#include <wavefront.hh>

Inheritance diagram for Wavefront:

Classes
struct	WavefrontStats

Public Types
enum	status_e { S_STOPPED, S_RETURNING, S_RUNNING, S_STALLED, S_STALLED_SLEEP, S_WAITCNT, S_BARRIER }

typedef WavefrontParams	Params

Public Types inherited from SimObject
typedef SimObjectParams	Params

Public Member Functions
void	freeResources ()

GPUDynInstPtr	nextInstr ()

void	setStatus (status_e newStatus)

status_e	getStatus ()

void	resizeRegFiles (int num_vregs, int num_sregs)

bool	isGmInstruction (GPUDynInstPtr ii)

bool	isLmInstruction (GPUDynInstPtr ii)

bool	isOldestInstWaitcnt ()

bool	isOldestInstSleep ()

bool	isOldestInstGMem ()

bool	isOldestInstLMem ()

bool	isOldestInstPrivMem ()

bool	isOldestInstFlatMem ()

bool	isOldestInstVectorALU ()

bool	isOldestInstScalarALU ()

bool	isOldestInstScalarMem ()

bool	isOldestInstBarrier ()

void	computeActualWgSz (HSAQueueEntry *task)

void	initRegState (HSAQueueEntry *task, int wgSizeInWorkItems)

	Wavefront (const Params &p)

	~Wavefront ()

virtual void	init ()
	init() is called after all C++ SimObjects have been created and all ports are connected. More...

void	setParent (ComputeUnit *cu)

void	validateRequestCounters ()

void	start (uint64_t _wfDynId, uint64_t _base_ptr)

void	exec ()

std::vector< int >	reserveResources ()

bool	stopFetch ()

Addr	pc () const

void	pc (Addr new_pc)

VectorMask &	execMask ()

bool	execMask (int lane) const

void	discardFetch ()

bool	waitCntsSatisfied ()

void	setWaitCnts (int vm_wait_cnt, int exp_wait_cnt, int lgkm_wait_cnt)

void	clearWaitCnts ()

void	incVMemInstsIssued ()

void	incExpInstsIssued ()

void	incLGKMInstsIssued ()

void	decVMemInstsIssued ()

void	decExpInstsIssued ()

void	decLGKMInstsIssued ()

void	freeRegisterFile ()
	Freeing VRF space. More...

bool	sleepDone ()

void	setSleepTime (int sleep_time)

TheGpuISA::GPUISA &	gpuISA ()

void	barrierId (int bar_id)

int	barrierId () const

bool	hasBarrier () const

void	releaseBarrier ()

Public Member Functions inherited from SimObject
const Params &	params () const

	SimObject (const Params &p)

virtual	~SimObject ()

virtual const std::string	name () const

virtual void	loadState (CheckpointIn &cp)
	loadState() is called on each SimObject when restoring from a checkpoint. More...

virtual void	initState ()
	initState() is called on each SimObject when not restoring from a checkpoint. More...

virtual void	regProbePoints ()
	Register probe points for this object. More...

virtual void	regProbeListeners ()
	Register probe listeners for this object. More...

ProbeManager *	getProbeManager ()
	Get the probe manager for this object. More...

virtual Port &	getPort (const std::string &if_name, PortID idx=InvalidPortID)
	Get a port with a given name and index. More...

virtual void	startup ()
	startup() is the final initialization call before simulation. More...

DrainState	drain () override
	Provide a default implementation of the drain interface for objects that don't need draining. More...

virtual void	memWriteback ()
	Write back dirty buffers to memory using functional writes. More...

virtual void	memInvalidate ()
	Invalidate the contents of memory buffers. More...

void	serialize (CheckpointOut &cp) const override
	Serialize an object. More...

void	unserialize (CheckpointIn &cp) override
	Unserialize an object. More...

Public Member Functions inherited from EventManager
EventQueue *	eventQueue () const

void	schedule (Event &event, Tick when)

void	deschedule (Event &event)

void	reschedule (Event &event, Tick when, bool always=false)

void	schedule (Event *event, Tick when)

void	deschedule (Event *event)

void	reschedule (Event *event, Tick when, bool always=false)

void	wakeupEventQueue (Tick when=(Tick) -1)
	This function is not needed by the usual gem5 event loop but may be necessary in derived EventQueues which host gem5 on other schedulers. More...

void	setCurTick (Tick newVal)

	EventManager (EventManager &em)
	Event manger manages events in the event queue. More...

	EventManager (EventManager *em)

	EventManager (EventQueue *eq)

Public Member Functions inherited from Serializable
	Serializable ()

virtual	~Serializable ()

void	serializeSection (CheckpointOut &cp, const char *name) const
	Serialize an object into a new section. More...

void	serializeSection (CheckpointOut &cp, const std::string &name) const

void	unserializeSection (CheckpointIn &cp, const char *name)
	Unserialize an a child object. More...

void	unserializeSection (CheckpointIn &cp, const std::string &name)

Public Member Functions inherited from Drainable
DrainState	drainState () const
	Return the current drain state of an object. More...

virtual void	notifyFork ()
	Notify a child process of a fork. More...

Public Member Functions inherited from Stats::Group
	Group (Group parent, const char name=nullptr)
	Construct a new statistics group. More...

virtual	~Group ()

virtual void	regStats ()
	Callback to set stat parameters. More...

virtual void	resetStats ()
	Callback to reset stats. More...

virtual void	preDumpStats ()
	Callback before stats are dumped. More...

void	addStat (Stats::Info *info)
	Register a stat with this group. More...

const std::map< std::string, Group * > &	getStatGroups () const
	Get all child groups associated with this object. More...

const std::vector< Info * > &	getStats () const
	Get all stats associated with this object. More...

void	addStatGroup (const char name, Group block)
	Add a stat block as a child of this block. More...

const Info *	resolveStat (std::string name) const
	Resolve a stat by its name within this group. More...

void	mergeStatGroup (Group *block)
	Merge the contents (stats & children) of a block to this block. More...

	Group ()=delete

	Group (const Group &)=delete

Group &	operator= (const Group &)=delete

Public Attributes
const int	wfSlotId

int	kernId

const int	simdId

int	execUnitId

int	flatLmUnitId

int	flatGmUnitId

ComputeUnit *	computeUnit

int	maxIbSize

std::deque< GPUDynInstPtr >	instructionBuffer

bool	pendingFetch

bool	dropFetch

Tick	lastNonIdleTick

int	scalarAlu

int	scalarAluGlobalIdx

int	globalMem

int	localMem

int	scalarMem

uint32_t	maxVgprs

uint32_t	maxSgprs

std::vector< Addr >	lastAddr

std::vector< uint32_t >	workItemId [3]

std::vector< uint32_t >	workItemFlatId

uint32_t	workGroupId [3]

uint32_t	workGroupSz [3]

uint32_t	gridSz [3]

uint32_t	wgId

uint32_t	wgSz

uint32_t	actualWgSz [3]

uint32_t	actualWgSzTotal

uint32_t	wfId

uint32_t	maxDynWaveId

uint32_t	dispatchId

int	outstandingReqs

int	outstandingReqsWrGm

int	outstandingReqsWrLm

int	outstandingReqsRdGm

int	outstandingReqsRdLm

int	scalarOutstandingReqsRdGm

int	scalarOutstandingReqsWrGm

int	rdLmReqsInPipe

int	rdGmReqsInPipe

int	wrLmReqsInPipe

int	wrGmReqsInPipe

int	scalarRdGmReqsInPipe

int	scalarWrGmReqsInPipe

int	memTraceBusy

uint64_t	lastTrace

int	reservedVectorRegs

int	reservedScalarRegs

uint32_t	startVgprIndex

uint32_t	startSgprIndex

std::vector< uint32_t >	oldVgpr

uint32_t	oldVgprId

uint64_t	oldVgprTcnt

std::vector< uint64_t >	oldDgpr

uint32_t	oldDgprId

uint64_t	oldDgprTcnt

VectorMask	initMask

LdsChunk *	ldsChunk

uint64_t	wfDynId

uint64_t	lastInstExec

std::unordered_map< int, uint64_t >	rawDist

std::vector< int >	vecReads

uint8_t *	context

Wavefront::WavefrontStats	stats

Private Member Functions
void	reserveGmResource (GPUDynInstPtr ii)

void	reserveLmResource (GPUDynInstPtr ii)

Private Attributes
TheGpuISA::GPUISA	_gpuISA

int	vmWaitCnt
	the following are used for waitcnt instructions vmWaitCnt: once set, we wait for the oustanding number of vector mem instructions to be at, or below vmWaitCnt. More...

int	expWaitCnt

int	lgkmWaitCnt

int	vmemInstsIssued

int	expInstsIssued

int	lgkmInstsIssued

int	sleepCnt

status_e	status

Addr	_pc

VectorMask	_execMask

int	barId

Additional Inherited Members
Static Public Member Functions inherited from SimObject
static void	serializeAll (CheckpointOut &cp)
	Serialize all SimObjects in the system. More...

static SimObject *	find (const char *name)
	Find the SimObject with the given name and return a pointer to it. More...

Static Public Member Functions inherited from Serializable
static const std::string &	currentSection ()
	Gets the fully-qualified name of the active section. More...

static void	serializeAll (const std::string &cpt_dir)
	Serializes all the SimObjects. More...

static void	unserializeGlobals (CheckpointIn &cp)

Protected Member Functions inherited from Drainable
	Drainable ()

virtual	~Drainable ()

virtual void	drainResume ()
	Resume execution after a successful drain. More...

void	signalDrainDone () const
	Signal that an object is drained. More...

Protected Attributes inherited from SimObject
const SimObjectParams &	_params
	Cached copy of the object parameters. More...

Protected Attributes inherited from EventManager
EventQueue *	eventq
	A pointer to this object's event queue. More...

Detailed Description

Definition at line 59 of file wavefront.hh.

Member Typedef Documentation

◆ Params

typedef WavefrontParams Wavefront::Params

Definition at line 242 of file wavefront.hh.

Member Enumeration Documentation

◆ status_e

enum Wavefront::status_e

Enumerator
S_STOPPED
S_RETURNING
S_RUNNING
S_STALLED
S_STALLED_SLEEP
S_WAITCNT	wavefront has unsatisfied wait counts while in this state the WF will only execute if the oldest instruction is the waitcnt. while in S_WAITCNT, the wavefront will not be ready until all of its waitcnts have been satisfied. the scoreboard ready() function will check the status of the waitcnts whenever the WF is in S_WAITCNT, and once they are satisfied, it will resume normal operation.
S_BARRIER	WF is stalled at a barrier.

Definition at line 62 of file wavefront.hh.

Constructor & Destructor Documentation

◆ Wavefront()

Wavefront::Wavefront ( const Params & p )

◆ ~Wavefront()

Wavefront::~Wavefront ( )

Definition at line 512 of file wavefront.cc.

Member Function Documentation

◆ barrierId() [1/2]

int Wavefront::barrierId ( ) const

Definition at line 1424 of file wavefront.cc.

References barId.

◆ barrierId() [2/2]

void Wavefront::barrierId ( int bar_id )

Definition at line 1416 of file wavefront.cc.

References barId, and WFBarrier::InvalidID.

Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), and ComputeUnit::releaseWFsFromBarrier().

◆ clearWaitCnts()

void Wavefront::clearWaitCnts ( )

Definition at line 1315 of file wavefront.cc.

References expWaitCnt, lgkmWaitCnt, S_RUNNING, status, and vmWaitCnt.

Referenced by waitCntsSatisfied().

◆ computeActualWgSz()

void Wavefront::computeActualWgSz ( HSAQueueEntry * task )

Definition at line 1405 of file wavefront.cc.

References actualWgSz, actualWgSzTotal, ArmISA::d, gridSz, HSAQueueEntry::MAX_DIM, HSAQueueEntry::wgId(), and workGroupSz.

◆ decExpInstsIssued()

void Wavefront::decExpInstsIssued ( )

Definition at line 1353 of file wavefront.cc.

References expInstsIssued.

◆ decLGKMInstsIssued()

void Wavefront::decLGKMInstsIssued ( )

Definition at line 1359 of file wavefront.cc.

References lgkmInstsIssued.

◆ decVMemInstsIssued()

void Wavefront::decVMemInstsIssued ( )

Definition at line 1347 of file wavefront.cc.

References vmemInstsIssued.

◆ discardFetch()

void Wavefront::discardFetch ( )

clear the fetch buffer for this wave in order to remove any stale inst data

Definition at line 1188 of file wavefront.cc.

References computeUnit, dropFetch, ComputeUnit::fetchStage, FetchStage::fetchUnit(), FetchUnit::flushBuf(), instructionBuffer, pendingFetch, simdId, and wfSlotId.

Referenced by exec().

◆ exec()

void Wavefront::exec ( )

if this wave is in S_WAITCNT state, then it should enter exec() precisely one time before the waitcnts are satisfied, in order to execute the waitcnt instruction itself thus we assert that the waitcnt is the oldest instruction. if we enter exec() with active waitcnts, and we're not executing the waitcnt instruction, something must be wrong

we return here to avoid spurious errors related to flat insts and their address segment resolution.

Definition at line 861 of file wavefront.cc.

Referenced by ExecStage::exec().

◆ execMask() [1/2]

VectorMask & Wavefront::execMask ( )

Definition at line 1377 of file wavefront.cc.

References _execMask.

◆ execMask() [2/2]

bool Wavefront::execMask ( int lane ) const

Definition at line 1383 of file wavefront.cc.

References _execMask.

◆ freeRegisterFile()

void Wavefront::freeRegisterFile ( )

Freeing VRF space.

Definition at line 1389 of file wavefront.cc.

References computeUnit, ArmISA::i, RegisterManager::mapVgpr(), maxVgprs, ComputeUnit::registerManager, reservedVectorRegs, simdId, startVgprIndex, ComputeUnit::vrf, and RegisterManager::vrfPoolMgrs.

◆ freeResources()

void Wavefront::freeResources ( )

Definition at line 740 of file wavefront.cc.

References execUnitId.

Referenced by ExecStage::exec().

◆ getStatus()

status_e Wavefront::getStatus ( )

inline

Definition at line 135 of file wavefront.hh.

References status.

Referenced by FetchUnit::exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and ComputeUnit::releaseWFsFromBarrier().

◆ gpuISA()

TheGpuISA::GPUISA& Wavefront::gpuISA ( )

inline

Definition at line 287 of file wavefront.hh.

References _gpuISA.

◆ hasBarrier()

bool Wavefront::hasBarrier ( ) const

Definition at line 1430 of file wavefront.cc.

References barId, and WFBarrier::InvalidID.

Referenced by ScheduleStage::addToSchList(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and Gcn3ISA::Inst_SOPP__S_BARRIER::execute().

◆ incExpInstsIssued()

void Wavefront::incExpInstsIssued ( )

Definition at line 1335 of file wavefront.cc.

References expInstsIssued.

Referenced by ScheduleStage::exec().

◆ incLGKMInstsIssued()

void Wavefront::incLGKMInstsIssued ( )

Definition at line 1341 of file wavefront.cc.

References lgkmInstsIssued.

Referenced by ScheduleStage::exec().

◆ incVMemInstsIssued()

void Wavefront::incVMemInstsIssued ( )

Definition at line 1329 of file wavefront.cc.

References vmemInstsIssued.

Referenced by ScheduleStage::exec().

◆ init()

void Wavefront::init ( )

virtual

init() is called after all C++ SimObjects have been created and all ports are connected.

Initializations that are independent of unserialization but rely on a fully instantiated and connected SimObject graph should be done here.

Reimplemented from SimObject.

Definition at line 101 of file wavefront.cc.

References computeUnit, globalMem, localMem, ComputeUnit::mapWaveToGlobalMem(), ComputeUnit::mapWaveToLocalMem(), ComputeUnit::mapWaveToScalarAlu(), ComputeUnit::mapWaveToScalarAluGlobalIdx(), ComputeUnit::mapWaveToScalarMem(), reservedScalarRegs, reservedVectorRegs, scalarAlu, scalarAluGlobalIdx, scalarMem, startSgprIndex, and startVgprIndex.

◆ initRegState()

void Wavefront::initRegState	(	HSAQueueEntry *	task,
		int	wgSizeInWorkItems
	)

Since flat scratch init is needed for this kernel, this kernel is going to have flat memory instructions and we need to initialize the hidden private base for this queue. scratch_resource_descriptor[0] has this queue's scratch base address. scratch_backing_memory_location has the offset to this queue's scratch base address from the SH_HIDDEN_PRIVATE_BASE_VMID. Ideally, we only require this queue's scratch base address for address calculation (stored in scratch_resource_descriptor[0]). But that address calculation shoule be done by first finding the queue's scratch base address using the calculation "SH_HIDDEN_PRIVATE_BASE_VMID + offset". So, we initialize SH_HIDDEN_PRIVATE_BASE_VMID.

For more details see: http://rocm-documentation.readthedocs.io/en/latest/ ROCm_Compiler_SDK/ROCm-Native-ISA.html#flat-scratch

https://github.com/ROCm-Developer-Tools/ ROCm-ComputeABI-Doc/blob/master/AMDGPU-ABI.md #flat-addressing

the compute_tmpring_size_wavesize specifies the number of kB allocated per wavefront, hence the multiplication by 1024.

to get the per wavefront offset into the scratch memory, we also multiply this by the wfId. the wfId stored in the Wavefront class, however, is the wave ID within the WG, whereas here we need the global WFID because the scratch space will be divided amongst all waves in the kernel. to get the global ID we multiply the WGID by the WG size, then add the WFID of the wave within its WG.

Definition at line 116 of file wavefront.cc.

References HSAQueueEntry::amdQueue, bits(), _amd_queue_s::compute_tmpring_size_wavesize, computeUnit, ComputeUnit::cu_id, DispatchPtr, divCeil(), DPRINTF, fatal, FlatScratchInit, HSAQueueEntry::gridSize(), GridWorkgroupCountX, GridWorkgroupCountY, GridWorkgroupCountZ, HSAQueueEntry::hostAMDQueueAddr, HSAQueueEntry::hostDispPktAddr(), Shader::initShHiddenPrivateBase(), HSAQueueEntry::kernargAddr(), KernargSegPtr, RegisterManager::mapSgpr(), RegisterManager::mapVgpr(), NumScalarInitFields, NumVectorInitFields, PrivateSegBuf, PrivSegWaveByteOffset, QueuePtr, rawDist, ComputeUnit::registerManager, _amd_queue_s::scratch_backing_memory_location, _amd_queue_s::scratch_resource_descriptor, _amd_queue_s::scratch_workitem_byte_size, HSAQueueEntry::sgprBitEnabled(), ComputeUnit::shader, simdId, ComputeUnit::srf, HSAQueueEntry::vgprBitEnabled(), ComputeUnit::vrf, wfDynId, wfId, ComputeUnit::wfSize(), wfSlotId, wgId, HSAQueueEntry::wgSize(), wgSz, workGroupId, WorkgroupIdX, WorkgroupIdY, WorkgroupIdZ, WorkgroupInfo, workItemId, WorkitemIdX, WorkitemIdY, and WorkitemIdZ.

◆ isGmInstruction()

bool Wavefront::isGmInstruction ( GPUDynInstPtr ii )

Definition at line 565 of file wavefront.cc.

Referenced by exec().

◆ isLmInstruction()

bool Wavefront::isLmInstruction ( GPUDynInstPtr ii )

Definition at line 576 of file wavefront.cc.

Referenced by exec().

◆ isOldestInstBarrier()

bool Wavefront::isOldestInstBarrier ( )

Definition at line 648 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

Referenced by ScheduleStage::addToSchList().

◆ isOldestInstFlatMem()

bool Wavefront::isOldestInstFlatMem ( )

Definition at line 713 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstGMem()

bool Wavefront::isOldestInstGMem ( )

Definition at line 661 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstLMem()

bool Wavefront::isOldestInstLMem ( )

Definition at line 687 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstPrivMem()

bool Wavefront::isOldestInstPrivMem ( )

Definition at line 700 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstScalarALU()

bool Wavefront::isOldestInstScalarALU ( )

Definition at line 618 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstScalarMem()

bool Wavefront::isOldestInstScalarMem ( )

Definition at line 674 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstSleep()

bool Wavefront::isOldestInstSleep ( )

Definition at line 587 of file wavefront.cc.

References instructionBuffer.

Referenced by ScheduleStage::addToSchList().

◆ isOldestInstVectorALU()

bool Wavefront::isOldestInstVectorALU ( )

Definition at line 633 of file wavefront.cc.

References instructionBuffer, S_STOPPED, and status.

◆ isOldestInstWaitcnt()

bool Wavefront::isOldestInstWaitcnt ( )

Definition at line 601 of file wavefront.cc.

References instructionBuffer.

Referenced by ScheduleStage::addToSchList(), and exec().

◆ nextInstr()

GPUDynInstPtr Wavefront::nextInstr ( )

Definition at line 1169 of file wavefront.cc.

References computeUnit, instructionBuffer, and ComputeUnit::pipeMap.

Referenced by ScoreboardCheckStage::exec().

◆ pc() [1/2]

Addr Wavefront::pc ( ) const

Definition at line 1365 of file wavefront.cc.

References _pc.

◆ pc() [2/2]

void Wavefront::pc ( Addr new_pc )

Definition at line 1371 of file wavefront.cc.

References _pc.

◆ releaseBarrier()

void Wavefront::releaseBarrier ( )

Definition at line 1436 of file wavefront.cc.

References barId, and WFBarrier::InvalidID.

Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute().

◆ reserveGmResource()

void Wavefront::reserveGmResource ( GPUDynInstPtr ii )

private

Definition at line 759 of file wavefront.cc.

References execUnitId, globalMem, panic, rdGmReqsInPipe, scalarMem, scalarRdGmReqsInPipe, scalarWrGmReqsInPipe, and wrGmReqsInPipe.

Referenced by reserveResources().

◆ reserveLmResource()

void Wavefront::reserveLmResource ( GPUDynInstPtr ii )

private

Definition at line 789 of file wavefront.cc.

References execUnitId, fatal_if, localMem, panic, rdLmReqsInPipe, and wrLmReqsInPipe.

Referenced by reserveResources().

◆ reserveResources()

std::vector< int > Wavefront::reserveResources ( )

Definition at line 807 of file wavefront.cc.

References execUnitId, fatal_if, flatGmUnitId, flatLmUnitId, instructionBuffer, panic, reserveGmResource(), reserveLmResource(), scalarAluGlobalIdx, and simdId.

Referenced by ScheduleStage::reserveResources().

◆ resizeRegFiles()

void Wavefront::resizeRegFiles	(	int	num_vregs,
		int	num_sregs
	)

Definition at line 506 of file wavefront.cc.

References maxSgprs, and maxVgprs.

◆ setParent()

void Wavefront::setParent ( ComputeUnit * cu )

inline

Definition at line 248 of file wavefront.hh.

References computeUnit.

◆ setSleepTime()

void Wavefront::setSleepTime ( int sleep_time )

Definition at line 1263 of file wavefront.cc.

References sleepCnt.

◆ setStatus()

void Wavefront::setStatus ( status_e newStatus )

Definition at line 517 of file wavefront.cc.

References computeUnit, ComputeUnit::cu_id, curTick(), ComputeUnit::idleCUTimeout, ComputeUnit::idleWfs, lastNonIdleTick, Shader::n_wf, ComputeUnit::numVectorALUs, panic_if, S_BARRIER, S_STALLED, S_STOPPED, S_WAITCNT, ComputeUnit::shader, and status.

Referenced by ScheduleStage::addToSchList(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and ComputeUnit::releaseWFsFromBarrier().

◆ setWaitCnts()

void Wavefront::setWaitCnts	(	int	vm_wait_cnt,
		int	exp_wait_cnt,
		int	lgkm_wait_cnt
	)

prior waitcnts should be satisfied, at which time the WF resets them back to -1, indicating they are no longer active

if the instruction encoding indicates a waitcnt of 0xf, that means the waitcnt is not being used

Definition at line 1270 of file wavefront.cc.

References expWaitCnt, lgkmWaitCnt, S_WAITCNT, status, and vmWaitCnt.

◆ sleepDone()

bool Wavefront::sleepDone ( )

Definition at line 1244 of file wavefront.cc.

References S_RUNNING, S_STALLED_SLEEP, sleepCnt, and status.

◆ start()

void Wavefront::start	(	uint64_t	_wfDynId,
		uint64_t	_base_ptr
	)

Definition at line 554 of file wavefront.cc.

References _pc, maxVgprs, S_RUNNING, status, vecReads, and wfDynId.

◆ stopFetch()

bool Wavefront::stopFetch ( )

Definition at line 726 of file wavefront.cc.

References instructionBuffer.

Referenced by FetchUnit::exec().

◆ validateRequestCounters()

void Wavefront::validateRequestCounters ( )

Definition at line 745 of file wavefront.cc.

References outstandingReqs, panic_if, rdGmReqsInPipe, rdLmReqsInPipe, simdId, wfDynId, wfSlotId, wrGmReqsInPipe, and wrLmReqsInPipe.

◆ waitCntsSatisfied()

bool Wavefront::waitCntsSatisfied ( )

If we reach here, that means an s_waitcnt instruction was executed and the waitcnts are set by the execute method. Check if waitcnts are satisfied.

Definition at line 1201 of file wavefront.cc.

References clearWaitCnts(), expInstsIssued, expWaitCnt, lgkmInstsIssued, lgkmWaitCnt, vmemInstsIssued, and vmWaitCnt.

Member Data Documentation

◆ _execMask

VectorMask Wavefront::_execMask

private

Definition at line 328 of file wavefront.hh.

Referenced by execMask(), and Wavefront().

◆ _gpuISA

TheGpuISA::GPUISA Wavefront::_gpuISA

private

Definition at line 298 of file wavefront.hh.

Referenced by exec(), and gpuISA().

◆ _pc

Addr Wavefront::_pc

private

Definition at line 327 of file wavefront.hh.

Referenced by pc(), and start().

◆ actualWgSz

uint32_t Wavefront::actualWgSz[3]

Definition at line 161 of file wavefront.hh.

Referenced by computeActualWgSz().

◆ actualWgSzTotal

uint32_t Wavefront::actualWgSzTotal

Definition at line 162 of file wavefront.hh.

Referenced by computeActualWgSz().

◆ barId

int Wavefront::barId

private

Definition at line 329 of file wavefront.hh.

Referenced by barrierId(), hasBarrier(), and releaseBarrier().

◆ computeUnit

ComputeUnit* Wavefront::computeUnit

Definition at line 104 of file wavefront.hh.

◆ context

uint8_t* Wavefront::context

Definition at line 240 of file wavefront.hh.

◆ dispatchId

uint32_t Wavefront::dispatchId

Definition at line 167 of file wavefront.hh.

Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute().

◆ dropFetch

bool Wavefront::dropFetch

Definition at line 110 of file wavefront.hh.

Referenced by discardFetch(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), FetchUnit::fetch(), FetchUnit::processFetchReturn(), ComputeUnit::ITLBPort::recvTimingResp(), and Wavefront().

◆ execUnitId

int Wavefront::execUnitId

Definition at line 100 of file wavefront.hh.

◆ expInstsIssued

int Wavefront::expInstsIssued

private

Definition at line 323 of file wavefront.hh.

Referenced by decExpInstsIssued(), incExpInstsIssued(), and waitCntsSatisfied().

◆ expWaitCnt

int Wavefront::expWaitCnt

private

Definition at line 320 of file wavefront.hh.

Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().

◆ flatGmUnitId

int Wavefront::flatGmUnitId

Definition at line 102 of file wavefront.hh.

Referenced by reserveResources(), and GPUDynInst::resolveFlatSegment().

◆ flatLmUnitId

int Wavefront::flatLmUnitId

Definition at line 101 of file wavefront.hh.

Referenced by reserveResources(), and GPUDynInst::resolveFlatSegment().

◆ globalMem

int Wavefront::globalMem

Definition at line 124 of file wavefront.hh.

Referenced by init(), reserveGmResource(), and ScheduleStage::reserveResources().

◆ gridSz

uint32_t Wavefront::gridSz[3]

Definition at line 157 of file wavefront.hh.

Referenced by computeActualWgSz().

◆ initMask

VectorMask Wavefront::initMask

Definition at line 217 of file wavefront.hh.

◆ instructionBuffer

◆ kernId

int Wavefront::kernId

Definition at line 95 of file wavefront.hh.

Referenced by GPUDynInst::GPUDynInst(), GPUDispatcher::isReachingKernelEnd(), and GPUDispatcher::notifyWgCompl().

◆ lastAddr

std::vector<Addr> Wavefront::lastAddr

Definition at line 151 of file wavefront.hh.

Referenced by Wavefront().

◆ lastInstExec

uint64_t Wavefront::lastInstExec

Definition at line 227 of file wavefront.hh.

Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and Wavefront().

◆ lastNonIdleTick

Tick Wavefront::lastNonIdleTick

Definition at line 112 of file wavefront.hh.

Referenced by setStatus(), and Wavefront().

◆ lastTrace

uint64_t Wavefront::lastTrace

Definition at line 190 of file wavefront.hh.

Referenced by Wavefront().

◆ ldsChunk

LdsChunk* Wavefront::ldsChunk

Definition at line 221 of file wavefront.hh.

Referenced by Gcn3ISA::Inst_DS::initDualMemRead(), Gcn3ISA::Inst_DS::initDualMemWrite(), Gcn3ISA::Inst_DS::initMemRead(), Gcn3ISA::Inst_DS::initMemWrite(), and Wavefront().

◆ lgkmInstsIssued

int Wavefront::lgkmInstsIssued

private

Definition at line 324 of file wavefront.hh.

Referenced by decLGKMInstsIssued(), incLGKMInstsIssued(), and waitCntsSatisfied().

◆ lgkmWaitCnt

int Wavefront::lgkmWaitCnt

private

Definition at line 321 of file wavefront.hh.

Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().

◆ localMem

int Wavefront::localMem

Definition at line 125 of file wavefront.hh.

Referenced by ScheduleStage::arbitrateVrfToLdsBus(), init(), reserveLmResource(), ScheduleStage::reserveResources(), and ScheduleStage::scheduleRfDestOperands().

◆ maxDynWaveId

uint32_t Wavefront::maxDynWaveId

Definition at line 166 of file wavefront.hh.

◆ maxIbSize

int Wavefront::maxIbSize

Definition at line 105 of file wavefront.hh.

Referenced by FetchUnit::FetchBufDesc::allocateBuf().

◆ maxSgprs

uint32_t Wavefront::maxSgprs

Definition at line 131 of file wavefront.hh.

Referenced by resizeRegFiles(), GPUDynInst::resolveFlatSegment(), and Wavefront().

◆ maxVgprs

uint32_t Wavefront::maxVgprs

Definition at line 129 of file wavefront.hh.

Referenced by freeRegisterFile(), resizeRegFiles(), start(), and Wavefront().

◆ memTraceBusy

int Wavefront::memTraceBusy

Definition at line 189 of file wavefront.hh.

Referenced by Wavefront().

◆ oldDgpr

std::vector<uint64_t> Wavefront::oldDgpr

Definition at line 210 of file wavefront.hh.

Referenced by Wavefront().

◆ oldDgprId

uint32_t Wavefront::oldDgprId

Definition at line 212 of file wavefront.hh.

◆ oldDgprTcnt

uint64_t Wavefront::oldDgprTcnt

Definition at line 214 of file wavefront.hh.

Referenced by Wavefront().

◆ oldVgpr

std::vector<uint32_t> Wavefront::oldVgpr

Definition at line 203 of file wavefront.hh.

Referenced by Wavefront().

◆ oldVgprId

uint32_t Wavefront::oldVgprId

Definition at line 205 of file wavefront.hh.

◆ oldVgprTcnt

uint64_t Wavefront::oldVgprTcnt

Definition at line 207 of file wavefront.hh.

Referenced by Wavefront().

◆ outstandingReqs

int Wavefront::outstandingReqs

Definition at line 169 of file wavefront.hh.

◆ outstandingReqsRdGm

int Wavefront::outstandingReqsRdGm

Definition at line 175 of file wavefront.hh.

◆ outstandingReqsRdLm

int Wavefront::outstandingReqsRdLm

Definition at line 177 of file wavefront.hh.

◆ outstandingReqsWrGm

int Wavefront::outstandingReqsWrGm

Definition at line 171 of file wavefront.hh.

◆ outstandingReqsWrLm

int Wavefront::outstandingReqsWrLm

Definition at line 173 of file wavefront.hh.

◆ pendingFetch

bool Wavefront::pendingFetch

Definition at line 109 of file wavefront.hh.

Referenced by discardFetch(), FetchUnit::exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), FetchUnit::fetch(), FetchUnit::processFetchReturn(), ComputeUnit::ITLBPort::recvTimingResp(), and Wavefront().

◆ rawDist

std::unordered_map<int,uint64_t> Wavefront::rawDist

Definition at line 231 of file wavefront.hh.

Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), initRegState(), and Wavefront().

◆ rdGmReqsInPipe

int Wavefront::rdGmReqsInPipe

Definition at line 183 of file wavefront.hh.

◆ rdLmReqsInPipe

int Wavefront::rdLmReqsInPipe

Definition at line 182 of file wavefront.hh.

◆ reservedScalarRegs

int Wavefront::reservedScalarRegs

Definition at line 194 of file wavefront.hh.

Referenced by init(), Gcn3ISA::ScalarOperand< DataType, Const, sizeof(DataType)/sizeof(VecElemU32) >::regIdx(), and Wavefront().

◆ reservedVectorRegs

int Wavefront::reservedVectorRegs

Definition at line 192 of file wavefront.hh.

Referenced by freeRegisterFile(), init(), and Wavefront().

◆ scalarAlu

int Wavefront::scalarAlu

Definition at line 119 of file wavefront.hh.

Referenced by ScheduleStage::dispatchReady(), exec(), and init().

◆ scalarAluGlobalIdx

int Wavefront::scalarAluGlobalIdx

Definition at line 123 of file wavefront.hh.

Referenced by init(), and reserveResources().

◆ scalarMem

int Wavefront::scalarMem

Definition at line 126 of file wavefront.hh.

Referenced by init(), and reserveGmResource().

◆ scalarOutstandingReqsRdGm

int Wavefront::scalarOutstandingReqsRdGm

Definition at line 179 of file wavefront.hh.

◆ scalarOutstandingReqsWrGm

int Wavefront::scalarOutstandingReqsWrGm

Definition at line 181 of file wavefront.hh.

Referenced by Gcn3ISA::Inst_SMEM__S_STORE_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX4::execute(), and Wavefront().

◆ scalarRdGmReqsInPipe

int Wavefront::scalarRdGmReqsInPipe

Definition at line 186 of file wavefront.hh.

◆ scalarWrGmReqsInPipe

int Wavefront::scalarWrGmReqsInPipe

Definition at line 187 of file wavefront.hh.

Referenced by ScheduleStage::dispatchReady(), Gcn3ISA::Inst_SMEM__S_STORE_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX4::execute(), reserveGmResource(), and Wavefront().

◆ simdId

const int Wavefront::simdId

Definition at line 97 of file wavefront.hh.

◆ sleepCnt

int Wavefront::sleepCnt

private

Definition at line 325 of file wavefront.hh.

Referenced by setSleepTime(), and sleepDone().

◆ startSgprIndex

uint32_t Wavefront::startSgprIndex

Definition at line 200 of file wavefront.hh.

Referenced by init(), and Wavefront().

◆ startVgprIndex

uint32_t Wavefront::startVgprIndex

Definition at line 197 of file wavefront.hh.

Referenced by freeRegisterFile(), init(), and Wavefront().

◆ stats

Wavefront::WavefrontStats Wavefront::stats

Referenced by ScheduleStage::addToSchList(), ScheduleStage::checkRfOperandReadComplete(), ScheduleStage::exec(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and ScheduleStage::schedRfWrites().

◆ status

status_e Wavefront::status

private

Definition at line 326 of file wavefront.hh.

Referenced by clearWaitCnts(), exec(), getStatus(), isOldestInstBarrier(), isOldestInstFlatMem(), isOldestInstGMem(), isOldestInstLMem(), isOldestInstPrivMem(), isOldestInstScalarALU(), isOldestInstScalarMem(), isOldestInstVectorALU(), setStatus(), setWaitCnts(), sleepDone(), start(), and Wavefront().

◆ vecReads

std::vector<int> Wavefront::vecReads

Definition at line 235 of file wavefront.hh.

Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), start(), and Wavefront().

◆ vmemInstsIssued

int Wavefront::vmemInstsIssued

private

Definition at line 322 of file wavefront.hh.

Referenced by decVMemInstsIssued(), incVMemInstsIssued(), and waitCntsSatisfied().

◆ vmWaitCnt

int Wavefront::vmWaitCnt

private

the following are used for waitcnt instructions vmWaitCnt: once set, we wait for the oustanding number of vector mem instructions to be at, or below vmWaitCnt.

expWaitCnt: once set, we wait for the outstanding number outstanding VM writes or EXP insts to be at, or below expWaitCnt.

lgkmWaitCnt: once set, we wait for the oustanding number of LDS, GDS, scalar memory, and message instructions to be at, or below lgkmCount. we currently do not support GDS/message ops.

Definition at line 319 of file wavefront.hh.

Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().

◆ wfDynId

uint64_t Wavefront::wfDynId

Definition at line 224 of file wavefront.hh.

Referenced by OFSchedulingPolicy::__chooseWave(), ScheduleStage::addToSchList(), ScheduleStage::checkRfOperandReadComplete(), ExecStage::dumpDispList(), ExecStage::exec(), ScoreboardCheckStage::exec(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), GPUDynInst::GPUDynInst(), FetchUnit::initiateFetch(), initRegState(), VectorRegisterFile::printReg(), ScalarRegisterFile::printReg(), ScheduleStage::reserveResources(), start(), and validateRequestCounters().

◆ wfId

uint32_t Wavefront::wfId

Definition at line 165 of file wavefront.hh.

Referenced by initRegState().

◆ wfSlotId

const int Wavefront::wfSlotId

Definition at line 94 of file wavefront.hh.

Referenced by discardFetch(), FetchUnit::exec(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), FetchUnit::fetch(), GPUDynInst::GPUDynInst(), FetchUnit::init(), FetchUnit::initiateFetch(), initRegState(), VectorRegisterFile::printReg(), ScalarRegisterFile::printReg(), FetchUnit::processFetchReturn(), and validateRequestCounters().

◆ wgId

uint32_t Wavefront::wgId

Definition at line 158 of file wavefront.hh.

Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), GPUDynInst::GPUDynInst(), initRegState(), and GPUDispatcher::notifyWgCompl().

◆ wgSz

uint32_t Wavefront::wgSz

Definition at line 159 of file wavefront.hh.

Referenced by initRegState().

◆ workGroupId

uint32_t Wavefront::workGroupId[3]

Definition at line 155 of file wavefront.hh.

Referenced by initRegState().

◆ workGroupSz

uint32_t Wavefront::workGroupSz[3]

Definition at line 156 of file wavefront.hh.

Referenced by computeActualWgSz().

◆ workItemFlatId

std::vector<uint32_t> Wavefront::workItemFlatId

Definition at line 153 of file wavefront.hh.

Referenced by Wavefront().

◆ workItemId

std::vector<uint32_t> Wavefront::workItemId[3]

Definition at line 152 of file wavefront.hh.

Referenced by initRegState(), and Wavefront().

◆ wrGmReqsInPipe

int Wavefront::wrGmReqsInPipe

Definition at line 185 of file wavefront.hh.

◆ wrLmReqsInPipe

int Wavefront::wrLmReqsInPipe

Definition at line 184 of file wavefront.hh.

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

gpu-compute/wavefront.hh
gpu-compute/wavefront.cc

Classes

Public Types

Public Member Functions

Public Attributes

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Member Typedef Documentation

◆ Params

Member Enumeration Documentation

◆ status_e

Constructor & Destructor Documentation

◆ Wavefront()

◆ ~Wavefront()

Member Function Documentation

◆ barrierId() [1/2]

◆ barrierId() [2/2]

◆ clearWaitCnts()

◆ computeActualWgSz()

◆ decExpInstsIssued()

◆ decLGKMInstsIssued()

◆ decVMemInstsIssued()

◆ discardFetch()

◆ exec()

◆ execMask() [1/2]

◆ execMask() [2/2]

◆ freeRegisterFile()

◆ freeResources()

◆ getStatus()

◆ gpuISA()

◆ hasBarrier()

◆ incExpInstsIssued()

◆ incLGKMInstsIssued()

◆ incVMemInstsIssued()

◆ init()

◆ initRegState()

◆ isGmInstruction()

◆ isLmInstruction()

◆ isOldestInstBarrier()

◆ isOldestInstFlatMem()

◆ isOldestInstGMem()

◆ isOldestInstLMem()

◆ isOldestInstPrivMem()

◆ isOldestInstScalarALU()

◆ isOldestInstScalarMem()

◆ isOldestInstSleep()

◆ isOldestInstVectorALU()

◆ isOldestInstWaitcnt()

◆ nextInstr()

◆ pc() [1/2]

◆ pc() [2/2]

◆ releaseBarrier()

◆ reserveGmResource()

◆ reserveLmResource()

◆ reserveResources()

◆ resizeRegFiles()

◆ setParent()

◆ setSleepTime()

◆ setStatus()

◆ setWaitCnts()

◆ sleepDone()

◆ start()

◆ stopFetch()

◆ validateRequestCounters()

◆ waitCntsSatisfied()

Member Data Documentation

◆ _execMask

◆ _gpuISA

◆ _pc

◆ actualWgSz

◆ actualWgSzTotal

◆ barId

◆ computeUnit

◆ context

◆ dispatchId

◆ dropFetch

◆ execUnitId

◆ expInstsIssued

◆ expWaitCnt