Member Functions

Public Attributes

Protected Member Functions

Detailed Description

Member Typedef Documentation

◆ EmitterType

◆ TrackerT

◆ TrackerType

Constructor & Destructor Documentation

◆ AMPEndgame()

◆ ~AMPEndgame()

Member Function Documentation

◆ AddObserver() [1/2]

◆ AddObserver() [2/2]

◆ AtActivePrecisionScalar()

◆ AtActivePrecisionVec()

◆ ChangePrecision()

◆ CrossSampsUp()

◆ CrossTimesUp()

◆ CrossVecUp()

◆ DowncastToDouble()

◆ EnsureAtPrecision() [1/4]

◆ EnsureAtPrecision() [2/4]

◆ EnsureAtPrecision() [3/4]

◆ EnsureAtPrecision() [4/4]

◆ EnsureAtUniformPrecision()

◆ GetSystem()

◆ GetTracker()

◆ HasObservers()

◆ NextEscalatedPrecision()

◆ NotifyObservers() [1/2]

◆ NotifyObservers() [2/2]

◆ ObservableIsA()

◆ RefineSampleImpl() [1/2]

◆ RefineSampleImpl() [2/2]

◆ RemoveObserver()

◆ SetTracker()

◆ ToBCT()

Member Data Documentation

◆ adaptive_numeric_type_active_

◆ current_endgame_precision_