MicroGPU Class Reference

Represents a microGPUThe MicroGPU class encapsulates the state and behavior of a microGPU, which consists of multiple compute units (CUs) that execute warps of threads. In this project miroGPU has 16 compute units, each capable of executing one warp at a time. MicroGPU schedules warps from its global warp collection to the compute units and then Compute Units execute the warps in a simple round-robin fashion. More...

#include <ugpu.hh>

Collaboration diagram for MicroGPU:

Public Member Functions
	MicroGPU ()
void	init ()
void	assignWarpToSM (int smId, const Warp &warp)
void	addWarpToGlobalCollection (const Warp &warp)
	Add a warp to the global warp collection.
int	getGlobalWarpCollectionSize () const
int	getCurrentCycle () const
void	incrementCycle ()
void	performWarpScheduling ()
void	performWarpSchedulingSimple ()
	Perform simple warp scheduling This function assigns alternating warps from the global collection to the first two compute units in a round-robin fashion. Just for testing purposes to verify that warps are being scheduled and executed correctly. In real life, all Compute Units would be utilized.
void	setMaxCycles (int cycles)
void	createGlobalWarpCollectionTest ()
	Create a Global Warp Collection Test object This method creates a set of test warps with simple instructions and adds them to the global warp collection for testing purposes. Each warp is initialized with a simple ADD instruction and all threads in the warp are set to active. Additionally, one warp with a branch instruction is added to test divergence handling in the compute units.
void	printComputeUnitStatus () const
bool	allComputeUnitsDone () const
bool	allWarpsCompleted () const
void	executeComputeUnits ()
	Execute all compute units for one cycle.
void	executeGPU ()
	Execute the entire GPU until all warps are completed.

Private Attributes
std::vector< Warp >	globalWarpCollection
std::array< ComputeUnit, CU_COUNT >	computeUnit
int	currentCycle
int	maxCycles

Detailed Description

Represents a microGPU

The MicroGPU class encapsulates the state and behavior of a microGPU, which consists of multiple compute units (CUs) that execute warps of threads. In this project miroGPU has 16 compute units, each capable of executing one warp at a time. MicroGPU schedules warps from its global warp collection to the compute units and then Compute Units execute the warps in a simple round-robin fashion.

Constructor & Destructor Documentation

MicroGPU()

MicroGPU::MicroGPU

(

)

Member Function Documentation

addWarpToGlobalCollection()

void MicroGPU::addWarpToGlobalCollection

(

const Warp &

warp

)

Add a warp to the global warp collection.

Parameters

warp	The warp to be added to the global collection

Here is the call graph for this function:

allComputeUnitsDone()

bool MicroGPU::allComputeUnitsDone

(

)

const

allWarpsCompleted()

bool MicroGPU::allWarpsCompleted

(

)

const

Here is the call graph for this function:

assignWarpToSM()

void MicroGPU::assignWarpToSM	(	int	smId,
		const Warp &	warp
	)

Here is the call graph for this function:

createGlobalWarpCollectionTest()

void MicroGPU::createGlobalWarpCollectionTest

(

)

Create a Global Warp Collection Test object This method creates a set of test warps with simple instructions and adds them to the global warp collection for testing purposes. Each warp is initialized with a simple ADD instruction and all threads in the warp are set to active. Additionally, one warp with a branch instruction is added to test divergence handling in the compute units.

Here is the call graph for this function:

executeComputeUnits()

void MicroGPU::executeComputeUnits

(

)

Execute all compute units for one cycle.

executeGPU()

void MicroGPU::executeGPU

(

)

Execute the entire GPU until all warps are completed.

Here is the call graph for this function:

getCurrentCycle()

int MicroGPU::getCurrentCycle ( ) const

inline

getGlobalWarpCollectionSize()

int MicroGPU::getGlobalWarpCollectionSize

(

)

const

incrementCycle()

void MicroGPU::incrementCycle ( )

inline

init()

void MicroGPU::init

(

)

performWarpScheduling()

void MicroGPU::performWarpScheduling

(

)

Here is the call graph for this function:

performWarpSchedulingSimple()

void MicroGPU::performWarpSchedulingSimple

(

)

Perform simple warp scheduling This function assigns alternating warps from the global collection to the first two compute units in a round-robin fashion. Just for testing purposes to verify that warps are being scheduled and executed correctly. In real life, all Compute Units would be utilized.

Here is the call graph for this function:

printComputeUnitStatus()

void MicroGPU::printComputeUnitStatus

(

)

const

setMaxCycles()

void MicroGPU::setMaxCycles ( int  cycles )

inline

Member Data Documentation

computeUnit

std::array<ComputeUnit, CU_COUNT> MicroGPU::computeUnit

private

currentCycle

int MicroGPU::currentCycle

private

globalWarpCollection

std::vector<Warp> MicroGPU::globalWarpCollection

private

maxCycles

int MicroGPU::maxCycles

private

---

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

• src/microGPU/ugpu.hh
• src/microGPU/ugpu.cc