2 * Copyright (c) 2007 MIPS Technologies, Inc.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 * Authors: Korey Sewell
34 #include "arch/utility.hh"
35 #include "config/full_system.hh"
36 #include "config/the_isa.hh"
37 #include "cpu/activity.hh"
38 #include "cpu/base.hh"
39 #include "cpu/exetrace.hh"
40 #include "cpu/inorder/cpu.hh"
41 #include "cpu/inorder/first_stage.hh"
42 #include "cpu/inorder/inorder_dyn_inst.hh"
43 #include "cpu/inorder/pipeline_traits.hh"
44 #include "cpu/inorder/resource_pool.hh"
45 #include "cpu/inorder/resources/resource_list.hh"
46 #include "cpu/inorder/thread_context.hh"
47 #include "cpu/inorder/thread_state.hh"
48 #include "cpu/simple_thread.hh"
49 #include "cpu/thread_context.hh"
50 #include "mem/translating_port.hh"
51 #include "params/InOrderCPU.hh"
52 #include "sim/process.hh"
53 #include "sim/stat_control.hh"
56 #include "cpu/quiesce_event.hh"
57 #include "sim/system.hh"
60 #if THE_ISA == ALPHA_ISA
61 #include "arch/alpha/osfpal.hh"
65 using namespace TheISA
;
66 using namespace ThePipeline
;
68 InOrderCPU::TickEvent::TickEvent(InOrderCPU
*c
)
69 : Event(CPU_Tick_Pri
), cpu(c
)
74 InOrderCPU::TickEvent::process()
81 InOrderCPU::TickEvent::description()
83 return "InOrderCPU tick event";
86 InOrderCPU::CPUEvent::CPUEvent(InOrderCPU
*_cpu
, CPUEventType e_type
,
87 Fault fault
, ThreadID _tid
, DynInstPtr inst
,
88 unsigned event_pri_offset
)
89 : Event(Event::Priority((unsigned int)CPU_Tick_Pri
+ event_pri_offset
)),
92 setEvent(e_type
, fault
, _tid
, inst
);
96 std::string
InOrderCPU::eventNames
[NumCPUEvents
] =
99 "ActivateNextReadyThread",
105 "SquashFromMemStall",
110 InOrderCPU::CPUEvent::process()
112 switch (cpuEventType
)
115 cpu
->activateThread(tid
);
118 case ActivateNextReadyThread
:
119 cpu
->activateNextReadyThread();
122 case DeactivateThread
:
123 cpu
->deactivateThread(tid
);
127 cpu
->haltThread(tid
);
131 cpu
->suspendThread(tid
);
134 case SquashFromMemStall
:
135 cpu
->squashDueToMemStall(inst
->squashingStage
, inst
->seqNum
, tid
);
139 cpu
->trapCPU(fault
, tid
, inst
);
143 fatal("Unrecognized Event Type %s", eventNames
[cpuEventType
]);
146 cpu
->cpuEventRemoveList
.push(this);
152 InOrderCPU::CPUEvent::description()
154 return "InOrderCPU event";
158 InOrderCPU::CPUEvent::scheduleEvent(int delay
)
160 assert(!scheduled() || squashed());
161 cpu
->reschedule(this, cpu
->nextCycle(curTick() + cpu
->ticks(delay
)), true);
165 InOrderCPU::CPUEvent::unscheduleEvent()
171 InOrderCPU::InOrderCPU(Params
*params
)
173 cpu_id(params
->cpu_id
),
177 stageWidth(params
->stageWidth
),
179 removeInstsThisCycle(false),
180 activityRec(params
->name
, NumStages
, 10, params
->activity
),
182 system(params
->system
),
183 physmem(system
->physmem
),
184 #endif // FULL_SYSTEM
190 deferRegistration(false/*params->deferRegistration*/),
191 stageTracing(params
->stageTracing
),
194 ThreadID active_threads
;
197 resPool
= new ResourcePool(this, params
);
199 // Resize for Multithreading CPUs
200 thread
.resize(numThreads
);
205 active_threads
= params
->workload
.size();
207 if (active_threads
> MaxThreads
) {
208 panic("Workload Size too large. Increase the 'MaxThreads'"
209 "in your InOrder implementation or "
210 "edit your workload size.");
214 if (active_threads
> 1) {
215 threadModel
= (InOrderCPU::ThreadModel
) params
->threadModel
;
217 if (threadModel
== SMT
) {
218 DPRINTF(InOrderCPU
, "Setting Thread Model to SMT.\n");
219 } else if (threadModel
== SwitchOnCacheMiss
) {
220 DPRINTF(InOrderCPU
, "Setting Thread Model to "
221 "Switch On Cache Miss\n");
225 threadModel
= Single
;
232 // Bind the fetch & data ports from the resource pool.
233 fetchPortIdx
= resPool
->getPortIdx(params
->fetchMemPort
);
234 if (fetchPortIdx
== 0) {
235 fatal("Unable to find port to fetch instructions from.\n");
238 dataPortIdx
= resPool
->getPortIdx(params
->dataMemPort
);
239 if (dataPortIdx
== 0) {
240 fatal("Unable to find port for data.\n");
243 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
) {
245 // SMT is not supported in FS mode yet.
246 assert(numThreads
== 1);
247 thread
[tid
] = new Thread(this, 0);
249 if (tid
< (ThreadID
)params
->workload
.size()) {
250 DPRINTF(InOrderCPU
, "Workload[%i] process is %#x\n",
251 tid
, params
->workload
[tid
]->prog_fname
);
253 new Thread(this, tid
, params
->workload
[tid
]);
255 //Allocate Empty thread so M5 can use later
256 //when scheduling threads to CPU
257 Process
* dummy_proc
= params
->workload
[0];
258 thread
[tid
] = new Thread(this, tid
, dummy_proc
);
261 // Eventually set this with parameters...
265 // Setup the TC that will serve as the interface to the threads/CPU.
266 InOrderThreadContext
*tc
= new InOrderThreadContext
;
268 tc
->thread
= thread
[tid
];
270 // Give the thread the TC.
271 thread
[tid
]->tc
= tc
;
272 thread
[tid
]->setFuncExeInst(0);
273 globalSeqNum
[tid
] = 1;
275 // Add the TC to the CPU's list of TC's.
276 this->threadContexts
.push_back(tc
);
279 // Initialize TimeBuffer Stage Queues
280 for (int stNum
=0; stNum
< NumStages
- 1; stNum
++) {
281 stageQueue
[stNum
] = new StageQueue(NumStages
, NumStages
);
282 stageQueue
[stNum
]->id(stNum
);
286 // Set Up Pipeline Stages
287 for (int stNum
=0; stNum
< NumStages
; stNum
++) {
289 pipelineStage
[stNum
] = new FirstStage(params
, stNum
);
291 pipelineStage
[stNum
] = new PipelineStage(params
, stNum
);
293 pipelineStage
[stNum
]->setCPU(this);
294 pipelineStage
[stNum
]->setActiveThreads(&activeThreads
);
295 pipelineStage
[stNum
]->setTimeBuffer(&timeBuffer
);
297 // Take Care of 1st/Nth stages
299 pipelineStage
[stNum
]->setPrevStageQueue(stageQueue
[stNum
- 1]);
300 if (stNum
< NumStages
- 1)
301 pipelineStage
[stNum
]->setNextStageQueue(stageQueue
[stNum
]);
304 // Initialize thread specific variables
305 for (ThreadID tid
= 0; tid
< numThreads
; tid
++) {
306 archRegDepMap
[tid
].setCPU(this);
308 nonSpecInstActive
[tid
] = false;
309 nonSpecSeqNum
[tid
] = 0;
311 squashSeqNum
[tid
] = MaxAddr
;
312 lastSquashCycle
[tid
] = 0;
314 memset(intRegs
[tid
], 0, sizeof(intRegs
[tid
]));
315 memset(floatRegs
.i
[tid
], 0, sizeof(floatRegs
.i
[tid
]));
318 // Define dummy instructions and resource requests to be used.
319 dummyInst
[tid
] = new InOrderDynInst(this,
325 dummyReq
[tid
] = new ResourceRequest(resPool
->getResource(0));
328 dummyReqInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
329 dummyReqInst
->setSquashed();
330 dummyReqInst
->resetInstCount();
332 dummyBufferInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
333 dummyBufferInst
->setSquashed();
334 dummyBufferInst
->resetInstCount();
336 endOfSkedIt
= skedCache
.end();
337 frontEndSked
= createFrontEndSked();
339 lastRunningCycle
= curTick();
341 // Reset CPU to reset state.
343 Fault resetFault
= new ResetFault();
344 resetFault
->invoke(tcBase());
348 // Schedule First Tick Event, CPU will reschedule itself from here on out.
349 scheduleTickEvent(0);
352 InOrderCPU::~InOrderCPU()
356 std::map
<SkedID
, ThePipeline::RSkedPtr
>::iterator sked_it
=
358 std::map
<SkedID
, ThePipeline::RSkedPtr
>::iterator sked_end
=
361 while (sked_it
!= sked_end
) {
362 delete (*sked_it
).second
;
368 std::map
<InOrderCPU::SkedID
, ThePipeline::RSkedPtr
> InOrderCPU::skedCache
;
371 InOrderCPU::createFrontEndSked()
373 RSkedPtr res_sked
= new ResourceSked();
375 StageScheduler
F(res_sked
, stage_num
++);
376 StageScheduler
D(res_sked
, stage_num
++);
379 F
.needs(FetchSeq
, FetchSeqUnit::AssignNextPC
);
380 F
.needs(ICache
, FetchUnit::InitiateFetch
);
383 D
.needs(ICache
, FetchUnit::CompleteFetch
);
384 D
.needs(Decode
, DecodeUnit::DecodeInst
);
385 D
.needs(BPred
, BranchPredictor::PredictBranch
);
386 D
.needs(FetchSeq
, FetchSeqUnit::UpdateTargetPC
);
389 DPRINTF(SkedCache
, "Resource Sked created for instruction \"front_end\"\n");
395 InOrderCPU::createBackEndSked(DynInstPtr inst
)
397 RSkedPtr res_sked
= lookupSked(inst
);
398 if (res_sked
!= NULL
) {
399 DPRINTF(SkedCache
, "Found %s in sked cache.\n",
403 res_sked
= new ResourceSked();
406 int stage_num
= ThePipeline::BackEndStartStage
;
407 StageScheduler
X(res_sked
, stage_num
++);
408 StageScheduler
M(res_sked
, stage_num
++);
409 StageScheduler
W(res_sked
, stage_num
++);
411 if (!inst
->staticInst
) {
412 warn_once("Static Instruction Object Not Set. Can't Create"
413 " Back End Schedule");
418 for (int idx
=0; idx
< inst
->numSrcRegs(); idx
++) {
419 if (!idx
|| !inst
->isStore()) {
420 X
.needs(RegManager
, UseDefUnit::ReadSrcReg
, idx
);
424 if ( inst
->isNonSpeculative() ) {
425 // skip execution of non speculative insts until later
426 } else if ( inst
->isMemRef() ) {
427 if ( inst
->isLoad() ) {
428 X
.needs(AGEN
, AGENUnit::GenerateAddr
);
430 } else if (inst
->opClass() == IntMultOp
|| inst
->opClass() == IntDivOp
) {
431 X
.needs(MDU
, MultDivUnit::StartMultDiv
);
433 X
.needs(ExecUnit
, ExecutionUnit::ExecuteInst
);
436 if (inst
->opClass() == IntMultOp
|| inst
->opClass() == IntDivOp
) {
437 X
.needs(MDU
, MultDivUnit::EndMultDiv
);
441 if ( inst
->isLoad() ) {
442 M
.needs(DCache
, CacheUnit::InitiateReadData
);
443 } else if ( inst
->isStore() ) {
444 if ( inst
->numSrcRegs() >= 2 ) {
445 M
.needs(RegManager
, UseDefUnit::ReadSrcReg
, 1);
447 M
.needs(AGEN
, AGENUnit::GenerateAddr
);
448 M
.needs(DCache
, CacheUnit::InitiateWriteData
);
453 if ( inst
->isLoad() ) {
454 W
.needs(DCache
, CacheUnit::CompleteReadData
);
455 } else if ( inst
->isStore() ) {
456 W
.needs(DCache
, CacheUnit::CompleteWriteData
);
459 if ( inst
->isNonSpeculative() ) {
460 if ( inst
->isMemRef() ) fatal("Non-Speculative Memory Instruction");
461 W
.needs(ExecUnit
, ExecutionUnit::ExecuteInst
);
464 W
.needs(Grad
, GraduationUnit::GraduateInst
);
466 for (int idx
=0; idx
< inst
->numDestRegs(); idx
++) {
467 W
.needs(RegManager
, UseDefUnit::WriteDestReg
, idx
);
470 // Insert Back Schedule into our cache of
471 // resource schedules
472 addToSkedCache(inst
, res_sked
);
474 DPRINTF(SkedCache
, "Back End Sked Created for instruction: %s (%08p)\n",
475 inst
->instName(), inst
->getMachInst());
482 InOrderCPU::regStats()
484 /* Register the Resource Pool's stats here.*/
487 /* Register for each Pipeline Stage */
488 for (int stage_num
=0; stage_num
< ThePipeline::NumStages
; stage_num
++) {
489 pipelineStage
[stage_num
]->regStats();
492 /* Register any of the InOrderCPU's stats here.*/
494 .name(name() + ".instsPerContextSwitch")
495 .desc("Instructions Committed Per Context Switch")
496 .prereq(instsPerCtxtSwitch
);
499 .name(name() + ".contextSwitches")
500 .desc("Number of context switches");
503 .name(name() + ".comLoads")
504 .desc("Number of Load instructions committed");
507 .name(name() + ".comStores")
508 .desc("Number of Store instructions committed");
511 .name(name() + ".comBranches")
512 .desc("Number of Branches instructions committed");
515 .name(name() + ".comNops")
516 .desc("Number of Nop instructions committed");
519 .name(name() + ".comNonSpec")
520 .desc("Number of Non-Speculative instructions committed");
523 .name(name() + ".comInts")
524 .desc("Number of Integer instructions committed");
527 .name(name() + ".comFloats")
528 .desc("Number of Floating Point instructions committed");
531 .name(name() + ".timesIdled")
532 .desc("Number of times that the entire CPU went into an idle state and"
533 " unscheduled itself")
537 .name(name() + ".idleCycles")
538 .desc("Number of cycles cpu's stages were not processed");
541 .name(name() + ".runCycles")
542 .desc("Number of cycles cpu stages are processed.");
545 .name(name() + ".activity")
546 .desc("Percentage of cycles cpu is active")
548 activity
= (runCycles
/ numCycles
) * 100;
552 .name(name() + ".threadCycles")
553 .desc("Total Number of Cycles A Thread Was Active in CPU (Per-Thread)");
556 .name(name() + ".smtCycles")
557 .desc("Total number of cycles that the CPU was in SMT-mode");
561 .name(name() + ".committedInsts")
562 .desc("Number of Instructions Simulated (Per-Thread)");
566 .name(name() + ".smtCommittedInsts")
567 .desc("Number of SMT Instructions Simulated (Per-Thread)");
570 .name(name() + ".committedInsts_total")
571 .desc("Number of Instructions Simulated (Total)");
574 .name(name() + ".cpi")
575 .desc("CPI: Cycles Per Instruction (Per-Thread)")
577 cpi
= numCycles
/ committedInsts
;
580 .name(name() + ".smt_cpi")
581 .desc("CPI: Total SMT-CPI")
583 smtCpi
= smtCycles
/ smtCommittedInsts
;
586 .name(name() + ".cpi_total")
587 .desc("CPI: Total CPI of All Threads")
589 totalCpi
= numCycles
/ totalCommittedInsts
;
592 .name(name() + ".ipc")
593 .desc("IPC: Instructions Per Cycle (Per-Thread)")
595 ipc
= committedInsts
/ numCycles
;
598 .name(name() + ".smt_ipc")
599 .desc("IPC: Total SMT-IPC")
601 smtIpc
= smtCommittedInsts
/ smtCycles
;
604 .name(name() + ".ipc_total")
605 .desc("IPC: Total IPC of All Threads")
607 totalIpc
= totalCommittedInsts
/ numCycles
;
616 DPRINTF(InOrderCPU
, "\n\nInOrderCPU: Ticking main, InOrderCPU.\n");
620 bool pipes_idle
= true;
622 //Tick each of the stages
623 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
624 pipelineStage
[stNum
]->tick();
626 pipes_idle
= pipes_idle
&& pipelineStage
[stNum
]->idle
;
634 // Now advance the time buffers one tick
635 timeBuffer
.advance();
636 for (int sqNum
=0; sqNum
< NumStages
- 1; sqNum
++) {
637 stageQueue
[sqNum
]->advance();
639 activityRec
.advance();
641 // Any squashed events, or insts then remove them now
642 cleanUpRemovedEvents();
643 cleanUpRemovedInsts();
645 // Re-schedule CPU for this cycle
646 if (!tickEvent
.scheduled()) {
647 if (_status
== SwitchedOut
) {
649 lastRunningCycle
= curTick();
650 } else if (!activityRec
.active()) {
651 DPRINTF(InOrderCPU
, "sleeping CPU.\n");
652 lastRunningCycle
= curTick();
655 //Tick next_tick = curTick() + cycles(1);
656 //tickEvent.schedule(next_tick);
657 schedule(&tickEvent
, nextCycle(curTick() + 1));
658 DPRINTF(InOrderCPU
, "Scheduled CPU for next tick @ %i.\n",
659 nextCycle(curTick() + 1));
664 updateThreadPriority();
671 if (!deferRegistration
) {
672 registerThreadContexts();
675 // Set inSyscall so that the CPU doesn't squash when initially
676 // setting up registers.
677 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
678 thread
[tid
]->inSyscall
= true;
681 for (ThreadID tid
= 0; tid
< numThreads
; tid
++) {
682 ThreadContext
*src_tc
= threadContexts
[tid
];
683 TheISA::initCPU(src_tc
, src_tc
->contextId());
688 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
689 thread
[tid
]->inSyscall
= false;
691 // Call Initializiation Routine for Resource Pool
696 InOrderCPU::getPort(const std::string
&if_name
, int idx
)
698 return resPool
->getPort(if_name
, idx
);
703 InOrderCPU::hwrei(ThreadID tid
)
705 panic("hwrei: Unimplemented");
712 InOrderCPU::simPalCheck(int palFunc
, ThreadID tid
)
714 panic("simPalCheck: Unimplemented");
721 InOrderCPU::getInterrupts()
723 // Check if there are any outstanding interrupts
724 return interrupts
->getInterrupt(threadContexts
[0]);
729 InOrderCPU::processInterrupts(Fault interrupt
)
731 // Check for interrupts here. For now can copy the code that
732 // exists within isa_fullsys_traits.hh. Also assume that thread 0
733 // is the one that handles the interrupts.
734 // @todo: Possibly consolidate the interrupt checking code.
735 // @todo: Allow other threads to handle interrupts.
737 assert(interrupt
!= NoFault
);
738 interrupts
->updateIntrInfo(threadContexts
[0]);
740 DPRINTF(InOrderCPU
, "Interrupt %s being handled\n", interrupt
->name());
742 // Note: Context ID ok here? Impl. of FS mode needs to revisit this
743 trap(interrupt
, threadContexts
[0]->contextId(), dummyBufferInst
);
748 InOrderCPU::updateMemPorts()
750 // Update all ThreadContext's memory ports (Functional/Virtual
752 ThreadID size
= thread
.size();
753 for (ThreadID i
= 0; i
< size
; ++i
)
754 thread
[i
]->connectMemPorts(thread
[i
]->getTC());
759 InOrderCPU::trap(Fault fault
, ThreadID tid
, DynInstPtr inst
, int delay
)
761 //@ Squash Pipeline during TRAP
762 scheduleCpuEvent(Trap
, fault
, tid
, inst
, delay
);
766 InOrderCPU::trapCPU(Fault fault
, ThreadID tid
, DynInstPtr inst
)
768 fault
->invoke(tcBase(tid
), inst
->staticInst
);
772 InOrderCPU::squashFromMemStall(DynInstPtr inst
, ThreadID tid
, int delay
)
774 scheduleCpuEvent(SquashFromMemStall
, NoFault
, tid
, inst
, delay
);
779 InOrderCPU::squashDueToMemStall(int stage_num
, InstSeqNum seq_num
,
782 DPRINTF(InOrderCPU
, "Squashing Pipeline Stages Due to Memory Stall...\n");
784 // Squash all instructions in each stage including
785 // instruction that caused the squash (seq_num - 1)
786 // NOTE: The stage bandwidth needs to be cleared so thats why
787 // the stalling instruction is squashed as well. The stalled
788 // instruction is previously placed in another intermediate buffer
789 // while it's stall is being handled.
790 InstSeqNum squash_seq_num
= seq_num
- 1;
792 for (int stNum
=stage_num
; stNum
>= 0 ; stNum
--) {
793 pipelineStage
[stNum
]->squashDueToMemStall(squash_seq_num
, tid
);
798 InOrderCPU::scheduleCpuEvent(CPUEventType c_event
, Fault fault
,
799 ThreadID tid
, DynInstPtr inst
,
800 unsigned delay
, unsigned event_pri_offset
)
802 CPUEvent
*cpu_event
= new CPUEvent(this, c_event
, fault
, tid
, inst
,
805 Tick sked_tick
= nextCycle(curTick() + ticks(delay
));
807 DPRINTF(InOrderCPU
, "Scheduling CPU Event (%s) for cycle %i, [tid:%i].\n",
808 eventNames
[c_event
], curTick() + delay
, tid
);
809 schedule(cpu_event
, sked_tick
);
811 cpu_event
->process();
812 cpuEventRemoveList
.push(cpu_event
);
815 // Broadcast event to the Resource Pool
816 // Need to reset tid just in case this is a dummy instruction
818 resPool
->scheduleEvent(c_event
, inst
, 0, 0, tid
);
822 InOrderCPU::isThreadActive(ThreadID tid
)
824 list
<ThreadID
>::iterator isActive
=
825 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
827 return (isActive
!= activeThreads
.end());
831 InOrderCPU::isThreadReady(ThreadID tid
)
833 list
<ThreadID
>::iterator isReady
=
834 std::find(readyThreads
.begin(), readyThreads
.end(), tid
);
836 return (isReady
!= readyThreads
.end());
840 InOrderCPU::isThreadSuspended(ThreadID tid
)
842 list
<ThreadID
>::iterator isSuspended
=
843 std::find(suspendedThreads
.begin(), suspendedThreads
.end(), tid
);
845 return (isSuspended
!= suspendedThreads
.end());
849 InOrderCPU::activateNextReadyThread()
851 if (readyThreads
.size() >= 1) {
852 ThreadID ready_tid
= readyThreads
.front();
854 // Activate in Pipeline
855 activateThread(ready_tid
);
857 // Activate in Resource Pool
858 resPool
->activateAll(ready_tid
);
860 list
<ThreadID
>::iterator ready_it
=
861 std::find(readyThreads
.begin(), readyThreads
.end(), ready_tid
);
862 readyThreads
.erase(ready_it
);
865 "Attempting to activate new thread, but No Ready Threads to"
868 "Unable to switch to next active thread.\n");
873 InOrderCPU::activateThread(ThreadID tid
)
875 if (isThreadSuspended(tid
)) {
877 "Removing [tid:%i] from suspended threads list.\n", tid
);
879 list
<ThreadID
>::iterator susp_it
=
880 std::find(suspendedThreads
.begin(), suspendedThreads
.end(),
882 suspendedThreads
.erase(susp_it
);
885 if (threadModel
== SwitchOnCacheMiss
&&
886 numActiveThreads() == 1) {
888 "Ignoring activation of [tid:%i], since [tid:%i] is "
889 "already running.\n", tid
, activeThreadId());
891 DPRINTF(InOrderCPU
,"Placing [tid:%i] on ready threads list\n",
894 readyThreads
.push_back(tid
);
896 } else if (!isThreadActive(tid
)) {
898 "Adding [tid:%i] to active threads list.\n", tid
);
899 activeThreads
.push_back(tid
);
901 activateThreadInPipeline(tid
);
903 thread
[tid
]->lastActivate
= curTick();
905 tcBase(tid
)->setStatus(ThreadContext::Active
);
914 InOrderCPU::activateThreadInPipeline(ThreadID tid
)
916 for (int stNum
=0; stNum
< NumStages
; stNum
++) {
917 pipelineStage
[stNum
]->activateThread(tid
);
922 InOrderCPU::deactivateContext(ThreadID tid
, int delay
)
924 DPRINTF(InOrderCPU
,"[tid:%i]: Deactivating ...\n", tid
);
926 scheduleCpuEvent(DeactivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
928 // Be sure to signal that there's some activity so the CPU doesn't
929 // deschedule itself.
930 activityRec
.activity();
936 InOrderCPU::deactivateThread(ThreadID tid
)
938 DPRINTF(InOrderCPU
, "[tid:%i]: Calling deactivate thread.\n", tid
);
940 if (isThreadActive(tid
)) {
941 DPRINTF(InOrderCPU
,"[tid:%i]: Removing from active threads list\n",
943 list
<ThreadID
>::iterator thread_it
=
944 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
946 removePipelineStalls(*thread_it
);
948 activeThreads
.erase(thread_it
);
950 // Ideally, this should be triggered from the
951 // suspendContext/Thread functions
952 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
955 assert(!isThreadActive(tid
));
959 InOrderCPU::removePipelineStalls(ThreadID tid
)
961 DPRINTF(InOrderCPU
,"[tid:%i]: Removing all pipeline stalls\n",
964 for (int stNum
= 0; stNum
< NumStages
; stNum
++) {
965 pipelineStage
[stNum
]->removeStalls(tid
);
971 InOrderCPU::updateThreadPriority()
973 if (activeThreads
.size() > 1)
975 //DEFAULT TO ROUND ROBIN SCHEME
976 //e.g. Move highest priority to end of thread list
977 list
<ThreadID
>::iterator list_begin
= activeThreads
.begin();
978 list
<ThreadID
>::iterator list_end
= activeThreads
.end();
980 unsigned high_thread
= *list_begin
;
982 activeThreads
.erase(list_begin
);
984 activeThreads
.push_back(high_thread
);
989 InOrderCPU::tickThreadStats()
991 /** Keep track of cycles that each thread is active */
992 list
<ThreadID
>::iterator thread_it
= activeThreads
.begin();
993 while (thread_it
!= activeThreads
.end()) {
994 threadCycles
[*thread_it
]++;
998 // Keep track of cycles where SMT is active
999 if (activeThreads
.size() > 1) {
1005 InOrderCPU::activateContext(ThreadID tid
, int delay
)
1007 DPRINTF(InOrderCPU
,"[tid:%i]: Activating ...\n", tid
);
1010 scheduleCpuEvent(ActivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
1012 // Be sure to signal that there's some activity so the CPU doesn't
1013 // deschedule itself.
1014 activityRec
.activity();
1020 InOrderCPU::activateNextReadyContext(int delay
)
1022 DPRINTF(InOrderCPU
,"Activating next ready thread\n");
1024 // NOTE: Add 5 to the event priority so that we always activate
1025 // threads after we've finished deactivating, squashing,etc.
1027 scheduleCpuEvent(ActivateNextReadyThread
, NoFault
, 0/*tid*/, dummyInst
[0],
1030 // Be sure to signal that there's some activity so the CPU doesn't
1031 // deschedule itself.
1032 activityRec
.activity();
1038 InOrderCPU::haltContext(ThreadID tid
, int delay
)
1040 DPRINTF(InOrderCPU
, "[tid:%i]: Calling Halt Context...\n", tid
);
1042 scheduleCpuEvent(HaltThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
1044 activityRec
.activity();
1048 InOrderCPU::haltThread(ThreadID tid
)
1050 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Halted Threads List...\n", tid
);
1051 deactivateThread(tid
);
1052 squashThreadInPipeline(tid
);
1053 haltedThreads
.push_back(tid
);
1055 tcBase(tid
)->setStatus(ThreadContext::Halted
);
1057 if (threadModel
== SwitchOnCacheMiss
) {
1058 activateNextReadyContext();
1063 InOrderCPU::suspendContext(ThreadID tid
, int delay
)
1065 scheduleCpuEvent(SuspendThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
1069 InOrderCPU::suspendThread(ThreadID tid
)
1071 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Suspended Threads List...\n",
1073 deactivateThread(tid
);
1074 suspendedThreads
.push_back(tid
);
1075 thread
[tid
]->lastSuspend
= curTick();
1077 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
1081 InOrderCPU::squashThreadInPipeline(ThreadID tid
)
1083 //Squash all instructions in each stage
1084 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
1085 pipelineStage
[stNum
]->squash(0 /*seq_num*/, tid
);
1090 InOrderCPU::getPipeStage(int stage_num
)
1092 return pipelineStage
[stage_num
];
1096 InOrderCPU::readIntReg(int reg_idx
, ThreadID tid
)
1098 return intRegs
[tid
][reg_idx
];
1102 InOrderCPU::readFloatReg(int reg_idx
, ThreadID tid
)
1104 return floatRegs
.f
[tid
][reg_idx
];
1108 InOrderCPU::readFloatRegBits(int reg_idx
, ThreadID tid
)
1110 return floatRegs
.i
[tid
][reg_idx
];
1114 InOrderCPU::setIntReg(int reg_idx
, uint64_t val
, ThreadID tid
)
1116 intRegs
[tid
][reg_idx
] = val
;
1121 InOrderCPU::setFloatReg(int reg_idx
, FloatReg val
, ThreadID tid
)
1123 floatRegs
.f
[tid
][reg_idx
] = val
;
1128 InOrderCPU::setFloatRegBits(int reg_idx
, FloatRegBits val
, ThreadID tid
)
1130 floatRegs
.i
[tid
][reg_idx
] = val
;
1134 InOrderCPU::readRegOtherThread(unsigned reg_idx
, ThreadID tid
)
1136 // If Default value is set, then retrieve target thread
1137 if (tid
== InvalidThreadID
) {
1138 tid
= TheISA::getTargetThread(tcBase(tid
));
1141 if (reg_idx
< FP_Base_DepTag
) {
1142 // Integer Register File
1143 return readIntReg(reg_idx
, tid
);
1144 } else if (reg_idx
< Ctrl_Base_DepTag
) {
1145 // Float Register File
1146 reg_idx
-= FP_Base_DepTag
;
1147 return readFloatRegBits(reg_idx
, tid
);
1149 reg_idx
-= Ctrl_Base_DepTag
;
1150 return readMiscReg(reg_idx
, tid
); // Misc. Register File
1154 InOrderCPU::setRegOtherThread(unsigned reg_idx
, const MiscReg
&val
,
1157 // If Default value is set, then retrieve target thread
1158 if (tid
== InvalidThreadID
) {
1159 tid
= TheISA::getTargetThread(tcBase(tid
));
1162 if (reg_idx
< FP_Base_DepTag
) { // Integer Register File
1163 setIntReg(reg_idx
, val
, tid
);
1164 } else if (reg_idx
< Ctrl_Base_DepTag
) { // Float Register File
1165 reg_idx
-= FP_Base_DepTag
;
1166 setFloatRegBits(reg_idx
, val
, tid
);
1168 reg_idx
-= Ctrl_Base_DepTag
;
1169 setMiscReg(reg_idx
, val
, tid
); // Misc. Register File
1174 InOrderCPU::readMiscRegNoEffect(int misc_reg
, ThreadID tid
)
1176 return isa
[tid
].readMiscRegNoEffect(misc_reg
);
1180 InOrderCPU::readMiscReg(int misc_reg
, ThreadID tid
)
1182 return isa
[tid
].readMiscReg(misc_reg
, tcBase(tid
));
1186 InOrderCPU::setMiscRegNoEffect(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1188 isa
[tid
].setMiscRegNoEffect(misc_reg
, val
);
1192 InOrderCPU::setMiscReg(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1194 isa
[tid
].setMiscReg(misc_reg
, val
, tcBase(tid
));
1199 InOrderCPU::addInst(DynInstPtr
&inst
)
1201 ThreadID tid
= inst
->readTid();
1203 instList
[tid
].push_back(inst
);
1205 return --(instList
[tid
].end());
1209 InOrderCPU::updateContextSwitchStats()
1211 // Set Average Stat Here, then reset to 0
1212 instsPerCtxtSwitch
= instsPerSwitch
;
1218 InOrderCPU::instDone(DynInstPtr inst
, ThreadID tid
)
1220 // Set the CPU's PCs - This contributes to the precise state of the CPU
1221 // which can be used when restoring a thread to the CPU after after any
1222 // type of context switching activity (fork, exception, etc.)
1223 pcState(inst
->pcState(), tid
);
1225 if (inst
->isControl()) {
1226 thread
[tid
]->lastGradIsBranch
= true;
1227 thread
[tid
]->lastBranchPC
= inst
->pcState();
1228 TheISA::advancePC(thread
[tid
]->lastBranchPC
, inst
->staticInst
);
1230 thread
[tid
]->lastGradIsBranch
= false;
1234 // Finalize Trace Data For Instruction
1235 if (inst
->traceData
) {
1236 //inst->traceData->setCycle(curTick());
1237 inst
->traceData
->setFetchSeq(inst
->seqNum
);
1238 //inst->traceData->setCPSeq(cpu->tcBase(tid)->numInst);
1239 inst
->traceData
->dump();
1240 delete inst
->traceData
;
1241 inst
->traceData
= NULL
;
1244 // Increment active thread's instruction count
1247 // Increment thread-state's instruction count
1248 thread
[tid
]->numInst
++;
1250 // Increment thread-state's instruction stats
1251 thread
[tid
]->numInsts
++;
1253 // Count committed insts per thread stats
1254 committedInsts
[tid
]++;
1256 // Count total insts committed stat
1257 totalCommittedInsts
++;
1259 // Count SMT-committed insts per thread stat
1260 if (numActiveThreads() > 1) {
1261 smtCommittedInsts
[tid
]++;
1264 // Instruction-Mix Stats
1265 if (inst
->isLoad()) {
1267 } else if (inst
->isStore()) {
1269 } else if (inst
->isControl()) {
1271 } else if (inst
->isNop()) {
1273 } else if (inst
->isNonSpeculative()) {
1275 } else if (inst
->isInteger()) {
1277 } else if (inst
->isFloating()) {
1281 // Check for instruction-count-based events.
1282 comInstEventQueue
[tid
]->serviceEvents(thread
[tid
]->numInst
);
1284 // Broadcast to other resources an instruction
1285 // has been completed
1286 resPool
->scheduleEvent((CPUEventType
)ResourcePool::InstGraduated
, inst
,
1289 // Finally, remove instruction from CPU
1293 // currently unused function, but substitute repetitive code w/this function
1296 InOrderCPU::addToRemoveList(DynInstPtr
&inst
)
1298 removeInstsThisCycle
= true;
1299 if (!inst
->isRemoveList()) {
1300 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1301 "[sn:%lli] to remove list\n",
1302 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1303 inst
->setRemoveList();
1304 removeList
.push(inst
->getInstListIt());
1306 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1307 "[sn:%lli], already remove list\n",
1308 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1314 InOrderCPU::removeInst(DynInstPtr
&inst
)
1316 DPRINTF(InOrderCPU
, "Removing graduated instruction [tid:%i] PC %s "
1318 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1320 removeInstsThisCycle
= true;
1322 // Remove the instruction.
1323 if (!inst
->isRemoveList()) {
1324 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1325 "[sn:%lli] to remove list\n",
1326 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1327 inst
->setRemoveList();
1328 removeList
.push(inst
->getInstListIt());
1330 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1331 "[sn:%lli], already on remove list\n",
1332 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1338 InOrderCPU::removeInstsUntil(const InstSeqNum
&seq_num
, ThreadID tid
)
1340 //assert(!instList[tid].empty());
1342 removeInstsThisCycle
= true;
1344 ListIt inst_iter
= instList
[tid
].end();
1348 DPRINTF(InOrderCPU
, "Squashing instructions from CPU instruction "
1349 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1350 tid
, seq_num
, (*inst_iter
)->seqNum
);
1352 while ((*inst_iter
)->seqNum
> seq_num
) {
1354 bool break_loop
= (inst_iter
== instList
[tid
].begin());
1356 squashInstIt(inst_iter
, tid
);
1367 InOrderCPU::squashInstIt(const ListIt
&instIt
, ThreadID tid
)
1369 if ((*instIt
)->threadNumber
== tid
) {
1370 DPRINTF(InOrderCPU
, "Squashing instruction, "
1371 "[tid:%i] [sn:%lli] PC %s\n",
1372 (*instIt
)->threadNumber
,
1374 (*instIt
)->pcState());
1376 (*instIt
)->setSquashed();
1378 if (!(*instIt
)->isRemoveList()) {
1379 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1380 "[sn:%lli] to remove list\n",
1381 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1383 (*instIt
)->setRemoveList();
1384 removeList
.push(instIt
);
1386 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i]"
1387 " PC %s [sn:%lli], already on remove list\n",
1388 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1398 InOrderCPU::cleanUpRemovedInsts()
1400 while (!removeList
.empty()) {
1401 DPRINTF(InOrderCPU
, "Removing instruction, "
1402 "[tid:%i] [sn:%lli] PC %s\n",
1403 (*removeList
.front())->threadNumber
,
1404 (*removeList
.front())->seqNum
,
1405 (*removeList
.front())->pcState());
1407 DynInstPtr inst
= *removeList
.front();
1408 ThreadID tid
= inst
->threadNumber
;
1410 // Remove From Register Dependency Map, If Necessary
1411 archRegDepMap
[(*removeList
.front())->threadNumber
].
1412 remove((*removeList
.front()));
1415 // Clear if Non-Speculative
1416 if (inst
->staticInst
&&
1417 inst
->seqNum
== nonSpecSeqNum
[tid
] &&
1418 nonSpecInstActive
[tid
] == true) {
1419 nonSpecInstActive
[tid
] = false;
1422 instList
[tid
].erase(removeList
.front());
1427 removeInstsThisCycle
= false;
1431 InOrderCPU::cleanUpRemovedEvents()
1433 while (!cpuEventRemoveList
.empty()) {
1434 Event
*cpu_event
= cpuEventRemoveList
.front();
1435 cpuEventRemoveList
.pop();
1442 InOrderCPU::dumpInsts()
1446 ListIt inst_list_it
= instList
[0].begin();
1448 cprintf("Dumping Instruction List\n");
1450 while (inst_list_it
!= instList
[0].end()) {
1451 cprintf("Instruction:%i\nPC:%s\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1453 num
, (*inst_list_it
)->pcState(),
1454 (*inst_list_it
)->threadNumber
,
1455 (*inst_list_it
)->seqNum
, (*inst_list_it
)->isIssued(),
1456 (*inst_list_it
)->isSquashed());
1463 InOrderCPU::wakeCPU()
1465 if (/*activityRec.active() || */tickEvent
.scheduled()) {
1466 DPRINTF(Activity
, "CPU already running.\n");
1470 DPRINTF(Activity
, "Waking up CPU\n");
1472 Tick extra_cycles
= tickToCycles((curTick() - 1) - lastRunningCycle
);
1474 idleCycles
+= extra_cycles
;
1475 for (int stage_num
= 0; stage_num
< NumStages
; stage_num
++) {
1476 pipelineStage
[stage_num
]->idleCycles
+= extra_cycles
;
1479 numCycles
+= extra_cycles
;
1481 schedule(&tickEvent
, nextCycle(curTick()));
1487 InOrderCPU::wakeup()
1489 if (thread
[0]->status() != ThreadContext::Suspended
)
1494 DPRINTF(Quiesce
, "Suspended Processor woken\n");
1495 threadContexts
[0]->activate();
1501 InOrderCPU::syscall(int64_t callnum
, ThreadID tid
)
1503 DPRINTF(InOrderCPU
, "[tid:%i] Executing syscall().\n\n", tid
);
1505 DPRINTF(Activity
,"Activity: syscall() called.\n");
1507 // Temporarily increase this by one to account for the syscall
1509 ++(this->thread
[tid
]->funcExeInst
);
1511 // Execute the actual syscall.
1512 this->thread
[tid
]->syscall(callnum
);
1514 // Decrease funcExeInst by one as the normal commit will handle
1516 --(this->thread
[tid
]->funcExeInst
);
1518 // Clear Non-Speculative Block Variable
1519 nonSpecInstActive
[tid
] = false;
1524 InOrderCPU::getITBPtr()
1526 CacheUnit
*itb_res
=
1527 dynamic_cast<CacheUnit
*>(resPool
->getResource(fetchPortIdx
));
1528 return itb_res
->tlb();
1533 InOrderCPU::getDTBPtr()
1535 CacheUnit
*dtb_res
=
1536 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1537 return dtb_res
->tlb();
1541 InOrderCPU::read(DynInstPtr inst
, Addr addr
,
1542 uint8_t *data
, unsigned size
, unsigned flags
)
1544 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1545 // you want to run w/out caches?
1546 CacheUnit
*cache_res
=
1547 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1549 return cache_res
->read(inst
, addr
, data
, size
, flags
);
1553 InOrderCPU::write(DynInstPtr inst
, uint8_t *data
, unsigned size
,
1554 Addr addr
, unsigned flags
, uint64_t *write_res
)
1556 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1557 // you want to run w/out caches?
1558 CacheUnit
*cache_res
=
1559 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1560 return cache_res
->write(inst
, data
, size
, addr
, flags
, write_res
);