Warp Class Reference

Represents a GPU warpThe Warp class encapsulates the state and behavior of a GPU warp, including its ID, program counter, threads, active mask, and current instruction. A warp is a group of threads that execute instructions in lockstep. Each warp has 32 threads in this project. To be added: reconvergence stack to handle divergent control flow. More...

#include <warp.hh>

Collaboration diagram for Warp:

Public Member Functions
	Warp ()
	Warp (int warpId, const ThreadGroup &threadGroup, WarpState warpState=WarpState::READY, PipelineStage pipelineStage=PipelineStage::NOT_STARTED, bool isBranchDivergent=false)
int	getId () const
int	getPc () const
void	setPc (int pc_)
void	setCurrentInstruction (const Instruction &instr)
Instruction	getCurrentInstruction () const
PipelineStage	getPipelineStage () const
void	setPipelineStage (PipelineStage stage)
void	setActiveMask (const ActiveMask &mask)
std::string	getPipelineStageString () const
const ActiveMask &	getActiveMask () const
WarpState	getState () const
bool	isDivergent () const
void	setDivergent ()
void	execute ()
std::string	getInstructionTypeString (Instruction inst) const
void	addReconvergencePoint (int pc, const ActiveMask &mask)
	Add a reconvergence point to the stack.
void	peekReconvergencePoint () const
	Peek the top reconvergence point on the stack This method is used for debugging purposes to check the current reconvergence points for a warp.
void	popReconvergencePoint ()
	Pop the top reconvergence point from the stack This method is called when a reconvergence point is reached during execution to restore the active mask and program counter for the warp.
size_t	getReconvergenceStackSize () const
std::bitset< WARP_THREAD_COUNT >	getActiveMaskFromReconvergenceStack () const
	Get the active mask from the top reconvergence point on the stack.

Private Attributes
int	id
int	pc
ThreadGroup	threads
ActiveMask	activeMask
Instruction	currentInstruction
std::vector< reconvergencePoint >	reconvergenceStack
	Reconvergence stack for handling divergent control flow.
WarpState	state
PipelineStage	pipelineStage
bool	isBranchDivergent = false

Detailed Description

Represents a GPU warp

The Warp class encapsulates the state and behavior of a GPU warp, including its ID, program counter, threads, active mask, and current instruction. A warp is a group of threads that execute instructions in lockstep. Each warp has 32 threads in this project. To be added: reconvergence stack to handle divergent control flow.

Constructor & Destructor Documentation

Warp() [1/2]

Warp::Warp

(

)

Warp() [2/2]

Warp::Warp	(	int	warpId,
		const ThreadGroup &	threadGroup,
		WarpState	warpState = WarpState::READY,
		PipelineStage	pipelineStage = PipelineStage::NOT_STARTED,
		bool	isBranchDivergent = false
	)

Member Function Documentation

addReconvergencePoint()

void Warp::addReconvergencePoint	(	int	pc,
		const ActiveMask &	mask
	)

Add a reconvergence point to the stack.

Parameters

pc	The program counter value at the reconvergence point
mask	The active mask at the reconvergence point This method is called when a divergent branch is encountered to save the reconvergence point information on the stack.

execute()

void Warp::execute

(

)

Here is the call graph for this function:

getActiveMask()

const ActiveMask & Warp::getActiveMask

(

)

const

getActiveMaskFromReconvergenceStack()

std::bitset< WARP_THREAD_COUNT > Warp::getActiveMaskFromReconvergenceStack

(

)

const

Get the active mask from the top reconvergence point on the stack.

Returns

ActiveMask The active mask from the top reconvergence point

getCurrentInstruction()

Instruction Warp::getCurrentInstruction

(

)

const

getId()

int Warp::getId

(

)

const

getInstructionTypeString()

std::string Warp::getInstructionTypeString

(

Instruction

inst

)

const

getPc()

int Warp::getPc

(

)

const

getPipelineStage()

PipelineStage Warp::getPipelineStage ( ) const

inline

getPipelineStageString()

std::string Warp::getPipelineStageString

(

)

const

getReconvergenceStackSize()

size_t Warp::getReconvergenceStackSize

(

)

const

getState()

WarpState Warp::getState ( ) const

inline

isDivergent()

bool Warp::isDivergent ( ) const

inline

peekReconvergencePoint()

void Warp::peekReconvergencePoint

(

)

const

Peek the top reconvergence point on the stack This method is used for debugging purposes to check the current reconvergence points for a warp.

popReconvergencePoint()

void Warp::popReconvergencePoint

(

)

Pop the top reconvergence point from the stack This method is called when a reconvergence point is reached during execution to restore the active mask and program counter for the warp.

setActiveMask()

void Warp::setActiveMask ( const ActiveMask &  mask )

inline

setCurrentInstruction()

void Warp::setCurrentInstruction

(

const Instruction &

instr

)

setDivergent()

void Warp::setDivergent ( )

inline

setPc()

void Warp::setPc

(

int

pc_

)

setPipelineStage()

void Warp::setPipelineStage ( PipelineStage  stage )

inline

Member Data Documentation

activeMask

ActiveMask Warp::activeMask

private

currentInstruction

Instruction Warp::currentInstruction

private

id

int Warp::id

private

isBranchDivergent

bool Warp::isBranchDivergent = false

private

pc

int Warp::pc

private

pipelineStage

PipelineStage Warp::pipelineStage

private

reconvergenceStack

std::vector<reconvergencePoint> Warp::reconvergenceStack

private

Reconvergence stack for handling divergent control flow.

state

WarpState Warp::state

private

threads

ThreadGroup Warp::threads

private

---

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

• src/warp/warp.hh
• src/warp/warp.cc