}
InOrderCPU::CPUEvent::CPUEvent(InOrderCPU *_cpu, CPUEventType e_type,
- Fault fault, ThreadID _tid, unsigned _vpe)
- : Event(CPU_Tick_Pri), cpu(_cpu)
+ Fault fault, ThreadID _tid, DynInstPtr inst,
+ unsigned event_pri_offset)
+ : Event(Event::Priority((unsigned int)CPU_Tick_Pri + event_pri_offset)),
+ cpu(_cpu)
{
- setEvent(e_type, fault, _tid, _vpe);
+ setEvent(e_type, fault, _tid, inst);
}
std::string InOrderCPU::eventNames[NumCPUEvents] =
{
"ActivateThread",
- "DeallocateThread",
+ "ActivateNextReadyThread",
+ "DeactivateThread",
+ "HaltThread",
"SuspendThread",
- "DisableThreads",
- "EnableThreads",
- "DisableVPEs",
- "EnableVPEs",
"Trap",
"InstGraduated",
- "SquashAll",
+ "SquashFromMemStall",
"UpdatePCs"
};
cpu->activateThread(tid);
break;
- //@TODO: Consider Implementing "Suspend Thread" as Separate from Deallocate
- case SuspendThread: // Suspend & Deallocate are same for now.
- //cpu->suspendThread(tid);
- //break;
- case DeallocateThread:
- cpu->deallocateThread(tid);
+ case ActivateNextReadyThread:
+ cpu->activateNextReadyThread();
break;
- case EnableVPEs:
- cpu->enableVPEs(vpe);
+ case DeactivateThread:
+ cpu->deactivateThread(tid);
break;
- case DisableVPEs:
- cpu->disableVPEs(tid, vpe);
+ case HaltThread:
+ cpu->haltThread(tid);
break;
- case EnableThreads:
- cpu->enableThreads(vpe);
+ case SuspendThread:
+ cpu->suspendThread(tid);
break;
- case DisableThreads:
- cpu->disableThreads(tid, vpe);
+ case SquashFromMemStall:
+ cpu->squashDueToMemStall(inst->squashingStage, inst->seqNum, tid);
break;
case Trap:
break;
default:
- fatal("Unrecognized Event Type %d", cpuEventType);
+ fatal("Unrecognized Event Type %s", eventNames[cpuEventType]);
}
-
+
cpu->cpuEventRemoveList.push(this);
}
+
+
const char *
InOrderCPU::CPUEvent::description()
{
InOrderCPU::CPUEvent::scheduleEvent(int delay)
{
if (squashed())
- mainEventQueue.reschedule(this,curTick + cpu->ticks(delay));
+ mainEventQueue.reschedule(this, cpu->nextCycle(curTick +
+ cpu->ticks(delay)));
else if (!scheduled())
- mainEventQueue.schedule(this,curTick + cpu->ticks(delay));
+ mainEventQueue.schedule(this, cpu->nextCycle(curTick +
+ cpu->ticks(delay)));
}
void
system(params->system),
physmem(system->physmem),
#endif // FULL_SYSTEM
+#ifdef DEBUG
+ cpuEventNum(0),
+ resReqCount(0),
+#endif // DEBUG
switchCount(0),
deferRegistration(false/*params->deferRegistration*/),
stageTracing(params->stageTracing),
- numVirtProcs(1)
-{
+ instsPerSwitch(0)
+{
ThreadID active_threads;
cpu_params = params;
"in your InOrder implementation or "
"edit your workload size.");
}
+
+
+ if (active_threads > 1) {
+ threadModel = (InOrderCPU::ThreadModel) params->threadModel;
+
+ if (threadModel == SMT) {
+ DPRINTF(InOrderCPU, "Setting Thread Model to SMT.\n");
+ } else if (threadModel == SwitchOnCacheMiss) {
+ DPRINTF(InOrderCPU, "Setting Thread Model to "
+ "Switch On Cache Miss\n");
+ }
+
+ } else {
+ threadModel = Single;
+ }
+
+
+
#endif
// Bind the fetch & data ports from the resource pool.
for (ThreadID tid = 0; tid < numThreads; ++tid) {
#if FULL_SYSTEM
// SMT is not supported in FS mode yet.
- assert(this->numThreads == 1);
- this->thread[tid] = new Thread(this, 0);
+ assert(numThreads == 1);
+ thread[tid] = new Thread(this, 0);
#else
if (tid < (ThreadID)params->workload.size()) {
DPRINTF(InOrderCPU, "Workload[%i] process is %#x\n",
- tid, this->thread[tid]);
- this->thread[tid] =
+ tid, params->workload[tid]->prog_fname);
+ thread[tid] =
new Thread(this, tid, params->workload[tid]);
} else {
//Allocate Empty thread so M5 can use later
//when scheduling threads to CPU
Process* dummy_proc = params->workload[0];
- this->thread[tid] = new Thread(this, tid, dummy_proc);
+ thread[tid] = new Thread(this, tid, dummy_proc);
}
+
+ // Eventually set this with parameters...
+ asid[tid] = tid;
#endif
// Setup the TC that will serve as the interface to the threads/CPU.
memset(floatRegs.i[tid], 0, sizeof(floatRegs.i[tid]));
isa[tid].clear();
- isa[tid].expandForMultithreading(numThreads, numVirtProcs);
+ isa[tid].expandForMultithreading(numThreads, 1/*numVirtProcs*/);
+
+ // Define dummy instructions and resource requests to be used.
+ dummyInst[tid] = new InOrderDynInst(this,
+ thread[tid],
+ 0,
+ tid,
+ asid[tid]);
+
+ dummyReq[tid] = new ResourceRequest(resPool->getResource(0),
+ dummyInst[tid],
+ 0,
+ 0,
+ 0,
+ 0);
}
- lastRunningCycle = curTick;
- contextSwitch = false;
+ dummyReqInst = new InOrderDynInst(this, NULL, 0, 0, 0);
+ dummyReqInst->setSquashed();
- // Define dummy instructions and resource requests to be used.
- DynInstPtr dummyBufferInst = new InOrderDynInst(this, NULL, 0, 0);
- dummyReq = new ResourceRequest(NULL, NULL, 0, 0, 0, 0);
+ dummyBufferInst = new InOrderDynInst(this, NULL, 0, 0, 0);
+ dummyBufferInst->setSquashed();
+
+ lastRunningCycle = curTick;
// Reset CPU to reset state.
#if FULL_SYSTEM
reset();
#endif
+ dummyBufferInst->resetInstCount();
+
// Schedule First Tick Event, CPU will reschedule itself from here on out.
scheduleTickEvent(0);
}
+InOrderCPU::~InOrderCPU()
+{
+ delete resPool;
+}
+
void
InOrderCPU::regStats()
/* Register the Resource Pool's stats here.*/
resPool->regStats();
+ /* Register for each Pipeline Stage */
+ for (int stage_num=0; stage_num < ThePipeline::NumStages; stage_num++) {
+ pipelineStage[stage_num]->regStats();
+ }
+
/* Register any of the InOrderCPU's stats here.*/
+ instsPerCtxtSwitch
+ .name(name() + ".instsPerContextSwitch")
+ .desc("Instructions Committed Per Context Switch")
+ .prereq(instsPerCtxtSwitch);
+
+ numCtxtSwitches
+ .name(name() + ".contextSwitches")
+ .desc("Number of context switches");
+
+ comLoads
+ .name(name() + ".comLoads")
+ .desc("Number of Load instructions committed");
+
+ comStores
+ .name(name() + ".comStores")
+ .desc("Number of Store instructions committed");
+
+ comBranches
+ .name(name() + ".comBranches")
+ .desc("Number of Branches instructions committed");
+
+ comNops
+ .name(name() + ".comNops")
+ .desc("Number of Nop instructions committed");
+
+ comNonSpec
+ .name(name() + ".comNonSpec")
+ .desc("Number of Non-Speculative instructions committed");
+
+ comInts
+ .name(name() + ".comInts")
+ .desc("Number of Integer instructions committed");
+
+ comFloats
+ .name(name() + ".comFloats")
+ .desc("Number of Floating Point instructions committed");
+
timesIdled
.name(name() + ".timesIdled")
.desc("Number of times that the entire CPU went into an idle state and"
idleCycles
.name(name() + ".idleCycles")
- .desc("Total number of cycles that the CPU has spent unscheduled due "
- "to idling")
- .prereq(idleCycles);
+ .desc("Number of cycles cpu's stages were not processed");
+
+ runCycles
+ .name(name() + ".runCycles")
+ .desc("Number of cycles cpu stages are processed.");
+
+ activity
+ .name(name() + ".activity")
+ .desc("Percentage of cycles cpu is active")
+ .precision(6);
+ activity = (runCycles / numCycles) * 100;
threadCycles
.init(numThreads)
smtCycles
.name(name() + ".smtCycles")
- .desc("Total number of cycles that the CPU was simultaneous multithreading.(SMT)");
+ .desc("Total number of cycles that the CPU was in SMT-mode");
committedInsts
.init(numThreads)
.name(name() + ".cpi")
.desc("CPI: Cycles Per Instruction (Per-Thread)")
.precision(6);
- cpi = threadCycles / committedInsts;
+ cpi = numCycles / committedInsts;
smtCpi
.name(name() + ".smt_cpi")
.name(name() + ".ipc")
.desc("IPC: Instructions Per Cycle (Per-Thread)")
.precision(6);
- ipc = committedInsts / threadCycles;
+ ipc = committedInsts / numCycles;
smtIpc
.name(name() + ".smt_ipc")
++numCycles;
+ bool pipes_idle = true;
+
//Tick each of the stages
for (int stNum=NumStages - 1; stNum >= 0 ; stNum--) {
pipelineStage[stNum]->tick();
+
+ pipes_idle = pipes_idle && pipelineStage[stNum]->idle;
}
+ if (pipes_idle)
+ idleCycles++;
+ else
+ runCycles++;
+
// Now advance the time buffers one tick
timeBuffer.advance();
for (int sqNum=0; sqNum < NumStages - 1; sqNum++) {
stageQueue[sqNum]->advance();
}
activityRec.advance();
-
+
// Any squashed requests, events, or insts then remove them now
cleanUpRemovedReqs();
cleanUpRemovedEvents();
//Tick next_tick = curTick + cycles(1);
//tickEvent.schedule(next_tick);
mainEventQueue.schedule(&tickEvent, nextCycle(curTick + 1));
- DPRINTF(InOrderCPU, "Scheduled CPU for next tick @ %i.\n", nextCycle(curTick + 1));
+ DPRINTF(InOrderCPU, "Scheduled CPU for next tick @ %i.\n",
+ nextCycle(curTick + 1));
}
}
{
for (int i = 0; i < numThreads; i++) {
isa[i].reset(coreType, numThreads,
- numVirtProcs, dynamic_cast<BaseCPU*>(this));
+ 1/*numVirtProcs*/, dynamic_cast<BaseCPU*>(this));
}
}
InOrderCPU::trap(Fault fault, ThreadID tid, int delay)
{
//@ Squash Pipeline during TRAP
- scheduleCpuEvent(Trap, fault, tid, 0/*vpe*/, delay);
+ scheduleCpuEvent(Trap, fault, tid, dummyInst[tid], delay);
}
void
fault->invoke(tcBase(tid));
}
+void
+InOrderCPU::squashFromMemStall(DynInstPtr inst, ThreadID tid, int delay)
+{
+ scheduleCpuEvent(SquashFromMemStall, NoFault, tid, inst, delay);
+}
+
+
+void
+InOrderCPU::squashDueToMemStall(int stage_num, InstSeqNum seq_num,
+ ThreadID tid)
+{
+ DPRINTF(InOrderCPU, "Squashing Pipeline Stages Due to Memory Stall...\n");
+
+ // Squash all instructions in each stage including
+ // instruction that caused the squash (seq_num - 1)
+ // NOTE: The stage bandwidth needs to be cleared so thats why
+ // the stalling instruction is squashed as well. The stalled
+ // instruction is previously placed in another intermediate buffer
+ // while it's stall is being handled.
+ InstSeqNum squash_seq_num = seq_num - 1;
+
+ for (int stNum=stage_num; stNum >= 0 ; stNum--) {
+ pipelineStage[stNum]->squashDueToMemStall(squash_seq_num, tid);
+ }
+}
+
void
InOrderCPU::scheduleCpuEvent(CPUEventType c_event, Fault fault,
- ThreadID tid, unsigned vpe, unsigned delay)
+ ThreadID tid, DynInstPtr inst,
+ unsigned delay, unsigned event_pri_offset)
{
- CPUEvent *cpu_event = new CPUEvent(this, c_event, fault, tid, vpe);
+ CPUEvent *cpu_event = new CPUEvent(this, c_event, fault, tid, inst,
+ event_pri_offset);
+ Tick sked_tick = nextCycle(curTick + ticks(delay));
if (delay >= 0) {
- DPRINTF(InOrderCPU, "Scheduling CPU Event (%s) for cycle %i.\n",
- eventNames[c_event], curTick + delay);
- mainEventQueue.schedule(cpu_event,curTick + delay);
+ DPRINTF(InOrderCPU, "Scheduling CPU Event (%s) for cycle %i, [tid:%i].\n",
+ eventNames[c_event], curTick + delay, tid);
+ mainEventQueue.schedule(cpu_event, sked_tick);
} else {
cpu_event->process();
cpuEventRemoveList.push(cpu_event);
}
// Broadcast event to the Resource Pool
- DynInstPtr dummy_inst =
- new InOrderDynInst(this, NULL, getNextEventNum(), tid);
- resPool->scheduleEvent(c_event, dummy_inst, 0, 0, tid);
+ // Need to reset tid just in case this is a dummy instruction
+ inst->setTid(tid);
+ resPool->scheduleEvent(c_event, inst, 0, 0, tid);
}
-inline bool
+bool
InOrderCPU::isThreadActive(ThreadID tid)
{
list<ThreadID>::iterator isActive =
return (isActive != activeThreads.end());
}
-
-void
-InOrderCPU::activateThread(ThreadID tid)
+bool
+InOrderCPU::isThreadReady(ThreadID tid)
{
- if (!isThreadActive(tid)) {
- DPRINTF(InOrderCPU,
- "Adding Thread %i to active threads list in CPU.\n", tid);
- activeThreads.push_back(tid);
+ list<ThreadID>::iterator isReady =
+ std::find(readyThreads.begin(), readyThreads.end(), tid);
- wakeCPU();
- }
+ return (isReady != readyThreads.end());
}
-void
-InOrderCPU::deactivateThread(ThreadID tid)
-{
- DPRINTF(InOrderCPU, "[tid:%i]: Calling deactivate thread.\n", tid);
-
- if (isThreadActive(tid)) {
- DPRINTF(InOrderCPU,"[tid:%i]: Removing from active threads list\n",
- tid);
- list<ThreadID>::iterator thread_it =
- std::find(activeThreads.begin(), activeThreads.end(), tid);
-
- removePipelineStalls(*thread_it);
-
- //@TODO: change stage status' to Idle?
-
- activeThreads.erase(thread_it);
- }
-}
-
-void
-InOrderCPU::removePipelineStalls(ThreadID tid)
-{
- DPRINTF(InOrderCPU,"[tid:%i]: Removing all pipeline stalls\n",
- tid);
-
- for (int stNum = 0; stNum < NumStages ; stNum++) {
- pipelineStage[stNum]->removeStalls(tid);
- }
-
-}
bool
-InOrderCPU::isThreadInCPU(ThreadID tid)
+InOrderCPU::isThreadSuspended(ThreadID tid)
{
- list<ThreadID>::iterator isCurrent =
- std::find(currentThreads.begin(), currentThreads.end(), tid);
+ list<ThreadID>::iterator isSuspended =
+ std::find(suspendedThreads.begin(), suspendedThreads.end(), tid);
- return (isCurrent != currentThreads.end());
+ return (isSuspended != suspendedThreads.end());
}
void
-InOrderCPU::addToCurrentThreads(ThreadID tid)
-{
- if (!isThreadInCPU(tid)) {
- DPRINTF(InOrderCPU, "Adding Thread %i to current threads list in CPU.\n",
- tid);
- currentThreads.push_back(tid);
- }
+InOrderCPU::activateNextReadyThread()
+{
+ if (readyThreads.size() >= 1) {
+ ThreadID ready_tid = readyThreads.front();
+
+ // Activate in Pipeline
+ activateThread(ready_tid);
+
+ // Activate in Resource Pool
+ resPool->activateAll(ready_tid);
+
+ list<ThreadID>::iterator ready_it =
+ std::find(readyThreads.begin(), readyThreads.end(), ready_tid);
+ readyThreads.erase(ready_it);
+ } else {
+ DPRINTF(InOrderCPU,
+ "Attempting to activate new thread, but No Ready Threads to"
+ "activate.\n");
+ DPRINTF(InOrderCPU,
+ "Unable to switch to next active thread.\n");
+ }
}
void
-InOrderCPU::removeFromCurrentThreads(ThreadID tid)
+InOrderCPU::activateThread(ThreadID tid)
{
- if (isThreadInCPU(tid)) {
+ if (isThreadSuspended(tid)) {
DPRINTF(InOrderCPU,
- "Adding Thread %i to current threads list in CPU.\n", tid);
- list<ThreadID>::iterator isCurrent =
- std::find(currentThreads.begin(), currentThreads.end(), tid);
- currentThreads.erase(isCurrent);
- }
-}
+ "Removing [tid:%i] from suspended threads list.\n", tid);
-bool
-InOrderCPU::isThreadSuspended(ThreadID tid)
-{
- list<ThreadID>::iterator isSuspended =
- std::find(suspendedThreads.begin(), suspendedThreads.end(), tid);
-
- return (isSuspended!= suspendedThreads.end());
-}
+ list<ThreadID>::iterator susp_it =
+ std::find(suspendedThreads.begin(), suspendedThreads.end(),
+ tid);
+ suspendedThreads.erase(susp_it);
+ }
-void
-InOrderCPU::enableVirtProcElement(unsigned vpe)
-{
- DPRINTF(InOrderCPU, "[vpe:%i]: Scheduling "
- "Enabling of concurrent virtual processor execution",
- vpe);
+ if (threadModel == SwitchOnCacheMiss &&
+ numActiveThreads() == 1) {
+ DPRINTF(InOrderCPU,
+ "Ignoring activation of [tid:%i], since [tid:%i] is "
+ "already running.\n", tid, activeThreadId());
+
+ DPRINTF(InOrderCPU,"Placing [tid:%i] on ready threads list\n",
+ tid);
+
+ readyThreads.push_back(tid);
+
+ } else if (!isThreadActive(tid)) {
+ DPRINTF(InOrderCPU,
+ "Adding [tid:%i] to active threads list.\n", tid);
+ activeThreads.push_back(tid);
+
+ activateThreadInPipeline(tid);
- scheduleCpuEvent(EnableVPEs, NoFault, 0/*tid*/, vpe);
-}
+ thread[tid]->lastActivate = curTick;
-void
-InOrderCPU::enableVPEs(unsigned vpe)
-{
- DPRINTF(InOrderCPU, "[vpe:%i]: Enabling Concurrent Execution "
- "virtual processors %i", vpe);
+ tcBase(tid)->setStatus(ThreadContext::Active);
- list<ThreadID>::iterator thread_it = currentThreads.begin();
+ wakeCPU();
- while (thread_it != currentThreads.end()) {
- if (!isThreadSuspended(*thread_it)) {
- activateThread(*thread_it);
- }
- thread_it++;
+ numCtxtSwitches++;
}
}
void
-InOrderCPU::disableVirtProcElement(ThreadID tid, unsigned vpe)
+InOrderCPU::activateThreadInPipeline(ThreadID tid)
{
- DPRINTF(InOrderCPU, "[vpe:%i]: Scheduling "
- "Disabling of concurrent virtual processor execution",
- vpe);
-
- scheduleCpuEvent(DisableVPEs, NoFault, 0/*tid*/, vpe);
+ for (int stNum=0; stNum < NumStages; stNum++) {
+ pipelineStage[stNum]->activateThread(tid);
+ }
}
void
-InOrderCPU::disableVPEs(ThreadID tid, unsigned vpe)
+InOrderCPU::deactivateContext(ThreadID tid, int delay)
{
- DPRINTF(InOrderCPU, "[vpe:%i]: Disabling Concurrent Execution of "
- "virtual processors %i", vpe);
+ DPRINTF(InOrderCPU,"[tid:%i]: Deactivating ...\n", tid);
- unsigned base_vpe = TheISA::getVirtProcNum(tcBase(tid));
-
- list<ThreadID>::iterator thread_it = activeThreads.begin();
+ scheduleCpuEvent(DeactivateThread, NoFault, tid, dummyInst[tid], delay);
- vector<list<ThreadID>::iterator> removeList;
-
- while (thread_it != activeThreads.end()) {
- if (base_vpe != vpe) {
- removeList.push_back(thread_it);
- }
- thread_it++;
- }
+ // Be sure to signal that there's some activity so the CPU doesn't
+ // deschedule itself.
+ activityRec.activity();
- for (int i = 0; i < removeList.size(); i++) {
- activeThreads.erase(removeList[i]);
- }
+ _status = Running;
}
void
-InOrderCPU::enableMultiThreading(unsigned vpe)
+InOrderCPU::deactivateThread(ThreadID tid)
{
- // Schedule event to take place at end of cycle
- DPRINTF(InOrderCPU, "[vpe:%i]: Scheduling Enable Multithreading on "
- "virtual processor %i", vpe);
+ DPRINTF(InOrderCPU, "[tid:%i]: Calling deactivate thread.\n", tid);
- scheduleCpuEvent(EnableThreads, NoFault, 0/*tid*/, vpe);
-}
+ if (isThreadActive(tid)) {
+ DPRINTF(InOrderCPU,"[tid:%i]: Removing from active threads list\n",
+ tid);
+ list<ThreadID>::iterator thread_it =
+ std::find(activeThreads.begin(), activeThreads.end(), tid);
-void
-InOrderCPU::enableThreads(unsigned vpe)
-{
- DPRINTF(InOrderCPU, "[vpe:%i]: Enabling Multithreading on "
- "virtual processor %i", vpe);
+ removePipelineStalls(*thread_it);
- list<ThreadID>::iterator thread_it = currentThreads.begin();
+ activeThreads.erase(thread_it);
- while (thread_it != currentThreads.end()) {
- if (TheISA::getVirtProcNum(tcBase(*thread_it)) == vpe) {
- if (!isThreadSuspended(*thread_it)) {
- activateThread(*thread_it);
- }
- }
- thread_it++;
+ // Ideally, this should be triggered from the
+ // suspendContext/Thread functions
+ tcBase(tid)->setStatus(ThreadContext::Suspended);
}
-}
-void
-InOrderCPU::disableMultiThreading(ThreadID tid, unsigned vpe)
-{
- // Schedule event to take place at end of cycle
- DPRINTF(InOrderCPU, "[tid:%i]: Scheduling Disable Multithreading on "
- "virtual processor %i", tid, vpe);
- scheduleCpuEvent(DisableThreads, NoFault, tid, vpe);
+ assert(!isThreadActive(tid));
}
void
-InOrderCPU::disableThreads(ThreadID tid, unsigned vpe)
+InOrderCPU::removePipelineStalls(ThreadID tid)
{
- DPRINTF(InOrderCPU, "[tid:%i]: Disabling Multithreading on "
- "virtual processor %i", tid, vpe);
-
- list<ThreadID>::iterator thread_it = activeThreads.begin();
-
- vector<list<ThreadID>::iterator> removeList;
+ DPRINTF(InOrderCPU,"[tid:%i]: Removing all pipeline stalls\n",
+ tid);
- while (thread_it != activeThreads.end()) {
- if (TheISA::getVirtProcNum(tcBase(*thread_it)) == vpe) {
- removeList.push_back(thread_it);
- }
- thread_it++;
+ for (int stNum = 0; stNum < NumStages ; stNum++) {
+ pipelineStage[stNum]->removeStalls(tid);
}
- for (int i = 0; i < removeList.size(); i++) {
- activeThreads.erase(removeList[i]);
- }
}
void
{
DPRINTF(InOrderCPU,"[tid:%i]: Activating ...\n", tid);
- scheduleCpuEvent(ActivateThread, NoFault, tid, 0/*vpe*/, delay);
+
+ scheduleCpuEvent(ActivateThread, NoFault, tid, dummyInst[tid], delay);
// Be sure to signal that there's some activity so the CPU doesn't
// deschedule itself.
_status = Running;
}
-
void
-InOrderCPU::suspendContext(ThreadID tid, int delay)
+InOrderCPU::activateNextReadyContext(int delay)
{
- scheduleCpuEvent(SuspendThread, NoFault, tid, 0/*vpe*/, delay);
- //_status = Idle;
-}
+ DPRINTF(InOrderCPU,"Activating next ready thread\n");
-void
-InOrderCPU::suspendThread(ThreadID tid)
-{
- DPRINTF(InOrderCPU,"[tid: %i]: Suspended ...\n", tid);
- deactivateThread(tid);
-}
+ // NOTE: Add 5 to the event priority so that we always activate
+ // threads after we've finished deactivating, squashing,etc.
+ // other threads
+ scheduleCpuEvent(ActivateNextReadyThread, NoFault, 0/*tid*/, dummyInst[0],
+ delay, 5);
-void
-InOrderCPU::deallocateContext(ThreadID tid, int delay)
-{
- scheduleCpuEvent(DeallocateThread, NoFault, tid, 0/*vpe*/, delay);
+ // Be sure to signal that there's some activity so the CPU doesn't
+ // deschedule itself.
+ activityRec.activity();
+
+ _status = Running;
}
void
-InOrderCPU::deallocateThread(ThreadID tid)
+InOrderCPU::haltContext(ThreadID tid, int delay)
{
- DPRINTF(InOrderCPU,"[tid:%i]: Deallocating ...", tid);
+ DPRINTF(InOrderCPU, "[tid:%i]: Calling Halt Context...\n", tid);
- removeFromCurrentThreads(tid);
+ scheduleCpuEvent(HaltThread, NoFault, tid, dummyInst[tid], delay);
- deactivateThread(tid);
-
- squashThreadInPipeline(tid);
+ activityRec.activity();
}
void
-InOrderCPU::squashThreadInPipeline(ThreadID tid)
+InOrderCPU::haltThread(ThreadID tid)
{
- //Squash all instructions in each stage
- for (int stNum=NumStages - 1; stNum >= 0 ; stNum--) {
- pipelineStage[stNum]->squash(0 /*seq_num*/, tid);
+ DPRINTF(InOrderCPU, "[tid:%i]: Placing on Halted Threads List...\n", tid);
+ deactivateThread(tid);
+ squashThreadInPipeline(tid);
+ haltedThreads.push_back(tid);
+
+ tcBase(tid)->setStatus(ThreadContext::Halted);
+
+ if (threadModel == SwitchOnCacheMiss) {
+ activateNextReadyContext();
}
}
void
-InOrderCPU::haltContext(ThreadID tid, int delay)
+InOrderCPU::suspendContext(ThreadID tid, int delay)
{
- DPRINTF(InOrderCPU, "[tid:%i]: Halt context called.\n", tid);
-
- // Halt is same thing as deallocate for now
- // @TODO: Differentiate between halt & deallocate in the CPU
- // model
- deallocateContext(tid, delay);
+ scheduleCpuEvent(SuspendThread, NoFault, tid, dummyInst[tid], delay);
}
void
-InOrderCPU::insertThread(ThreadID tid)
+InOrderCPU::suspendThread(ThreadID tid)
{
- panic("Unimplemented Function\n.");
+ DPRINTF(InOrderCPU, "[tid:%i]: Placing on Suspended Threads List...\n",
+ tid);
+ deactivateThread(tid);
+ suspendedThreads.push_back(tid);
+ thread[tid]->lastSuspend = curTick;
+
+ tcBase(tid)->setStatus(ThreadContext::Suspended);
}
void
-InOrderCPU::removeThread(ThreadID tid)
+InOrderCPU::squashThreadInPipeline(ThreadID tid)
{
- DPRINTF(InOrderCPU, "Removing Thread %i from CPU.\n", tid);
-
- /** Broadcast to CPU resources*/
+ //Squash all instructions in each stage
+ for (int stNum=NumStages - 1; stNum >= 0 ; stNum--) {
+ pipelineStage[stNum]->squash(0 /*seq_num*/, tid);
+ }
}
PipelineStage*
return pipelineStage[stage_num];
}
-
-void
-InOrderCPU::activateWhenReady(ThreadID tid)
-{
- panic("Unimplemented Function\n.");
-}
-
-
uint64_t
InOrderCPU::readPC(ThreadID tid)
{
tid = TheISA::getTargetThread(tcBase(tid));
}
- if (reg_idx < FP_Base_DepTag) { // Integer Register File
+ if (reg_idx < FP_Base_DepTag) {
+ // Integer Register File
return readIntReg(reg_idx, tid);
- } else if (reg_idx < Ctrl_Base_DepTag) { // Float Register File
+ } else if (reg_idx < Ctrl_Base_DepTag) {
+ // Float Register File
reg_idx -= FP_Base_DepTag;
return readFloatRegBits(reg_idx, tid);
} else {
return --(instList[tid].end());
}
+void
+InOrderCPU::updateContextSwitchStats()
+{
+ // Set Average Stat Here, then reset to 0
+ instsPerCtxtSwitch = instsPerSwitch;
+ instsPerSwitch = 0;
+}
+
+
void
InOrderCPU::instDone(DynInstPtr inst, ThreadID tid)
{
- // Set the CPU's PCs - This contributes to the precise state of the CPU which can be used
- // when restoring a thread to the CPU after a fork or after an exception
- // @TODO: Set-Up Grad-Info/Committed-Info to let ThreadState know if it's a branch or not
+ // Set the CPU's PCs - This contributes to the precise state of the CPU
+ // which can be used when restoring a thread to the CPU after after any
+ // type of context switching activity (fork, exception, etc.)
setPC(inst->readPC(), tid);
setNextPC(inst->readNextPC(), tid);
setNextNPC(inst->readNextNPC(), tid);
+ if (inst->isControl()) {
+ thread[tid]->lastGradIsBranch = true;
+ thread[tid]->lastBranchPC = inst->readPC();
+ thread[tid]->lastBranchNextPC = inst->readNextPC();
+ thread[tid]->lastBranchNextNPC = inst->readNextNPC();
+ } else {
+ thread[tid]->lastGradIsBranch = false;
+ }
+
+
// Finalize Trace Data For Instruction
if (inst->traceData) {
//inst->traceData->setCycle(curTick);
inst->traceData = NULL;
}
- // Set Last Graduated Instruction In Thread State
- //thread[tid]->lastGradInst = inst;
-
+ // Increment active thread's instruction count
+ instsPerSwitch++;
+
// Increment thread-state's instruction count
thread[tid]->numInst++;
smtCommittedInsts[tid]++;
}
+ // Instruction-Mix Stats
+ if (inst->isLoad()) {
+ comLoads++;
+ } else if (inst->isStore()) {
+ comStores++;
+ } else if (inst->isControl()) {
+ comBranches++;
+ } else if (inst->isNop()) {
+ comNops++;
+ } else if (inst->isNonSpeculative()) {
+ comNonSpec++;
+ } else if (inst->isInteger()) {
+ comInts++;
+ } else if (inst->isFloating()) {
+ comFloats++;
+ }
+
// Check for instruction-count-based events.
comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
// Broadcast to other resources an instruction
// has been completed
- resPool->scheduleEvent((CPUEventType)ResourcePool::InstGraduated, inst, tid);
+ resPool->scheduleEvent((CPUEventType)ResourcePool::InstGraduated, inst,
+ 0, 0, tid);
// Finally, remove instruction from CPU
removeInst(inst);
}
+// currently unused function, but substitute repetitive code w/this function
+// call
void
InOrderCPU::addToRemoveList(DynInstPtr &inst)
{
removeInstsThisCycle = true;
-
- removeList.push(inst->getInstListIt());
+ if (!inst->isRemoveList()) {
+ DPRINTF(InOrderCPU, "Pushing instruction [tid:%i] PC %#x "
+ "[sn:%lli] to remove list\n",
+ inst->threadNumber, inst->readPC(), inst->seqNum);
+ inst->setRemoveList();
+ removeList.push(inst->getInstListIt());
+ } else {
+ DPRINTF(InOrderCPU, "Ignoring instruction removal for [tid:%i] PC %#x "
+ "[sn:%lli], already remove list\n",
+ inst->threadNumber, inst->readPC(), inst->seqNum);
+ }
+
}
void
removeInstsThisCycle = true;
// Remove the instruction.
- removeList.push(inst->getInstListIt());
+ if (!inst->isRemoveList()) {
+ DPRINTF(InOrderCPU, "Pushing instruction [tid:%i] PC %#x "
+ "[sn:%lli] to remove list\n",
+ inst->threadNumber, inst->readPC(), inst->seqNum);
+ inst->setRemoveList();
+ removeList.push(inst->getInstListIt());
+ } else {
+ DPRINTF(InOrderCPU, "Ignoring instruction removal for [tid:%i] PC %#x "
+ "[sn:%lli], already on remove list\n",
+ inst->threadNumber, inst->readPC(), inst->seqNum);
+ }
+
}
void
inst_iter--;
- DPRINTF(InOrderCPU, "Deleting instructions from CPU instruction "
+ DPRINTF(InOrderCPU, "Squashing instructions from CPU instruction "
"list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
tid, seq_num, (*inst_iter)->seqNum);
(*instIt)->setSquashed();
- removeList.push(instIt);
+ if (!(*instIt)->isRemoveList()) {
+ DPRINTF(InOrderCPU, "Pushing instruction [tid:%i] PC %#x "
+ "[sn:%lli] to remove list\n",
+ (*instIt)->threadNumber, (*instIt)->readPC(),
+ (*instIt)->seqNum);
+ (*instIt)->setRemoveList();
+ removeList.push(instIt);
+ } else {
+ DPRINTF(InOrderCPU, "Ignoring instruction removal for [tid:%i]"
+ " PC %#x [sn:%lli], already on remove list\n",
+ (*instIt)->threadNumber, (*instIt)->readPC(),
+ (*instIt)->seqNum);
+ }
+
}
+
}
"[tid:%i] [sn:%lli] PC %#x\n",
(*removeList.front())->threadNumber,
(*removeList.front())->seqNum,
- (*removeList.front())->readPC());
+ (*removeList.front())->readPC());
DynInstPtr inst = *removeList.front();
ThreadID tid = inst->threadNumber;
// Make Sure Resource Schedule Is Emptied Out
ThePipeline::ResSchedule *inst_sched = &inst->resSched;
while (!inst_sched->empty()) {
- ThePipeline::ScheduleEntry* sch_entry = inst_sched->top();
+ ScheduleEntry* sch_entry = inst_sched->top();
inst_sched->pop();
delete sch_entry;
}
instList[tid].erase(removeList.front());
removeList.pop();
-
- DPRINTF(RefCount, "pop from remove list: [sn:%i]: Refcount = %i.\n",
- inst->seqNum,
- 0/*inst->curCount()*/);
-
}
removeInstsThisCycle = false;
while (!reqRemoveList.empty()) {
ResourceRequest *res_req = reqRemoveList.front();
- DPRINTF(RefCount, "[tid:%i]: Removing Request, "
- "[sn:%lli] [slot:%i] [stage_num:%i] [res:%s] [refcount:%i].\n",
+ DPRINTF(RefCount, "[tid:%i] [sn:%lli]: Removing Request "
+ "[stage_num:%i] [res:%s] [slot:%i] [completed:%i].\n",
res_req->inst->threadNumber,
res_req->inst->seqNum,
- res_req->getSlot(),
res_req->getStageNum(),
res_req->res->name(),
- 0/*res_req->inst->curCount()*/);
+ (res_req->isCompleted()) ?
+ res_req->getComplSlot() : res_req->getSlot(),
+ res_req->isCompleted());
reqRemoveList.pop();
delete res_req;
-
- DPRINTF(RefCount, "after remove request: [sn:%i]: Refcount = %i.\n",
- res_req->inst->seqNum,
- 0/*res_req->inst->curCount()*/);
}
}
DPRINTF(Activity, "Waking up CPU\n");
- //@todo: figure out how to count idleCycles correctly
- //idleCycles += (curTick - 1) - lastRunningCycle;
+ Tick extra_cycles = tickToCycles((curTick - 1) - lastRunningCycle);
+
+ idleCycles += extra_cycles;
+ for (int stage_num = 0; stage_num < NumStages; stage_num++) {
+ pipelineStage[stage_num]->idleCycles += extra_cycles;
+ }
+
+ numCycles += extra_cycles;
- mainEventQueue.schedule(&tickEvent, curTick);
+ mainEventQueue.schedule(&tickEvent, nextCycle(curTick));
}
#if FULL_SYSTEM
return dtb_res->tlb();
}
-template <class T>
Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, T &data, unsigned flags)
+InOrderCPU::read(DynInstPtr inst, Addr addr,
+ uint8_t *data, unsigned size, unsigned flags)
{
//@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
// you want to run w/out caches?
- CacheUnit *cache_res = dynamic_cast<CacheUnit*>(resPool->getResource(dataPortIdx));
-
- return cache_res->read(inst, addr, data, flags);
-}
-
-#ifndef DOXYGEN_SHOULD_SKIP_THIS
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, Twin32_t &data, unsigned flags);
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, Twin64_t &data, unsigned flags);
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, uint64_t &data, unsigned flags);
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, uint32_t &data, unsigned flags);
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, uint16_t &data, unsigned flags);
-
-template
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, uint8_t &data, unsigned flags);
-
-#endif //DOXYGEN_SHOULD_SKIP_THIS
-
-template<>
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, double &data, unsigned flags)
-{
- return read(inst, addr, *(uint64_t*)&data, flags);
-}
-
-template<>
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, float &data, unsigned flags)
-{
- return read(inst, addr, *(uint32_t*)&data, flags);
-}
-
+ CacheUnit *cache_res =
+ dynamic_cast<CacheUnit*>(resPool->getResource(dataPortIdx));
-template<>
-Fault
-InOrderCPU::read(DynInstPtr inst, Addr addr, int32_t &data, unsigned flags)
-{
- return read(inst, addr, (uint32_t&)data, flags);
+ return cache_res->read(inst, addr, data, size, flags);
}
-template <class T>
Fault
-InOrderCPU::write(DynInstPtr inst, T data, Addr addr, unsigned flags,
- uint64_t *write_res)
+InOrderCPU::write(DynInstPtr inst, uint8_t *data, unsigned size,
+ Addr addr, unsigned flags, uint64_t *write_res)
{
//@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
// you want to run w/out caches?
CacheUnit *cache_res =
dynamic_cast<CacheUnit*>(resPool->getResource(dataPortIdx));
- return cache_res->write(inst, data, addr, flags, write_res);
-}
-
-#ifndef DOXYGEN_SHOULD_SKIP_THIS
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, Twin32_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, Twin64_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, uint64_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, uint32_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, uint16_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-template
-Fault
-InOrderCPU::write(DynInstPtr inst, uint8_t data, Addr addr,
- unsigned flags, uint64_t *res);
-
-#endif //DOXYGEN_SHOULD_SKIP_THIS
-
-template<>
-Fault
-InOrderCPU::write(DynInstPtr inst, double data, Addr addr, unsigned flags, uint64_t *res)
-{
- return write(inst, *(uint64_t*)&data, addr, flags, res);
-}
-
-template<>
-Fault
-InOrderCPU::write(DynInstPtr inst, float data, Addr addr, unsigned flags, uint64_t *res)
-{
- return write(inst, *(uint32_t*)&data, addr, flags, res);
-}
-
-
-template<>
-Fault
-InOrderCPU::write(DynInstPtr inst, int32_t data, Addr addr, unsigned flags, uint64_t *res)
-{
- return write(inst, (uint32_t)data, addr, flags, res);
+ return cache_res->write(inst, data, size, addr, flags, write_res);
}