#include <global_memory_pipeline.hh>

Public Member Functions
	GlobalMemPipeline (const ComputeUnitParams *params)

void	init (ComputeUnit *cu)

void	exec ()

std::queue< GPUDynInstPtr > &	getGMStRespFIFO ()

std::queue< GPUDynInstPtr > &	getGMLdRespFIFO ()

GPUDynInstPtr	getNextReadyResp ()
	find the next ready response to service. More...

void	completeRequest (GPUDynInstPtr gpuDynInst)
	once a memory request is finished we remove it from the buffer. More...

void	issueRequest (GPUDynInstPtr gpuDynInst)
	issues a request to the pipeline - i.e., enqueue it in the request buffer. More...

void	handleResponse (GPUDynInstPtr gpuDynInst)
	this method handles responses sent to this GM pipeline by the CU. More...

bool	isGMLdRespFIFOWrRdy () const

bool	isGMStRespFIFOWrRdy () const

bool	isGMReqFIFOWrRdy (uint32_t pendReqs=0) const

const std::string &	name () const

void	regStats ()

void	incLoadVRFBankConflictCycles (int num_cycles)

bool	coalescerReady (GPUDynInstPtr mp) const

Private Attributes
ComputeUnit *	computeUnit

std::string	_name

int	gmQueueSize

bool	outOfOrderDataDelivery

Stats::Scalar	loadVrfBankConflictCycles

int	inflightStores

int	inflightLoads

int	globalMemSize

std::map< uint64_t, std::pair< GPUDynInstPtr, bool > >	gmOrderedRespBuffer

std::queue< GPUDynInstPtr >	gmIssuedRequests

std::queue< GPUDynInstPtr >	gmReturnedStores

std::queue< GPUDynInstPtr >	gmReturnedLoads

Detailed Description

Definition at line 56 of file global_memory_pipeline.hh.

Constructor & Destructor Documentation

◆ GlobalMemPipeline()

GlobalMemPipeline::GlobalMemPipeline ( const ComputeUnitParams * params )

Definition at line 45 of file global_memory_pipeline.cc.

Member Function Documentation

◆ coalescerReady()

bool GlobalMemPipeline::coalescerReady ( GPUDynInstPtr mp ) const

Definition at line 61 of file global_memory_pipeline.cc.

References DPRINTF.

Referenced by incLoadVRFBankConflictCycles(), and Wavefront::ready().

◆ completeRequest()

void GlobalMemPipeline::completeRequest ( GPUDynInstPtr gpuDynInst )

once a memory request is finished we remove it from the buffer.

this method determines which response buffer we're using based on the mode (in-order vs. OoO).

Definition at line 201 of file global_memory_pipeline.cc.

References gmOrderedRespBuffer, gmReturnedLoads, gmReturnedStores, inflightLoads, inflightStores, and outOfOrderDataDelivery.

Referenced by exec(), and getGMLdRespFIFO().

◆ exec()

void GlobalMemPipeline::exec ( )

if we are not in out-of-order data delivery mode then we keep the responses sorted in program order. in order to do so we must reserve an entry in the resp buffer before we issue the request to the mem system. mem fence requests will not be stored here because once they are issued from the GM pipeline, they do not send any response back to it.

Definition at line 80 of file global_memory_pipeline.cc.

References Shader::coissue_return, completeRequest(), computeUnit, Wavefront::computeUnit, ComputeUnit::cu_id, DPRINTF, getNextReadyResp(), ComputeUnit::glbMemToVrfBus, gmIssuedRequests, gmOrderedRespBuffer, gmQueueSize, inflightLoads, inflightStores, ArmISA::m, X86ISA::mp, outOfOrderDataDelivery, Wavefront::outstandingReqs, Wavefront::outstandingReqsRdGm, Wavefront::outstandingReqsWrGm, WaitClass::rdy(), Shader::ScheduleAdd(), WaitClass::set(), ComputeUnit::shader, Wavefront::simdId, ComputeUnit::vrf, MipsISA::w, ComputeUnit::wfWait, and WRITE.

Referenced by ComputeUnit::exec().

◆ getGMLdRespFIFO()

std::queue<GPUDynInstPtr>& GlobalMemPipeline::getGMLdRespFIFO ( )

inline

Definition at line 64 of file global_memory_pipeline.hh.

References completeRequest(), getNextReadyResp(), gmReturnedLoads, handleResponse(), and issueRequest().

◆ getGMStRespFIFO()

std::queue<GPUDynInstPtr>& GlobalMemPipeline::getGMStRespFIFO ( )

inline

Definition at line 63 of file global_memory_pipeline.hh.

References gmReturnedStores.

◆ getNextReadyResp()

GPUDynInstPtr GlobalMemPipeline::getNextReadyResp ( )

find the next ready response to service.

for OoO mode we simply pop the oldest (based on when the response was received) response in the response FIFOs. for in-order mode we pop the oldest (in program order) response, and only if it is marked as done.

Definition at line 179 of file global_memory_pipeline.cc.

References gmOrderedRespBuffer, gmReturnedLoads, gmReturnedStores, and outOfOrderDataDelivery.

Referenced by exec(), and getGMLdRespFIFO().

◆ handleResponse()

void GlobalMemPipeline::handleResponse ( GPUDynInstPtr gpuDynInst )

this method handles responses sent to this GM pipeline by the CU.

in the case of in-order delivery it simply marks the reqeust as done in the ordered buffer to indicate that the requst is finished. for out-of-order data delivery, the requests are enqueued (in the order in which they are received) in the response FIFOs.

Definition at line 238 of file global_memory_pipeline.cc.

References gmOrderedRespBuffer, gmReturnedLoads, gmReturnedStores, isGMLdRespFIFOWrRdy(), isGMStRespFIFOWrRdy(), and outOfOrderDataDelivery.

Referenced by getGMLdRespFIFO(), and ComputeUnit::DataPort::processMemRespEvent().