WirelengthOptimizer builds numerical models based on the element locations and calls solvers to find an optimal solution of the placement. More...

#include <WirelengthOptimizer.h>

Collaboration diagram for WirelengthOptimizer:

[legend]

Classes
struct	_CellWithScore

struct	_PUWithScore

struct	_slackEnhanceTuple

Public Member Functions
	WirelengthOptimizer (PlacementInfo *placementInfo, std::map< std::string, std::string > &JSONCfg, bool verbose=true)
	Construct a new Wirelength Optimizer object. More...

	~WirelengthOptimizer ()

void	GlobalPlacementQPSolve (float pesudoNetWeight, bool firstIteration=true, bool forwardSolutionToNextIteration=false, bool enableMacroPseudoNet2Site=false, bool considerNetNum=true, bool enableUserDefinedClusterOpt=false, float displacementLimit=-10, PlacementTimingOptimizer *timingOptimizer=nullptr)
	use quadratic model to estimate the HPWL and some timing/user-defined pseudo nets are involved for specific optimization More...

void	updateB2BNetWeight (float pesudoNetWeight, bool enableMacroPseudoNet2Site=false, bool considerNetNum=true, bool enableUserDefinedClusterOpt=false, PlacementTimingOptimizer *timingOptimizer=nullptr)
	update the net weights in the quadratic model according to B2B net HPWL model. More...

void	reloadPlacementInfo ()
	re-initialize some parameters and optimizer configuration according to the PlacementInfo More...

void	setMacroLegalizationParameters (int cnt, float macroLegalizationWeight)
	Set the macro legalization parameters. More...

int	getMacroPseudoNetEnhanceCnt ()
	Get the number of the conducted macro iterations. More...

float	getMacroLegalizationWeight ()
	Get the net weight for macro legalization. More...

void	clearNetPinEnhanceRate ()

void	setGeneralTimingNetWeight (float _generalTimingNetWeight)

void	setSlackPowerFactor (double _slackPowerFactor)

Static Public Member Functions
static void	updateB2BNetWeightWorker (PlacementInfo *placementInfo, std::vector< Eigen::Triplet< float >> &objectiveMatrixTripletList, std::vector< float > &objectiveMatrixDiag, Eigen::VectorXd &objectiveVector, float generalNetWeight, float y2xRatio, bool updateX, bool updateY)
	a worker funtion for multi-threading net weight updating More...

Private Types
typedef struct WirelengthOptimizer::_slackEnhanceTuple	slackEnhanceTuple

typedef struct WirelengthOptimizer::_PUWithScore	PUWithScore

typedef struct WirelengthOptimizer::_CellWithScore	CellWithScore

Private Member Functions
void	solverLoadData ()
	load to placement location from PlacementInfo to the solver More...

void	solverLoadFixedData ()
	load to placement location of fixed PlacementUnits from PlacementInfo to the solver More...

void	solverWriteBackData (float displacementLimit)
	write placement location from the solver to the PlacementInfo More...

void	addPseudoNetForMacros (float pesudoNetWeight, bool considerNetNum)
	add the legalization pseudo nets to force macros move to the legal sites More...

void	addPseudoNet2LoctionForAllPUs (float pesudoNetWeight, bool considerNetNum)
	as convential QP placers do, we add pseudo nets between anchors and PlacementUnit to constrain the movement of PlacementUnit for WL optimization. More...

void	updatePseudoNetForUserDefinedClusters (float pesudoNetWeight)
	add pseudo net for user-defined clusters More...

void	addPseudoNet_SlackBased (float timingWeight, double slackPowFactor, PlacementTimingOptimizer *timingOptimizer, bool calculate=false)
	add pseudo net for timing optimization based on the timing slack of each elements in the design netlist More...

void	LUTLUTPairing_TimingDriven (float timingWeight, float disThreshold, PlacementTimingOptimizer *timingOptimizer)

void	updatePseudoNetForClockRegion (float pesudoNetWeight)
	add pseudo nets for clock region More...

double	manhattanDis (double x0, double y0, double x1, double y1)
	evaluate the Mahattan distance between two locations More...

float	getCellDistance (PlacementInfo::Location &A, PlacementInfo::Location &B)

Static Private Member Functions
static void	QPSolve (QPSolverWrapper *&curSolver)
	call the cooresponding solver to solve the QP problem defined in the given QPSolverWrapper More...

Private Attributes
std::vector< std::vector< slackEnhanceTuple > * >	PNetId2SlackEnhanceTuples

PlacementInfo *	placementInfo

QPSolverWrapper *	xSolver = nullptr

QPSolverWrapper *	ySolver = nullptr

std::map< std::string, std::string > &	JSONCfg

bool	verbose

float	generalNetWeight = 1.0
	a common factor indicate the overall strength of the nets in the QP model from external setting More...

float	generalTimingNetWeight = 1.0

double	slackPowerFactor = 1.1

float	y2xRatio = 1.0
	a factor to tune the weights of the net spanning in Y-coordinate relative to the net spanning in X-coordinate More...

bool	useUnconstrainedCG = true
	indicate whether wirelength optimizer uses Eigen3, which cannot solve QP problem with constraints. If false, OSQP solver which can set constraints for the quadratic model, will be involved to replace Eigen3. More...

bool	MKLorNot = false
	indicate whether wirelength optimizer is based on MKL library when using OSQP placer, which can set constraints for the quadratic model More...

bool	directMacroLegalize = false
	indicate whether we use direct macro legalization instread of the progressive legalization (2-phase legalization) More...

bool	DSPCritical = false

float	pin2pinEnhance = 5.0

float	userDefinedClusterFadeOutFactor = 1.0
	the fade-out factor for the user-defined clusters More...

float	oriMacroLegalizationWeight = 1.0
	the net weight for macro legalization More...

int	macroPseudoNetCnt = 0
	the number of the conducted macro iterations More...

std::map< DesignInfo::DesignNet *, std::vector< float > >	netPinEnhanceRate

float	slackThr = 0

std::string	targetCellName = "design_1_i/DigitRec_0/inst/ap_phi_reg_pp2_iter31_knn_set_392_1_reg_27413_reg[20]"

int	targetCellId = -1

Detailed Description

WirelengthOptimizer builds numerical models based on the element locations and calls solvers to find an optimal solution of the placement.

Definition at line 53 of file WirelengthOptimizer.h.

Member Typedef Documentation

◆ CellWithScore

typedef struct WirelengthOptimizer::_CellWithScore WirelengthOptimizer::CellWithScore

private

◆ PUWithScore

typedef struct WirelengthOptimizer::_PUWithScore WirelengthOptimizer::PUWithScore

private

◆ slackEnhanceTuple

typedef struct WirelengthOptimizer::_slackEnhanceTuple WirelengthOptimizer::slackEnhanceTuple

private

Constructor & Destructor Documentation

◆ WirelengthOptimizer()

WirelengthOptimizer::WirelengthOptimizer	(	PlacementInfo *	placementInfo,
		std::map< std::string, std::string > &	JSONCfg,
		bool	verbose = `true`
	)

Construct a new Wirelength Optimizer object.

Parameters

placementInfo	the PlacementInfo for this placer to handle
JSONCfg	the user-defined placement configuration
verbose	whether the WirelengthOptimizer prints out more information or dumps log files

Definition at line 31 of file WirelengthOptimizer.cc.

Here is the call graph for this function:

◆ ~WirelengthOptimizer()

WirelengthOptimizer::~WirelengthOptimizer ( )

inline

Definition at line 64 of file WirelengthOptimizer.h.

Member Function Documentation

◆ addPseudoNet2LoctionForAllPUs()

void WirelengthOptimizer::addPseudoNet2LoctionForAllPUs	(	float	pesudoNetWeight,
		bool	considerNetNum
	)

private

as convential QP placers do, we add pseudo nets between anchors and PlacementUnit to constrain the movement of PlacementUnit for WL optimization.

Please note that we can add the interconnect-aware factor into the pseudo nets so the macros can move slower compared to the fine-grained elements.

Parameters

pesudoNetWeight	the common weight factor for pseudo nets
considerNetNum	whether add the interconnection-density-aware factor to pseudo net weights

Definition at line 821 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ addPseudoNet_SlackBased()

void WirelengthOptimizer::addPseudoNet_SlackBased	(	float	timingWeight,
		double	slackPowFactor,
		PlacementTimingOptimizer *	timingOptimizer,
		bool	calculate = `false`
	)

private

add pseudo net for timing optimization based on the timing slack of each elements in the design netlist

Parameters

timingWeight	the common weight factor for pseudo nets for timing
slackPowFactor	a factor for the sensitivity of negative timing slack
timingOptimizer	the handler of timing-related analysis

Definition at line 434 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacementQPSolve(), and updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ addPseudoNetForMacros()

void WirelengthOptimizer::addPseudoNetForMacros	(	float	pesudoNetWeight,
		bool	considerNetNum
	)

private

add the legalization pseudo nets to force macros move to the legal sites

Parameters

pesudoNetWeight	the common weight factor for pseudo nets
considerNetNum	whether add the interconnection-density-aware factor to pseudo net weights

Definition at line 356 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ clearNetPinEnhanceRate()

void WirelengthOptimizer::clearNetPinEnhanceRate ( )

inline

Definition at line 181 of file WirelengthOptimizer.h.

◆ getCellDistance()

float WirelengthOptimizer::getCellDistance	(	PlacementInfo::Location &	A,
		PlacementInfo::Location &	B
	)

inlineprivate

Definition at line 305 of file WirelengthOptimizer.h.

Referenced by LUTLUTPairing_TimingDriven().

Here is the caller graph for this function:

◆ getMacroLegalizationWeight()

float WirelengthOptimizer::getMacroLegalizationWeight ( )

inline

Get the net weight for macro legalization.

Returns: float

Definition at line 176 of file WirelengthOptimizer.h.

Referenced by GlobalPlacer::getMacroLegalizationWeight(), and GlobalPlacer::GlobalPlacement_CLBElements().

Here is the caller graph for this function:

◆ getMacroPseudoNetEnhanceCnt()

int WirelengthOptimizer::getMacroPseudoNetEnhanceCnt ( )

inline

Get the number of the conducted macro iterations.

Returns: int

Definition at line 166 of file WirelengthOptimizer.h.

Referenced by GlobalPlacer::getMacroPseudoNetEnhanceCnt(), and GlobalPlacer::GlobalPlacement_CLBElements().

Here is the caller graph for this function:

◆ GlobalPlacementQPSolve()

void WirelengthOptimizer::GlobalPlacementQPSolve	(	float	pesudoNetWeight,
		bool	firstIteration = `true`,
		bool	forwardSolutionToNextIteration = `false`,
		bool	enableMacroPseudoNet2Site = `false`,
		bool	considerNetNum = `true`,
		bool	enableUserDefinedClusterOpt = `false`,
		float	displacementLimit = `-10`,
		PlacementTimingOptimizer *	timingOptimizer = `nullptr`
	)

use quadratic model to estimate the HPWL and some timing/user-defined pseudo nets are involved for specific optimization

Parameters

pesudoNetWeight	the common weight factor for pseudo nets
firstIteration	indicates whether this is the first iteration of current round of WL optimization to control the data loading
forwardSolutionToNextIteration	if this round of WL optimization includes multiple iterations, do we forward the location data obtained in last iteration for later processing?
enableMacroPseudoNet2Site	enable the legalization pseudo net to force macros move to the legal site
considerNetNum	whether add the interconnection-density-aware factor to pseudo net weights
enableUserDefinedClusterOpt	whether check user-defined cluster information to add pseudo nets
timingOpt	whether turn on the timing-oriented optimizations

Definition at line 100 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacer::GlobalPlacement_CLBElements(), GlobalPlacer::GlobalPlacement_fixedCLB(), and ParallelCLBPacker::packCLBs().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ LUTLUTPairing_TimingDriven()

void WirelengthOptimizer::LUTLUTPairing_TimingDriven	(	float	timingWeight,
		float	disThreshold,
		PlacementTimingOptimizer *	timingOptimizer
	)

private

Definition at line 666 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ manhattanDis()

double WirelengthOptimizer::manhattanDis	(	double	x0,
		double	y0,
		double	x1,
		double	y1
	)

inlineprivate

evaluate the Mahattan distance between two locations

Parameters

x0
y0
x1
y1

Returns: double

Definition at line 301 of file WirelengthOptimizer.h.

◆ QPSolve()

static void WirelengthOptimizer::QPSolve ( QPSolverWrapper *& curSolver )

staticprivate

call the cooresponding solver to solve the QP problem defined in the given QPSolverWrapper

Parameters

curSolver a given QPSolverWrapper

◆ reloadPlacementInfo()

void WirelengthOptimizer::reloadPlacementInfo ( )

re-initialize some parameters and optimizer configuration according to the PlacementInfo

PlacementInfo might change due to the loading of checkpoint or re-initialization of bin grid. Therefore, we need to let the WL optimizer to know the changes.

Definition at line 69 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacer::GlobalPlacement_CLBElements(), and GlobalPlacer::GlobalPlacement_fixedCLB().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ setGeneralTimingNetWeight()

void WirelengthOptimizer::setGeneralTimingNetWeight ( float _generalTimingNetWeight )

inline

Definition at line 186 of file WirelengthOptimizer.h.

◆ setMacroLegalizationParameters()

void WirelengthOptimizer::setMacroLegalizationParameters	(	int	cnt,
		float	macroLegalizationWeight
	)

inline

Set the macro legalization parameters.

The convergence of the placement is determined by some parameters and when we load the checkpoint from files, we also need to load the corresponding parameters for the optimizers.

Parameters

cnt	the number of the conducted macro iterations
macroLegalizationWeight	the net weight for macro legalization

Definition at line 155 of file WirelengthOptimizer.h.

Referenced by GlobalPlacer::setMacroLegalizationParameters().

Here is the caller graph for this function:

◆ setSlackPowerFactor()

void WirelengthOptimizer::setSlackPowerFactor ( double _slackPowerFactor )

inline

Definition at line 191 of file WirelengthOptimizer.h.

◆ solverLoadData()

void WirelengthOptimizer::solverLoadData ( )

private

load to placement location from PlacementInfo to the solver

Definition at line 142 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacementQPSolve().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ solverLoadFixedData()

void WirelengthOptimizer::solverLoadFixedData ( )

private

load to placement location of fixed PlacementUnits from PlacementInfo to the solver

Definition at line 152 of file WirelengthOptimizer.cc.

Here is the call graph for this function:

◆ solverWriteBackData()

void WirelengthOptimizer::solverWriteBackData ( float displacementLimit )

private

write placement location from the solver to the PlacementInfo

Definition at line 181 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacementQPSolve().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ updateB2BNetWeight()

void WirelengthOptimizer::updateB2BNetWeight	(	float	pesudoNetWeight,
		bool	enableMacroPseudoNet2Site = `false`,
		bool	considerNetNum = `true`,
		bool	enableUserDefinedClusterOpt = `false`,
		PlacementTimingOptimizer *	timingOptimizer = `nullptr`
	)

update the net weights in the quadratic model according to B2B net HPWL model.

In this procedure, we will not only consider the conventional nets/pseudonets in previous QP placer, but also include the legalization pseudo nets and timing-oriented pseudo nets to comprehensively optimize the placement in this part. Meanwhile, we allow users to define some clusters and let the WL optimizer take them into consideration during optimization.

Parameters

pesudoNetWeight	the common weight factor for pseudo nets
enableMacroPseudoNet2Site	enable the legalization pseudo net to force macros move to the legal site
considerNetNum	whether add the interconnection-density-aware factor to pseudo net weights
enableUserDefinedClusterOpt	whether check user-defined cluster information to add pseudo nets
timingOpt	whether turn on the timing-oriented optimizations

Definition at line 268 of file WirelengthOptimizer.cc.

Referenced by GlobalPlacementQPSolve().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ updateB2BNetWeightWorker()

void WirelengthOptimizer::updateB2BNetWeightWorker	(	PlacementInfo *	placementInfo,
		std::vector< Eigen::Triplet< float >> &	objectiveMatrixTripletList,
		std::vector< float > &	objectiveMatrixDiag,
		Eigen::VectorXd &	objectiveVector,
		float	generalNetWeight,
		float	y2xRatio,
		bool	updateX,
		bool	updateY
	)

static

a worker funtion for multi-threading net weight updating

min_x 0.5 * x'Px + q'x s.t. l <= Ax <= u

Parameters

placementInfo	the PlacementInfo for this placer to handle
objectiveMatrixTripletList	a list of Eigen::Triplet used to describe the sparse matrix P of QP model
objectiveMatrixDiag	a vector used to store the diagonal values in the sparse matrix P of QP model
objectiveVector	a vector (i.e., Eigen::VectorXd) used to store the linear factors q' of QP model
generalNetWeight	a common factor indicate the overall strength of the nets in the QP model from external setting
y2xRatio	a factor to tune the weights of the net spanning in Y-coordinate relative to the net spanning in X-coordinate
updateX	update the X-coordinate term in the quadratic problem
updateY	update the X-coordinate term in the quadratic problem

Definition at line 334 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ updatePseudoNetForClockRegion()

void WirelengthOptimizer::updatePseudoNetForClockRegion ( float pesudoNetWeight )

private

add pseudo nets for clock region

we find that when the nets route across the boundaries of clock regions will lead to high delay, therefore, we will try to force the long path located in the same clock regions for the timing optimization.

Parameters

pesudoNetWeight the common weight factor for pseudo nets

Definition at line 1043 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ updatePseudoNetForUserDefinedClusters()

void WirelengthOptimizer::updatePseudoNetForUserDefinedClusters ( float pesudoNetWeight )

private

add pseudo net for user-defined clusters

We allow users to specify some clusters in the design and WL optimizer will try to push them togather during solving the QP problem

Parameters

pesudoNetWeight the common weight factor for pseudo nets

Definition at line 955 of file WirelengthOptimizer.cc.

Referenced by updateB2BNetWeight().

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

◆ directMacroLegalize

bool WirelengthOptimizer::directMacroLegalize = false

private

indicate whether we use direct macro legalization instread of the progressive legalization (2-phase legalization)

If we conduct direct macro legalization, pseudo net for legalization will be removed and macro will be placed to the cooresponding location according to the legalization result.

Definition at line 374 of file WirelengthOptimizer.h.

Referenced by updateB2BNetWeight(), and WirelengthOptimizer().

◆ DSPCritical

bool WirelengthOptimizer::DSPCritical = false

private

Definition at line 376 of file WirelengthOptimizer.h.

Referenced by updatePseudoNetForClockRegion(), and WirelengthOptimizer().

◆ generalNetWeight

float WirelengthOptimizer::generalNetWeight = 1.0

private

a common factor indicate the overall strength of the nets in the QP model from external setting

Definition at line 341 of file WirelengthOptimizer.h.

Referenced by updateB2BNetWeight(), and updateB2BNetWeightWorker().

◆ generalTimingNetWeight

float WirelengthOptimizer::generalTimingNetWeight = 1.0

private

Definition at line 342 of file WirelengthOptimizer.h.

Referenced by GlobalPlacementQPSolve(), setGeneralTimingNetWeight(), and updateB2BNetWeight().

◆ JSONCfg

std::map<std::string, std::string>& WirelengthOptimizer::JSONCfg

private

Definition at line 333 of file WirelengthOptimizer.h.

Referenced by updateB2BNetWeight(), and WirelengthOptimizer().

◆ macroPseudoNetCnt

int WirelengthOptimizer::macroPseudoNetCnt = 0

private

the number of the conducted macro iterations

Definition at line 399 of file WirelengthOptimizer.h.

Referenced by addPseudoNetForMacros(), getMacroPseudoNetEnhanceCnt(), and setMacroLegalizationParameters().

◆ MKLorNot

bool WirelengthOptimizer::MKLorNot = false

private

indicate whether wirelength optimizer is based on MKL library when using OSQP placer, which can set constraints for the quadratic model

Definition at line 364 of file WirelengthOptimizer.h.

Referenced by reloadPlacementInfo(), and WirelengthOptimizer().

◆ netPinEnhanceRate

std::map<DesignInfo::DesignNet *, std::vector<float> > WirelengthOptimizer::netPinEnhanceRate

private

Definition at line 401 of file WirelengthOptimizer.h.

Referenced by addPseudoNet_SlackBased(), clearNetPinEnhanceRate(), and reloadPlacementInfo().

◆ oriMacroLegalizationWeight

float WirelengthOptimizer::oriMacroLegalizationWeight = 1.0

private

the net weight for macro legalization

Definition at line 393 of file WirelengthOptimizer.h.

Referenced by addPseudoNetForMacros(), getMacroLegalizationWeight(), and setMacroLegalizationParameters().

◆ pin2pinEnhance

float WirelengthOptimizer::pin2pinEnhance = 5.0

private

Definition at line 378 of file WirelengthOptimizer.h.

Referenced by updateB2BNetWeight(), and WirelengthOptimizer().

◆ placementInfo

PlacementInfo* WirelengthOptimizer::placementInfo

private

Definition at line 329 of file WirelengthOptimizer.h.

Referenced by addPseudoNet2LoctionForAllPUs(), addPseudoNet_SlackBased(), addPseudoNetForMacros(), LUTLUTPairing_TimingDriven(), reloadPlacementInfo(), solverLoadData(), solverLoadFixedData(), solverWriteBackData(), updateB2BNetWeight(), updateB2BNetWeightWorker(), updatePseudoNetForClockRegion(), updatePseudoNetForUserDefinedClusters(), and WirelengthOptimizer().

◆ PNetId2SlackEnhanceTuples

std::vector<std::vector<slackEnhanceTuple> *> WirelengthOptimizer::PNetId2SlackEnhanceTuples

private

Definition at line 267 of file WirelengthOptimizer.h.

Referenced by addPseudoNet_SlackBased(), and ~WirelengthOptimizer().

◆ slackPowerFactor

double WirelengthOptimizer::slackPowerFactor = 1.1

private

Definition at line 343 of file WirelengthOptimizer.h.

Referenced by GlobalPlacementQPSolve(), LUTLUTPairing_TimingDriven(), setSlackPowerFactor(), and updateB2BNetWeight().

◆ slackThr

float WirelengthOptimizer::slackThr = 0

private

Definition at line 403 of file WirelengthOptimizer.h.

Referenced by addPseudoNet_SlackBased(), and GlobalPlacementQPSolve().

◆ targetCellId

int WirelengthOptimizer::targetCellId = -1

private

Definition at line 407 of file WirelengthOptimizer.h.

Referenced by addPseudoNet_SlackBased(), LUTLUTPairing_TimingDriven(), and reloadPlacementInfo().

◆ targetCellName

std::string WirelengthOptimizer::targetCellName = "design_1_i/DigitRec_0/inst/ap_phi_reg_pp2_iter31_knn_set_392_1_reg_27413_reg[20]"

private

Definition at line 406 of file WirelengthOptimizer.h.

Referenced by addPseudoNet_SlackBased(), and reloadPlacementInfo().

◆ userDefinedClusterFadeOutFactor

float WirelengthOptimizer::userDefinedClusterFadeOutFactor = 1.0

private

the fade-out factor for the user-defined clusters

The user-defined clusters might not be perfect so we might need to gradually reduce their terms in the QP problem by this factor

Definition at line 387 of file WirelengthOptimizer.h.

Referenced by updatePseudoNetForUserDefinedClusters().

◆ useUnconstrainedCG

bool WirelengthOptimizer::useUnconstrainedCG = true

private

indicate whether wirelength optimizer uses Eigen3, which cannot solve QP problem with constraints. If false, OSQP solver which can set constraints for the quadratic model, will be involved to replace Eigen3.

Definition at line 357 of file WirelengthOptimizer.h.

Referenced by reloadPlacementInfo(), and WirelengthOptimizer().

◆ verbose

bool WirelengthOptimizer::verbose

private

Definition at line 334 of file WirelengthOptimizer.h.

Referenced by GlobalPlacementQPSolve(), reloadPlacementInfo(), updateB2BNetWeight(), and WirelengthOptimizer().

◆ xSolver

QPSolverWrapper* WirelengthOptimizer::xSolver = nullptr

private

Definition at line 331 of file WirelengthOptimizer.h.

Referenced by addPseudoNet2LoctionForAllPUs(), addPseudoNet_SlackBased(), addPseudoNetForMacros(), GlobalPlacementQPSolve(), LUTLUTPairing_TimingDriven(), reloadPlacementInfo(), solverLoadData(), solverLoadFixedData(), solverWriteBackData(), updateB2BNetWeight(), updatePseudoNetForClockRegion(), updatePseudoNetForUserDefinedClusters(), WirelengthOptimizer(), and ~WirelengthOptimizer().

◆ y2xRatio

float WirelengthOptimizer::y2xRatio = 1.0

private

a factor to tune the weights of the net spanning in Y-coordinate relative to the net spanning in X-coordinate

Definition at line 350 of file WirelengthOptimizer.h.

Referenced by addPseudoNet2LoctionForAllPUs(), addPseudoNet_SlackBased(), addPseudoNetForMacros(), getCellDistance(), LUTLUTPairing_TimingDriven(), manhattanDis(), updateB2BNetWeight(), updateB2BNetWeightWorker(), updatePseudoNetForClockRegion(), updatePseudoNetForUserDefinedClusters(), and WirelengthOptimizer().

◆ ySolver

QPSolverWrapper* WirelengthOptimizer::ySolver = nullptr

private

Definition at line 332 of file WirelengthOptimizer.h.

Referenced by addPseudoNet2LoctionForAllPUs(), addPseudoNet_SlackBased(), addPseudoNetForMacros(), GlobalPlacementQPSolve(), LUTLUTPairing_TimingDriven(), reloadPlacementInfo(), solverLoadData(), solverLoadFixedData(), solverWriteBackData(), updateB2BNetWeight(), updatePseudoNetForUserDefinedClusters(), WirelengthOptimizer(), and ~WirelengthOptimizer().

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

src/lib/HiFPlacer/placement/globalPlacement/WirelengthOptimizer.h
src/lib/HiFPlacer/placement/globalPlacement/WirelengthOptimizer.cc

Classes

Public Member Functions

Static Public Member Functions

Private Types

Private Member Functions

Static Private Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

◆ CellWithScore

◆ PUWithScore

◆ slackEnhanceTuple

Constructor & Destructor Documentation

◆ WirelengthOptimizer()

◆ ~WirelengthOptimizer()

Member Function Documentation

◆ addPseudoNet2LoctionForAllPUs()

◆ addPseudoNet_SlackBased()

◆ addPseudoNetForMacros()

◆ clearNetPinEnhanceRate()

◆ getCellDistance()

◆ getMacroLegalizationWeight()

◆ getMacroPseudoNetEnhanceCnt()

◆ GlobalPlacementQPSolve()

◆ LUTLUTPairing_TimingDriven()

◆ manhattanDis()

◆ QPSolve()

◆ reloadPlacementInfo()

◆ setGeneralTimingNetWeight()

◆ setMacroLegalizationParameters()

◆ setSlackPowerFactor()

◆ solverLoadData()

◆ solverLoadFixedData()

◆ solverWriteBackData()

◆ updateB2BNetWeight()

◆ updateB2BNetWeightWorker()

◆ updatePseudoNetForClockRegion()

◆ updatePseudoNetForUserDefinedClusters()

Member Data Documentation

◆ directMacroLegalize

◆ DSPCritical

◆ generalNetWeight

◆ generalTimingNetWeight

◆ JSONCfg

◆ macroPseudoNetCnt

◆ MKLorNot

◆ netPinEnhanceRate

◆ oriMacroLegalizationWeight

◆ pin2pinEnhance

◆ placementInfo

◆ PNetId2SlackEnhanceTuples

◆ slackPowerFactor

◆ slackThr

◆ targetCellId

◆ targetCellName

◆ userDefinedClusterFadeOutFactor

◆ useUnconstrainedCG

◆ verbose

◆ xSolver

◆ y2xRatio

◆ ySolver