gem5
v20.1.0.0
|
#include <wavefront.hh>
Public Types | |
enum | status_e { S_STOPPED, S_RETURNING, S_RUNNING, S_STALLED, 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 | 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 () |
void | regStats () |
Callback to set stat parameters. More... | |
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... | |
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 *_params) | |
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 | 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... | |
Group ()=delete | |
Group (const Group &)=delete | |
Group & | operator= (const Group &)=delete |
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 |
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... | |
Definition at line 57 of file wavefront.hh.
typedef WavefrontParams Wavefront::Params |
Definition at line 275 of file wavefront.hh.
enum Wavefront::status_e |
Definition at line 60 of file wavefront.hh.
Wavefront::Wavefront | ( | const Params * | p | ) |
Definition at line 52 of file wavefront.cc.
References _execMask, dropFetch, execUnitId, ArmISA::i, lastAddr, lastInstExec, lastNonIdleTick, lastTrace, ldsChunk, maxSgprs, maxVgprs, memTraceBusy, oldDgpr, oldDgprTcnt, oldVgpr, oldVgprTcnt, outstandingReqs, outstandingReqsRdGm, outstandingReqsRdLm, outstandingReqsWrGm, outstandingReqsWrLm, MipsISA::p, pendingFetch, rawDist, rdGmReqsInPipe, rdLmReqsInPipe, reservedScalarRegs, reservedVectorRegs, S_STOPPED, scalarOutstandingReqsRdGm, scalarOutstandingReqsWrGm, scalarRdGmReqsInPipe, scalarWrGmReqsInPipe, startSgprIndex, startVgprIndex, status, vecReads, workItemFlatId, workItemId, wrGmReqsInPipe, and wrLmReqsInPipe.
Wavefront::~Wavefront | ( | ) |
Definition at line 586 of file wavefront.cc.
int Wavefront::barrierId | ( | ) | const |
Definition at line 1445 of file wavefront.cc.
References barId.
void Wavefront::barrierId | ( | int | bar_id | ) |
Definition at line 1437 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().
void Wavefront::clearWaitCnts | ( | ) |
Definition at line 1336 of file wavefront.cc.
References expWaitCnt, lgkmWaitCnt, S_RUNNING, status, and vmWaitCnt.
Referenced by waitCntsSatisfied().
void Wavefront::computeActualWgSz | ( | HSAQueueEntry * | task | ) |
Definition at line 1426 of file wavefront.cc.
References actualWgSz, actualWgSzTotal, ArmISA::d, gridSz, HSAQueueEntry::MAX_DIM, HSAQueueEntry::wgId(), and workGroupSz.
void Wavefront::decExpInstsIssued | ( | ) |
Definition at line 1374 of file wavefront.cc.
References expInstsIssued.
void Wavefront::decLGKMInstsIssued | ( | ) |
Definition at line 1380 of file wavefront.cc.
References lgkmInstsIssued.
Referenced by Gcn3ISA::Inst_DS__DS_PERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_BPERMUTE_B32::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), and GPUDynInst::resolveFlatSegment().
void Wavefront::decVMemInstsIssued | ( | ) |
Definition at line 1368 of file wavefront.cc.
References vmemInstsIssued.
Referenced by Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), and GPUDynInst::resolveFlatSegment().
void Wavefront::discardFetch | ( | ) |
clear the fetch buffer for this wave in order to remove any stale inst data
Definition at line 1235 of file wavefront.cc.
References computeUnit, dropFetch, ComputeUnit::fetchStage, FetchStage::fetchUnit(), FetchUnit::flushBuf(), instructionBuffer, pendingFetch, simdId, and wfSlotId.
Referenced by 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 921 of file wavefront.cc.
References _gpuISA, ComputeUnit::activeLanesPerGMemInstrDist, ComputeUnit::activeLanesPerLMemInstrDist, computeUnit, ComputeUnit::controlFlowDivergenceDist, ComputeUnit::cu_id, Clocked::cyclesToTicks(), ComputeUnit::deleteFromPipeMap(), discardFetch(), DPRINTF, execMask(), ComputeUnit::execRateDist, fatal, ComputeUnit::getTokenManager(), Shader::hsail_mode, ArmISA::i, ComputeUnit::instCyclesLdsPerSimd, ComputeUnit::instCyclesScMemPerSimd, ComputeUnit::instCyclesVMemPerSimd, ComputeUnit::instExecPerSimd, ComputeUnit::instInterleave, instructionBuffer, isGmInstruction(), isLmInstruction(), isOldestInstWaitcnt(), ComputeUnit::issuePeriod, ComputeUnit::lastExecCycle, lastInstExec, ArmISA::n, Stats::none, numInstrExecuted, ComputeUnit::numInstrExecuted, ComputeUnit::numVecOpsExecuted, ComputeUnit::numVecOpsExecutedF16, ComputeUnit::numVecOpsExecutedF32, ComputeUnit::numVecOpsExecutedF64, ComputeUnit::numVecOpsExecutedFMA16, ComputeUnit::numVecOpsExecutedFMA32, ComputeUnit::numVecOpsExecutedFMA64, ComputeUnit::numVecOpsExecutedMAC16, ComputeUnit::numVecOpsExecutedMAC32, ComputeUnit::numVecOpsExecutedMAC64, ComputeUnit::numVecOpsExecutedMAD16, ComputeUnit::numVecOpsExecutedMAD32, ComputeUnit::numVecOpsExecutedMAD64, ComputeUnit::numVecOpsExecutedTwoOpFP, panic, pc(), rawDist, readsPerWrite, TokenManager::recvTokens(), S_RETURNING, S_STALLED, S_STOPPED, S_WAITCNT, Stats::DistBase< Derived, Stor >::sample(), scalarAlu, ComputeUnit::scalarALUs, ComputeUnit::scalarMemUnit, WaitClass::set(), ComputeUnit::shader, simdId, Shader::SIMT, ComputeUnit::srf, ComputeUnit::srf_scm_bus_latency, ComputeUnit::srfToScalarMemPipeBus, status, ComputeUnit::totalCycles, ComputeUnit::updateInstStats(), Stats::ScalarBase< Derived, Stor >::value(), vecRawDistance, vecReads, ComputeUnit::vectorALUs, ComputeUnit::vectorGlobalMemUnit, Shader::vectorInstDstOperand, Shader::vectorInstSrcOperand, ComputeUnit::vectorSharedMemUnit, ComputeUnit::vrf, ComputeUnit::vrf_gm_bus_latency, ComputeUnit::vrf_lm_bus_latency, ComputeUnit::vrfToGlobalMemPipeBus, ComputeUnit::vrfToLocalMemPipeBus, wfDynId, and wfSlotId.
Referenced by ExecStage::exec().
VectorMask & Wavefront::execMask | ( | ) |
Definition at line 1398 of file wavefront.cc.
References _execMask.
Referenced by Gcn3ISA::Inst_DS::calcAddr(), exec(), Gcn3ISA::Inst_SOP1__S_AND_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_OR_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_XOR_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_ANDN2_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_ORN2_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_NAND_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_NOR_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOP1__S_XNOR_SAVEEXEC_B64::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_EXECZ::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_EXECNZ::execute(), Gcn3ISA::Inst_VOP2__V_CNDMASK_B32::execute(), Gcn3ISA::Inst_VOP2__V_ADD_F32::execute(), Gcn3ISA::Inst_VOP2__V_SUB_F32::execute(), Gcn3ISA::Inst_VOP2__V_SUBREV_F32::execute(), Gcn3ISA::Inst_VOP2__V_MUL_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP2__V_MUL_F32::execute(), Gcn3ISA::Inst_VOP2__V_MUL_I32_I24::execute(), Gcn3ISA::Inst_VOP2__V_MUL_HI_I32_I24::execute(), Gcn3ISA::Inst_VOP2__V_MUL_U32_U24::execute(), Gcn3ISA::Inst_VOP2__V_MUL_HI_U32_U24::execute(), Gcn3ISA::Inst_VOP2__V_MIN_F32::execute(), Gcn3ISA::Inst_VOP2__V_MAX_F32::execute(), Gcn3ISA::Inst_VOP2__V_MIN_I32::execute(), Gcn3ISA::Inst_VOP2__V_MAX_I32::execute(), Gcn3ISA::Inst_VOP2__V_MIN_U32::execute(), Gcn3ISA::Inst_VOP2__V_MAX_U32::execute(), Gcn3ISA::Inst_VOP2__V_LSHRREV_B32::execute(), Gcn3ISA::Inst_VOP2__V_ASHRREV_I32::execute(), Gcn3ISA::Inst_VOP2__V_LSHLREV_B32::execute(), Gcn3ISA::Inst_VOP2__V_AND_B32::execute(), Gcn3ISA::Inst_VOP2__V_OR_B32::execute(), Gcn3ISA::Inst_VOP2__V_XOR_B32::execute(), Gcn3ISA::Inst_VOP2__V_MAC_F32::execute(), Gcn3ISA::Inst_VOP2__V_MADMK_F32::execute(), Gcn3ISA::Inst_VOP2__V_MADAK_F32::execute(), Gcn3ISA::Inst_VOP2__V_ADD_U32::execute(), Gcn3ISA::Inst_VOP2__V_SUB_U32::execute(), Gcn3ISA::Inst_VOP2__V_SUBREV_U32::execute(), Gcn3ISA::Inst_VOP2__V_ADDC_U32::execute(), Gcn3ISA::Inst_VOP2__V_SUBB_U32::execute(), Gcn3ISA::Inst_VOP2__V_SUBBREV_U32::execute(), Gcn3ISA::Inst_VOP2__V_ADD_U16::execute(), Gcn3ISA::Inst_VOP2__V_SUB_U16::execute(), Gcn3ISA::Inst_VOP2__V_SUBREV_U16::execute(), Gcn3ISA::Inst_VOP2__V_MUL_LO_U16::execute(), Gcn3ISA::Inst_VOP2__V_LSHLREV_B16::execute(), Gcn3ISA::Inst_VOP2__V_LSHRREV_B16::execute(), Gcn3ISA::Inst_VOP2__V_ASHRREV_I16::execute(), Gcn3ISA::Inst_VOP2__V_MAX_U16::execute(), Gcn3ISA::Inst_VOP2__V_MAX_I16::execute(), Gcn3ISA::Inst_VOP2__V_MIN_U16::execute(), Gcn3ISA::Inst_VOP2__V_MIN_I16::execute(), Gcn3ISA::Inst_VOP1__V_MOV_B32::execute(), Gcn3ISA::Inst_VOP1__V_READFIRSTLANE_B32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_I32_F64::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F64_I32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_I32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_U32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_U32_F32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_I32_F32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_RPI_I32_F32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_FLR_I32_F32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_F64::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F64_F32::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_UBYTE0::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_UBYTE1::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_UBYTE2::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F32_UBYTE3::execute(), Gcn3ISA::Inst_VOP1__V_CVT_U32_F64::execute(), Gcn3ISA::Inst_VOP1__V_CVT_F64_U32::execute(), Gcn3ISA::Inst_VOP1__V_TRUNC_F64::execute(), Gcn3ISA::Inst_VOP1__V_CEIL_F64::execute(), Gcn3ISA::Inst_VOP1__V_RNDNE_F64::execute(), Gcn3ISA::Inst_VOP1__V_FLOOR_F64::execute(), Gcn3ISA::Inst_VOP1__V_FRACT_F32::execute(), Gcn3ISA::Inst_VOP1__V_TRUNC_F32::execute(), Gcn3ISA::Inst_VOP1__V_CEIL_F32::execute(), Gcn3ISA::Inst_VOP1__V_RNDNE_F32::execute(), Gcn3ISA::Inst_VOP1__V_FLOOR_F32::execute(), Gcn3ISA::Inst_VOP1__V_EXP_F32::execute(), Gcn3ISA::Inst_VOP1__V_LOG_F32::execute(), Gcn3ISA::Inst_VOP1__V_RCP_F32::execute(), Gcn3ISA::Inst_VOP1__V_RCP_IFLAG_F32::execute(), Gcn3ISA::Inst_VOP1__V_RSQ_F32::execute(), Gcn3ISA::Inst_VOP1__V_RCP_F64::execute(), Gcn3ISA::Inst_VOP1__V_RSQ_F64::execute(), Gcn3ISA::Inst_VOP1__V_SQRT_F32::execute(), Gcn3ISA::Inst_VOP1__V_SQRT_F64::execute(), Gcn3ISA::Inst_VOP1__V_SIN_F32::execute(), Gcn3ISA::Inst_VOP1__V_COS_F32::execute(), Gcn3ISA::Inst_VOP1__V_NOT_B32::execute(), Gcn3ISA::Inst_VOP1__V_BFREV_B32::execute(), Gcn3ISA::Inst_VOP1__V_FFBH_U32::execute(), Gcn3ISA::Inst_VOP1__V_FFBL_B32::execute(), Gcn3ISA::Inst_VOP1__V_FFBH_I32::execute(), Gcn3ISA::Inst_VOP1__V_FREXP_EXP_I32_F64::execute(), Gcn3ISA::Inst_VOP1__V_FREXP_MANT_F64::execute(), Gcn3ISA::Inst_VOP1__V_FRACT_F64::execute(), Gcn3ISA::Inst_VOP1__V_FREXP_EXP_I32_F32::execute(), Gcn3ISA::Inst_VOP1__V_FREXP_MANT_F32::execute(), Gcn3ISA::Inst_VOP1__V_EXP_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP1__V_LOG_LEGACY_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_CLASS_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_CLASS_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_CLASS_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_CLASS_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LG_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_O_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_U_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NGE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLG_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NGT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NEQ_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_TRU_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LG_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_O_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_U_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NGE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLG_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NGT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLE_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NEQ_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLT_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_TRU_F32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LG_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_O_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_U_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NGE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLG_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NGT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NEQ_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NLT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_TRU_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LG_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_O_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_U_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NGE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLG_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NGT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLE_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NEQ_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NLT_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_TRU_F64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_I16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_U16::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_I32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_U32::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_F_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LT_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_EQ_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_LE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GT_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_NE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_GE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMP_T_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_I64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_F_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LT_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_EQ_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_LE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GT_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_NE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_GE_U64::execute(), Gcn3ISA::Inst_VOPC__V_CMPX_T_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_CLASS_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_CLASS_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_CLASS_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_CLASS_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_TRU_F16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_F16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_TRU_F16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LG_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_O_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_U_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NGE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLG_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NGT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NEQ_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_TRU_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LG_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_O_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_U_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NGE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLG_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NGT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLE_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NEQ_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLT_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_TRU_F32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LG_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_O_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_U_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NGE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLG_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NGT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NEQ_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NLT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_TRU_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LG_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_O_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_U_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NGE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLG_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NGT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLE_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NEQ_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NLT_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_TRU_F64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_I16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_U16::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_I32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_U32::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_F_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LT_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_EQ_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_LE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GT_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_NE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_GE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMP_T_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_I64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_F_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LT_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_EQ_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_LE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GT_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_NE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_GE_U64::execute(), Gcn3ISA::Inst_VOP3__V_CMPX_T_U64::execute(), Gcn3ISA::Inst_VOP3__V_CNDMASK_B32::execute(), Gcn3ISA::Inst_VOP3__V_ADD_F32::execute(), Gcn3ISA::Inst_VOP3__V_SUB_F32::execute(), Gcn3ISA::Inst_VOP3__V_SUBREV_F32::execute(), Gcn3ISA::Inst_VOP3__V_MUL_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP3__V_MUL_F32::execute(), Gcn3ISA::Inst_VOP3__V_MUL_I32_I24::execute(), Gcn3ISA::Inst_VOP3__V_MUL_HI_I32_I24::execute(), Gcn3ISA::Inst_VOP3__V_MUL_U32_U24::execute(), Gcn3ISA::Inst_VOP3__V_MUL_HI_U32_U24::execute(), Gcn3ISA::Inst_VOP3__V_MIN_F32::execute(), Gcn3ISA::Inst_VOP3__V_MAX_F32::execute(), Gcn3ISA::Inst_VOP3__V_MIN_I32::execute(), Gcn3ISA::Inst_VOP3__V_MAX_I32::execute(), Gcn3ISA::Inst_VOP3__V_MIN_U32::execute(), Gcn3ISA::Inst_VOP3__V_MAX_U32::execute(), Gcn3ISA::Inst_VOP3__V_LSHRREV_B32::execute(), Gcn3ISA::Inst_VOP3__V_ASHRREV_I32::execute(), Gcn3ISA::Inst_VOP3__V_LSHLREV_B32::execute(), Gcn3ISA::Inst_VOP3__V_AND_B32::execute(), Gcn3ISA::Inst_VOP3__V_OR_B32::execute(), Gcn3ISA::Inst_VOP3__V_XOR_B32::execute(), Gcn3ISA::Inst_VOP3__V_MAC_F32::execute(), Gcn3ISA::Inst_VOP3__V_ADD_U32::execute(), Gcn3ISA::Inst_VOP3__V_SUB_U32::execute(), Gcn3ISA::Inst_VOP3__V_SUBREV_U32::execute(), Gcn3ISA::Inst_VOP3__V_ADDC_U32::execute(), Gcn3ISA::Inst_VOP3__V_SUBB_U32::execute(), Gcn3ISA::Inst_VOP3__V_SUBBREV_U32::execute(), Gcn3ISA::Inst_VOP3__V_ADD_U16::execute(), Gcn3ISA::Inst_VOP3__V_SUB_U16::execute(), Gcn3ISA::Inst_VOP3__V_SUBREV_U16::execute(), Gcn3ISA::Inst_VOP3__V_MUL_LO_U16::execute(), Gcn3ISA::Inst_VOP3__V_LSHLREV_B16::execute(), Gcn3ISA::Inst_VOP3__V_LSHRREV_B16::execute(), Gcn3ISA::Inst_VOP3__V_ASHRREV_I16::execute(), Gcn3ISA::Inst_VOP3__V_MAX_U16::execute(), Gcn3ISA::Inst_VOP3__V_MAX_I16::execute(), Gcn3ISA::Inst_VOP3__V_MIN_U16::execute(), Gcn3ISA::Inst_VOP3__V_MIN_I16::execute(), Gcn3ISA::Inst_VOP3__V_MOV_B32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_I32_F64::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F64_I32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_I32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_U32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_U32_F32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_I32_F32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_RPI_I32_F32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_FLR_I32_F32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_F64::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F64_F32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_UBYTE0::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_UBYTE1::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_UBYTE2::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F32_UBYTE3::execute(), Gcn3ISA::Inst_VOP3__V_CVT_U32_F64::execute(), Gcn3ISA::Inst_VOP3__V_CVT_F64_U32::execute(), Gcn3ISA::Inst_VOP3__V_TRUNC_F64::execute(), Gcn3ISA::Inst_VOP3__V_CEIL_F64::execute(), Gcn3ISA::Inst_VOP3__V_RNDNE_F64::execute(), Gcn3ISA::Inst_VOP3__V_FLOOR_F64::execute(), Gcn3ISA::Inst_VOP3__V_FRACT_F32::execute(), Gcn3ISA::Inst_VOP3__V_TRUNC_F32::execute(), Gcn3ISA::Inst_VOP3__V_CEIL_F32::execute(), Gcn3ISA::Inst_VOP3__V_RNDNE_F32::execute(), Gcn3ISA::Inst_VOP3__V_FLOOR_F32::execute(), Gcn3ISA::Inst_VOP3__V_EXP_F32::execute(), Gcn3ISA::Inst_VOP3__V_LOG_F32::execute(), Gcn3ISA::Inst_VOP3__V_RCP_F32::execute(), Gcn3ISA::Inst_VOP3__V_RCP_IFLAG_F32::execute(), Gcn3ISA::Inst_VOP3__V_RSQ_F32::execute(), Gcn3ISA::Inst_VOP3__V_RCP_F64::execute(), Gcn3ISA::Inst_VOP3__V_RSQ_F64::execute(), Gcn3ISA::Inst_VOP3__V_SQRT_F32::execute(), Gcn3ISA::Inst_VOP3__V_SQRT_F64::execute(), Gcn3ISA::Inst_VOP3__V_SIN_F32::execute(), Gcn3ISA::Inst_VOP3__V_COS_F32::execute(), Gcn3ISA::Inst_VOP3__V_NOT_B32::execute(), Gcn3ISA::Inst_VOP3__V_BFREV_B32::execute(), Gcn3ISA::Inst_VOP3__V_FFBH_U32::execute(), Gcn3ISA::Inst_VOP3__V_FFBL_B32::execute(), Gcn3ISA::Inst_VOP3__V_FFBH_I32::execute(), Gcn3ISA::Inst_VOP3__V_FREXP_EXP_I32_F64::execute(), Gcn3ISA::Inst_VOP3__V_FREXP_MANT_F64::execute(), Gcn3ISA::Inst_VOP3__V_FRACT_F64::execute(), Gcn3ISA::Inst_VOP3__V_FREXP_EXP_I32_F32::execute(), Gcn3ISA::Inst_VOP3__V_FREXP_MANT_F32::execute(), Gcn3ISA::Inst_VOP3__V_EXP_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP3__V_LOG_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP3__V_MAD_LEGACY_F32::execute(), Gcn3ISA::Inst_VOP3__V_MAD_F32::execute(), Gcn3ISA::Inst_VOP3__V_MAD_I32_I24::execute(), Gcn3ISA::Inst_VOP3__V_MAD_U32_U24::execute(), Gcn3ISA::Inst_VOP3__V_BFE_U32::execute(), Gcn3ISA::Inst_VOP3__V_BFE_I32::execute(), Gcn3ISA::Inst_VOP3__V_BFI_B32::execute(), Gcn3ISA::Inst_VOP3__V_FMA_F32::execute(), Gcn3ISA::Inst_VOP3__V_FMA_F64::execute(), Gcn3ISA::Inst_VOP3__V_LERP_U8::execute(), Gcn3ISA::Inst_VOP3__V_ALIGNBIT_B32::execute(), Gcn3ISA::Inst_VOP3__V_ALIGNBYTE_B32::execute(), Gcn3ISA::Inst_VOP3__V_MIN3_F32::execute(), Gcn3ISA::Inst_VOP3__V_MIN3_I32::execute(), Gcn3ISA::Inst_VOP3__V_MIN3_U32::execute(), Gcn3ISA::Inst_VOP3__V_MAX3_F32::execute(), Gcn3ISA::Inst_VOP3__V_MAX3_I32::execute(), Gcn3ISA::Inst_VOP3__V_MAX3_U32::execute(), Gcn3ISA::Inst_VOP3__V_MED3_F32::execute(), Gcn3ISA::Inst_VOP3__V_MED3_I32::execute(), Gcn3ISA::Inst_VOP3__V_MED3_U32::execute(), Gcn3ISA::Inst_VOP3__V_SAD_U8::execute(), Gcn3ISA::Inst_VOP3__V_SAD_HI_U8::execute(), Gcn3ISA::Inst_VOP3__V_SAD_U16::execute(), Gcn3ISA::Inst_VOP3__V_SAD_U32::execute(), Gcn3ISA::Inst_VOP3__V_CVT_PK_U8_F32::execute(), Gcn3ISA::Inst_VOP3__V_DIV_FIXUP_F32::execute(), Gcn3ISA::Inst_VOP3__V_DIV_FIXUP_F64::execute(), Gcn3ISA::Inst_VOP3__V_DIV_SCALE_F32::execute(), Gcn3ISA::Inst_VOP3__V_DIV_SCALE_F64::execute(), Gcn3ISA::Inst_VOP3__V_DIV_FMAS_F32::execute(), Gcn3ISA::Inst_VOP3__V_DIV_FMAS_F64::execute(), Gcn3ISA::Inst_VOP3__V_MAD_U64_U32::execute(), Gcn3ISA::Inst_VOP3__V_MAD_I64_I32::execute(), Gcn3ISA::Inst_VOP3__V_MAD_U16::execute(), Gcn3ISA::Inst_VOP3__V_MAD_I16::execute(), Gcn3ISA::Inst_VOP3__V_PERM_B32::execute(), Gcn3ISA::Inst_VOP3__V_ADD_F64::execute(), Gcn3ISA::Inst_VOP3__V_MUL_F64::execute(), Gcn3ISA::Inst_VOP3__V_MIN_F64::execute(), Gcn3ISA::Inst_VOP3__V_MAX_F64::execute(), Gcn3ISA::Inst_VOP3__V_LDEXP_F64::execute(), Gcn3ISA::Inst_VOP3__V_MUL_LO_U32::execute(), Gcn3ISA::Inst_VOP3__V_MUL_HI_U32::execute(), Gcn3ISA::Inst_VOP3__V_MUL_HI_I32::execute(), Gcn3ISA::Inst_VOP3__V_LDEXP_F32::execute(), Gcn3ISA::Inst_VOP3__V_BCNT_U32_B32::execute(), Gcn3ISA::Inst_VOP3__V_MBCNT_LO_U32_B32::execute(), Gcn3ISA::Inst_VOP3__V_MBCNT_HI_U32_B32::execute(), Gcn3ISA::Inst_VOP3__V_LSHLREV_B64::execute(), Gcn3ISA::Inst_VOP3__V_LSHRREV_B64::execute(), Gcn3ISA::Inst_VOP3__V_ASHRREV_I64::execute(), Gcn3ISA::Inst_VOP3__V_BFM_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_PERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_BPERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), VectorRegisterFile::printReg(), Gcn3ISA::VecOperand< DataType, Const, NumDwords >::write(), and Gcn3ISA::ScalarOperand< DataType, Const, sizeof(DataType)/sizeof(VecElemU32) >::write().
bool Wavefront::execMask | ( | int | lane | ) | const |
Definition at line 1404 of file wavefront.cc.
References _execMask.
void Wavefront::freeRegisterFile | ( | ) |
Freeing VRF space.
Definition at line 1410 of file wavefront.cc.
References computeUnit, ArmISA::i, RegisterManager::mapVgpr(), maxVgprs, ComputeUnit::registerManager, reservedVectorRegs, simdId, startVgprIndex, ComputeUnit::vrf, and RegisterManager::vrfPoolMgrs.
void Wavefront::freeResources | ( | ) |
Definition at line 800 of file wavefront.cc.
References execUnitId.
Referenced by ExecStage::exec().
|
inline |
Definition at line 130 of file wavefront.hh.
References status.
Referenced by FetchUnit::exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), and ComputeUnit::releaseWFsFromBarrier().
|
inline |
Definition at line 318 of file wavefront.hh.
References _gpuISA.
bool Wavefront::hasBarrier | ( | ) | const |
Definition at line 1451 of file wavefront.cc.
References barId, and WFBarrier::InvalidID.
Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and Gcn3ISA::Inst_SOPP__S_BARRIER::execute().
void Wavefront::incExpInstsIssued | ( | ) |
Definition at line 1356 of file wavefront.cc.
References expInstsIssued.
void Wavefront::incLGKMInstsIssued | ( | ) |
Definition at line 1362 of file wavefront.cc.
References lgkmInstsIssued.
Referenced by ScheduleStage::exec().
void Wavefront::incVMemInstsIssued | ( | ) |
Definition at line 1350 of file wavefront.cc.
References vmemInstsIssued.
Referenced by ScheduleStage::exec().
|
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 175 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.
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 190 of file wavefront.cc.
References HSAQueueEntry::amdQueue, _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.
bool Wavefront::isGmInstruction | ( | GPUDynInstPtr | ii | ) |
Definition at line 639 of file wavefront.cc.
Referenced by exec().
bool Wavefront::isLmInstruction | ( | GPUDynInstPtr | ii | ) |
Definition at line 650 of file wavefront.cc.
Referenced by exec().
bool Wavefront::isOldestInstBarrier | ( | ) |
Definition at line 708 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstFlatMem | ( | ) |
Definition at line 773 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstGMem | ( | ) |
Definition at line 721 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstLMem | ( | ) |
Definition at line 747 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstPrivMem | ( | ) |
Definition at line 760 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstScalarALU | ( | ) |
Definition at line 678 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstScalarMem | ( | ) |
Definition at line 734 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstVectorALU | ( | ) |
Definition at line 693 of file wavefront.cc.
References instructionBuffer, S_STOPPED, and status.
bool Wavefront::isOldestInstWaitcnt | ( | ) |
Definition at line 661 of file wavefront.cc.
References instructionBuffer.
Referenced by ScheduleStage::addToSchList(), and exec().
GPUDynInstPtr Wavefront::nextInstr | ( | ) |
Definition at line 1216 of file wavefront.cc.
References computeUnit, instructionBuffer, and ComputeUnit::pipeMap.
Referenced by ScoreboardCheckStage::exec().
Addr Wavefront::pc | ( | ) | const |
Definition at line 1386 of file wavefront.cc.
References _pc.
Referenced by Gcn3ISA::GPUISA::advancePC(), exec(), Gcn3ISA::Inst_SOP1__S_GETPC_B64::execute(), Gcn3ISA::Inst_SOP1__S_SETPC_B64::execute(), Gcn3ISA::Inst_SOP1__S_SWAPPC_B64::execute(), Gcn3ISA::Inst_SOPP__S_BRANCH::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_SCC0::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_SCC1::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_VCCZ::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_VCCNZ::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_EXECZ::execute(), Gcn3ISA::Inst_SOPP__S_CBRANCH_EXECNZ::execute(), and FetchUnit::initiateFetch().
void Wavefront::pc | ( | Addr | new_pc | ) |
Definition at line 1392 of file wavefront.cc.
References _pc.
|
virtual |
Callback to set stat parameters.
This callback is typically used for complex stats (e.g., distributions) that need parameters in addition to a name and a description. Stat names and descriptions should typically be set from the constructor usingo from the constructor using the ADD_STAT macro.
Reimplemented from Stats::Group.
Definition at line 106 of file wavefront.cc.
References Stats::DataWrap< Derived, InfoProxyType >::desc(), Stats::Distribution::init(), SimObject::name(), Stats::DataWrap< Derived, InfoProxyType >::name(), numInstrExecuted, numTimesBlockedDueRAWDependencies, numTimesBlockedDueWAXDependencies, readsPerWrite, Stats::Group::regStats(), schCycles, schLdsArbStalls, schOpdNrdyStalls, schResourceStalls, schRfAccessStalls, schStalls, and vecRawDistance.
void Wavefront::releaseBarrier | ( | ) |
Definition at line 1457 of file wavefront.cc.
References barId, and WFBarrier::InvalidID.
Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute().
|
private |
Definition at line 819 of file wavefront.cc.
References execUnitId, globalMem, panic, rdGmReqsInPipe, scalarMem, scalarRdGmReqsInPipe, scalarWrGmReqsInPipe, and wrGmReqsInPipe.
Referenced by reserveResources().
|
private |
Definition at line 849 of file wavefront.cc.
References execUnitId, fatal_if, localMem, panic, rdLmReqsInPipe, and wrLmReqsInPipe.
Referenced by reserveResources().
std::vector< int > Wavefront::reserveResources | ( | ) |
Definition at line 867 of file wavefront.cc.
References execUnitId, fatal_if, flatGmUnitId, flatLmUnitId, instructionBuffer, panic, reserveGmResource(), reserveLmResource(), scalarAluGlobalIdx, and simdId.
Referenced by ScheduleStage::reserveResources().
void Wavefront::resizeRegFiles | ( | int | num_vregs, |
int | num_sregs | ||
) |
Definition at line 580 of file wavefront.cc.
|
inline |
Definition at line 281 of file wavefront.hh.
References computeUnit.
void Wavefront::setStatus | ( | status_e | newStatus | ) |
Definition at line 591 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(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), and ComputeUnit::releaseWFsFromBarrier().
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 1291 of file wavefront.cc.
References expWaitCnt, lgkmWaitCnt, S_WAITCNT, status, and vmWaitCnt.
void Wavefront::start | ( | uint64_t | _wfDynId, |
uint64_t | _base_ptr | ||
) |
bool Wavefront::stopFetch | ( | ) |
Definition at line 786 of file wavefront.cc.
References instructionBuffer.
Referenced by FetchUnit::exec().
void Wavefront::validateRequestCounters | ( | ) |
Definition at line 805 of file wavefront.cc.
References outstandingReqs, panic_if, rdGmReqsInPipe, rdLmReqsInPipe, simdId, wfDynId, wfSlotId, wrGmReqsInPipe, and wrLmReqsInPipe.
Referenced by Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_SWIZZLE_B32::execute(), Gcn3ISA::Inst_DS__DS_PERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_BPERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), and Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute().
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 1248 of file wavefront.cc.
References clearWaitCnts(), expInstsIssued, expWaitCnt, lgkmInstsIssued, lgkmWaitCnt, vmemInstsIssued, and vmWaitCnt.
|
private |
Definition at line 358 of file wavefront.hh.
Referenced by execMask(), and Wavefront().
|
private |
Definition at line 329 of file wavefront.hh.
|
private |
Definition at line 357 of file wavefront.hh.
uint32_t Wavefront::actualWgSz[3] |
Definition at line 155 of file wavefront.hh.
Referenced by computeActualWgSz().
uint32_t Wavefront::actualWgSzTotal |
Definition at line 156 of file wavefront.hh.
Referenced by computeActualWgSz().
|
private |
Definition at line 359 of file wavefront.hh.
Referenced by barrierId(), hasBarrier(), and releaseBarrier().
ComputeUnit* Wavefront::computeUnit |
Definition at line 99 of file wavefront.hh.
Referenced by discardFetch(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), freeRegisterFile(), init(), initRegState(), nextInstr(), GPUDispatcher::notifyWgCompl(), GPUDynInst::resolveFlatSegment(), setParent(), and setStatus().
uint8_t* Wavefront::context |
Definition at line 273 of file wavefront.hh.
uint32_t Wavefront::dispatchId |
Definition at line 161 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute().
bool Wavefront::dropFetch |
Definition at line 105 of file wavefront.hh.
Referenced by discardFetch(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), FetchUnit::processFetchReturn(), ComputeUnit::ITLBPort::recvTimingResp(), and Wavefront().
int Wavefront::execUnitId |
Definition at line 95 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_SMEM__S_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX16::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX16::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_STORE_DWORDX4::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_SWIZZLE_B32::execute(), Gcn3ISA::Inst_DS__DS_PERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_BPERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), freeResources(), reserveGmResource(), reserveLmResource(), reserveResources(), GPUDynInst::resolveFlatSegment(), and Wavefront().
|
private |
Definition at line 354 of file wavefront.hh.
Referenced by decExpInstsIssued(), incExpInstsIssued(), and waitCntsSatisfied().
|
private |
Definition at line 351 of file wavefront.hh.
Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().
int Wavefront::flatGmUnitId |
Definition at line 97 of file wavefront.hh.
Referenced by reserveResources(), and GPUDynInst::resolveFlatSegment().
int Wavefront::flatLmUnitId |
Definition at line 96 of file wavefront.hh.
Referenced by reserveResources(), and GPUDynInst::resolveFlatSegment().
int Wavefront::globalMem |
Definition at line 119 of file wavefront.hh.
Referenced by init(), reserveGmResource(), and ScheduleStage::reserveResources().
uint32_t Wavefront::gridSz[3] |
Definition at line 151 of file wavefront.hh.
Referenced by computeActualWgSz().
VectorMask Wavefront::initMask |
Definition at line 211 of file wavefront.hh.
std::deque<GPUDynInstPtr> Wavefront::instructionBuffer |
Definition at line 102 of file wavefront.hh.
Referenced by discardFetch(), ExecStage::dumpDispList(), ScheduleStage::exec(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), isOldestInstBarrier(), isOldestInstFlatMem(), isOldestInstGMem(), isOldestInstLMem(), isOldestInstPrivMem(), isOldestInstScalarALU(), isOldestInstScalarMem(), isOldestInstVectorALU(), isOldestInstWaitcnt(), nextInstr(), FetchUnit::processFetchReturn(), ComputeUnit::ITLBPort::recvTimingResp(), reserveResources(), ScheduleStage::scheduleRfDestOperands(), and stopFetch().
int Wavefront::kernId |
Definition at line 90 of file wavefront.hh.
Referenced by GPUDynInst::GPUDynInst(), GPUDispatcher::isReachingKernelEnd(), and GPUDispatcher::notifyWgCompl().
std::vector<Addr> Wavefront::lastAddr |
Definition at line 145 of file wavefront.hh.
Referenced by Wavefront().
uint64_t Wavefront::lastInstExec |
Definition at line 254 of file wavefront.hh.
Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and Wavefront().
Tick Wavefront::lastNonIdleTick |
Definition at line 107 of file wavefront.hh.
Referenced by setStatus(), and Wavefront().
uint64_t Wavefront::lastTrace |
Definition at line 184 of file wavefront.hh.
Referenced by Wavefront().
LdsChunk* Wavefront::ldsChunk |
Definition at line 215 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_DS::initDualMemRead(), Gcn3ISA::Inst_DS::initDualMemWrite(), Gcn3ISA::Inst_DS::initMemRead(), Gcn3ISA::Inst_DS::initMemWrite(), and Wavefront().
|
private |
Definition at line 355 of file wavefront.hh.
Referenced by decLGKMInstsIssued(), incLGKMInstsIssued(), and waitCntsSatisfied().
|
private |
Definition at line 352 of file wavefront.hh.
Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().
int Wavefront::localMem |
Definition at line 120 of file wavefront.hh.
Referenced by ScheduleStage::arbitrateVrfToLdsBus(), init(), reserveLmResource(), ScheduleStage::reserveResources(), and ScheduleStage::scheduleRfDestOperands().
uint32_t Wavefront::maxDynWaveId |
Definition at line 160 of file wavefront.hh.
int Wavefront::maxIbSize |
Definition at line 100 of file wavefront.hh.
Referenced by FetchUnit::FetchBufDesc::allocateBuf().
uint32_t Wavefront::maxSgprs |
Definition at line 126 of file wavefront.hh.
Referenced by resizeRegFiles(), GPUDynInst::resolveFlatSegment(), and Wavefront().
uint32_t Wavefront::maxVgprs |
Definition at line 124 of file wavefront.hh.
Referenced by freeRegisterFile(), resizeRegFiles(), start(), and Wavefront().
int Wavefront::memTraceBusy |
Definition at line 183 of file wavefront.hh.
Referenced by Wavefront().
Stats::Scalar Wavefront::numInstrExecuted |
Definition at line 224 of file wavefront.hh.
Referenced by exec(), and regStats().
Stats::Scalar Wavefront::numTimesBlockedDueRAWDependencies |
Definition at line 251 of file wavefront.hh.
Referenced by regStats().
Stats::Scalar Wavefront::numTimesBlockedDueWAXDependencies |
Definition at line 248 of file wavefront.hh.
Referenced by regStats().
std::vector<uint64_t> Wavefront::oldDgpr |
Definition at line 204 of file wavefront.hh.
Referenced by Wavefront().
uint32_t Wavefront::oldDgprId |
Definition at line 206 of file wavefront.hh.
uint64_t Wavefront::oldDgprTcnt |
Definition at line 208 of file wavefront.hh.
Referenced by Wavefront().
std::vector<uint32_t> Wavefront::oldVgpr |
Definition at line 197 of file wavefront.hh.
Referenced by Wavefront().
uint32_t Wavefront::oldVgprId |
Definition at line 199 of file wavefront.hh.
uint64_t Wavefront::oldVgprTcnt |
Definition at line 201 of file wavefront.hh.
Referenced by Wavefront().
int Wavefront::outstandingReqs |
Definition at line 163 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), validateRequestCounters(), and Wavefront().
int Wavefront::outstandingReqsRdGm |
Definition at line 169 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), and Wavefront().
int Wavefront::outstandingReqsRdLm |
Definition at line 171 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), and Wavefront().
int Wavefront::outstandingReqsWrGm |
Definition at line 165 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), and Wavefront().
int Wavefront::outstandingReqsWrLm |
Definition at line 167 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), and Wavefront().
bool Wavefront::pendingFetch |
Definition at line 104 of file wavefront.hh.
Referenced by discardFetch(), FetchUnit::exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), FetchUnit::processFetchReturn(), ComputeUnit::ITLBPort::recvTimingResp(), and Wavefront().
std::unordered_map<int,uint64_t> Wavefront::rawDist |
Definition at line 261 of file wavefront.hh.
Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), initRegState(), and Wavefront().
int Wavefront::rdGmReqsInPipe |
Definition at line 177 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), reserveGmResource(), GPUDynInst::resolveFlatSegment(), validateRequestCounters(), and Wavefront().
int Wavefront::rdLmReqsInPipe |
Definition at line 176 of file wavefront.hh.
Referenced by ScheduleStage::dispatchReady(), Gcn3ISA::Inst_DS__DS_READ_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2_B32::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_U8::execute(), Gcn3ISA::Inst_DS__DS_READ_U16::execute(), Gcn3ISA::Inst_DS__DS_SWIZZLE_B32::execute(), Gcn3ISA::Inst_DS__DS_PERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_BPERMUTE_B32::execute(), Gcn3ISA::Inst_DS__DS_READ_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2_B64::execute(), Gcn3ISA::Inst_DS__DS_READ2ST64_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_UBYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_USHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), reserveLmResource(), GPUDynInst::resolveFlatSegment(), validateRequestCounters(), and Wavefront().
Stats::Distribution Wavefront::readsPerWrite |
Definition at line 265 of file wavefront.hh.
Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), and regStats().
int Wavefront::reservedScalarRegs |
Definition at line 188 of file wavefront.hh.
Referenced by init(), Gcn3ISA::ScalarOperand< DataType, Const, sizeof(DataType)/sizeof(VecElemU32) >::regIdx(), and Wavefront().
int Wavefront::reservedVectorRegs |
Definition at line 186 of file wavefront.hh.
Referenced by freeRegisterFile(), init(), and Wavefront().
int Wavefront::scalarAlu |
Definition at line 114 of file wavefront.hh.
Referenced by ScheduleStage::dispatchReady(), exec(), and init().
int Wavefront::scalarAluGlobalIdx |
Definition at line 118 of file wavefront.hh.
Referenced by init(), and reserveResources().
int Wavefront::scalarMem |
Definition at line 121 of file wavefront.hh.
Referenced by init(), and reserveGmResource().
int Wavefront::scalarOutstandingReqsRdGm |
Definition at line 173 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_SMEM__S_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX16::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX16::execute(), and Wavefront().
int Wavefront::scalarOutstandingReqsWrGm |
Definition at line 175 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().
int Wavefront::scalarRdGmReqsInPipe |
Definition at line 180 of file wavefront.hh.
Referenced by ScheduleStage::dispatchReady(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_LOAD_DWORDX16::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORD::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX8::execute(), Gcn3ISA::Inst_SMEM__S_BUFFER_LOAD_DWORDX16::execute(), reserveGmResource(), and Wavefront().
int Wavefront::scalarWrGmReqsInPipe |
Definition at line 181 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().
Stats::Scalar Wavefront::schCycles |
Definition at line 227 of file wavefront.hh.
Referenced by ScheduleStage::checkRfOperandReadComplete(), ScheduleStage::exec(), and regStats().
Stats::Scalar Wavefront::schLdsArbStalls |
Definition at line 244 of file wavefront.hh.
Referenced by regStats().
Stats::Scalar Wavefront::schOpdNrdyStalls |
Definition at line 240 of file wavefront.hh.
Referenced by ScheduleStage::checkRfOperandReadComplete(), and regStats().
Stats::Scalar Wavefront::schResourceStalls |
Definition at line 238 of file wavefront.hh.
Referenced by regStats().
Stats::Scalar Wavefront::schRfAccessStalls |
Definition at line 236 of file wavefront.hh.
Referenced by ScheduleStage::addToSchList(), regStats(), and ScheduleStage::schedRfWrites().
Stats::Scalar Wavefront::schStalls |
Definition at line 230 of file wavefront.hh.
Referenced by ScheduleStage::addToSchList(), ScheduleStage::checkRfOperandReadComplete(), regStats(), and ScheduleStage::schedRfWrites().
const int Wavefront::simdId |
Definition at line 92 of file wavefront.hh.
Referenced by ScheduleStage::addToSchList(), ScheduleStage::checkRfOperandReadComplete(), discardFetch(), ScheduleStage::dispatchReady(), ExecStage::dumpDispList(), ScoreboardCheckStage::exec(), ExecStage::exec(), exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), Gcn3ISA::Inst_SOPP__S_BARRIER::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_UBYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_USHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_LOAD_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), FetchUnit::fetch(), FetchStage::fetch(), freeRegisterFile(), GPUDynInst::GPUDynInst(), FetchUnit::initiateFetch(), initRegState(), VectorRegisterFile::printReg(), ScalarRegisterFile::printReg(), FetchStage::processFetchReturn(), FetchUnit::processFetchReturn(), Gcn3ISA::VecOperand< DataType, Const, NumDwords >::read(), Gcn3ISA::ScalarOperand< DataType, Const, sizeof(DataType)/sizeof(VecElemU32) >::read(), ScheduleStage::reserveResources(), reserveResources(), ScheduleStage::schedRfWrites(), validateRequestCounters(), Gcn3ISA::VecOperand< DataType, Const, NumDwords >::write(), and Gcn3ISA::ScalarOperand< DataType, Const, sizeof(DataType)/sizeof(VecElemU32) >::write().
uint32_t Wavefront::startSgprIndex |
Definition at line 194 of file wavefront.hh.
Referenced by init(), and Wavefront().
uint32_t Wavefront::startVgprIndex |
Definition at line 191 of file wavefront.hh.
Referenced by freeRegisterFile(), init(), and Wavefront().
|
private |
Definition at line 356 of file wavefront.hh.
Referenced by clearWaitCnts(), exec(), getStatus(), isOldestInstBarrier(), isOldestInstFlatMem(), isOldestInstGMem(), isOldestInstLMem(), isOldestInstPrivMem(), isOldestInstScalarALU(), isOldestInstScalarMem(), isOldestInstVectorALU(), setStatus(), setWaitCnts(), start(), and Wavefront().
Stats::Distribution Wavefront::vecRawDistance |
Definition at line 258 of file wavefront.hh.
Referenced by exec(), and regStats().
std::vector<int> Wavefront::vecReads |
Definition at line 268 of file wavefront.hh.
Referenced by exec(), Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), start(), and Wavefront().
|
private |
Definition at line 353 of file wavefront.hh.
Referenced by decVMemInstsIssued(), incVMemInstsIssued(), and waitCntsSatisfied().
|
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 350 of file wavefront.hh.
Referenced by clearWaitCnts(), setWaitCnts(), and waitCntsSatisfied().
uint64_t Wavefront::wfDynId |
Definition at line 218 of file wavefront.hh.
Referenced by OFSchedulingPolicy::__chooseWave(), ScheduleStage::addToSchList(), ScheduleStage::checkRfOperandReadComplete(), ExecStage::dumpDispList(), ScoreboardCheckStage::exec(), ExecStage::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().
uint32_t Wavefront::wfId |
Definition at line 159 of file wavefront.hh.
Referenced by initRegState().
const int Wavefront::wfSlotId |
Definition at line 89 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().
uint32_t Wavefront::wgId |
Definition at line 152 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_SOPP__S_ENDPGM::execute(), GPUDynInst::GPUDynInst(), initRegState(), and GPUDispatcher::notifyWgCompl().
uint32_t Wavefront::wgSz |
Definition at line 153 of file wavefront.hh.
Referenced by initRegState().
uint32_t Wavefront::workGroupId[3] |
Definition at line 149 of file wavefront.hh.
Referenced by initRegState().
uint32_t Wavefront::workGroupSz[3] |
Definition at line 150 of file wavefront.hh.
Referenced by computeActualWgSz().
std::vector<uint32_t> Wavefront::workItemFlatId |
Definition at line 147 of file wavefront.hh.
Referenced by Wavefront().
std::vector<uint32_t> Wavefront::workItemId[3] |
Definition at line 146 of file wavefront.hh.
Referenced by initRegState(), and Wavefront().
int Wavefront::wrGmReqsInPipe |
Definition at line 179 of file wavefront.hh.
Referenced by Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_WBINVL1_VOL::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), reserveGmResource(), GPUDynInst::resolveFlatSegment(), validateRequestCounters(), and Wavefront().
int Wavefront::wrLmReqsInPipe |
Definition at line 178 of file wavefront.hh.
Referenced by ScheduleStage::dispatchReady(), Gcn3ISA::Inst_DS__DS_WRITE_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE2ST64_B32::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B8::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B16::execute(), Gcn3ISA::Inst_DS__DS_WRITE_B64::execute(), Gcn3ISA::Inst_DS__DS_WRITE2_B64::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_BYTE::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_SHORT::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORD::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX2::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX3::execute(), Gcn3ISA::Inst_MUBUF__BUFFER_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_BYTE::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_SHORT::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORD::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX2::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX3::execute(), Gcn3ISA::Inst_FLAT__FLAT_STORE_DWORDX4::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_CMPSWAP_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_ADD_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_SUB_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_INC_X2::execute(), Gcn3ISA::Inst_FLAT__FLAT_ATOMIC_DEC_X2::execute(), reserveLmResource(), GPUDynInst::resolveFlatSegment(), validateRequestCounters(), and Wavefront().