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
)
161 mainEventQueue
.reschedule(this, cpu
->nextCycle(curTick
+
163 else if (!scheduled())
164 mainEventQueue
.schedule(this, cpu
->nextCycle(curTick
+
169 InOrderCPU::CPUEvent::unscheduleEvent()
175 InOrderCPU::InOrderCPU(Params
*params
)
177 cpu_id(params
->cpu_id
),
182 removeInstsThisCycle(false),
183 activityRec(params
->name
, NumStages
, 10, params
->activity
),
185 system(params
->system
),
186 physmem(system
->physmem
),
187 #endif // FULL_SYSTEM
193 deferRegistration(false/*params->deferRegistration*/),
194 stageTracing(params
->stageTracing
),
197 ThreadID active_threads
;
200 resPool
= new ResourcePool(this, params
);
202 // Resize for Multithreading CPUs
203 thread
.resize(numThreads
);
208 active_threads
= params
->workload
.size();
210 if (active_threads
> MaxThreads
) {
211 panic("Workload Size too large. Increase the 'MaxThreads'"
212 "in your InOrder implementation or "
213 "edit your workload size.");
217 if (active_threads
> 1) {
218 threadModel
= (InOrderCPU::ThreadModel
) params
->threadModel
;
220 if (threadModel
== SMT
) {
221 DPRINTF(InOrderCPU
, "Setting Thread Model to SMT.\n");
222 } else if (threadModel
== SwitchOnCacheMiss
) {
223 DPRINTF(InOrderCPU
, "Setting Thread Model to "
224 "Switch On Cache Miss\n");
228 threadModel
= Single
;
235 // Bind the fetch & data ports from the resource pool.
236 fetchPortIdx
= resPool
->getPortIdx(params
->fetchMemPort
);
237 if (fetchPortIdx
== 0) {
238 fatal("Unable to find port to fetch instructions from.\n");
241 dataPortIdx
= resPool
->getPortIdx(params
->dataMemPort
);
242 if (dataPortIdx
== 0) {
243 fatal("Unable to find port for data.\n");
246 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
) {
248 // SMT is not supported in FS mode yet.
249 assert(numThreads
== 1);
250 thread
[tid
] = new Thread(this, 0);
252 if (tid
< (ThreadID
)params
->workload
.size()) {
253 DPRINTF(InOrderCPU
, "Workload[%i] process is %#x\n",
254 tid
, params
->workload
[tid
]->prog_fname
);
256 new Thread(this, tid
, params
->workload
[tid
]);
258 //Allocate Empty thread so M5 can use later
259 //when scheduling threads to CPU
260 Process
* dummy_proc
= params
->workload
[0];
261 thread
[tid
] = new Thread(this, tid
, dummy_proc
);
264 // Eventually set this with parameters...
268 // Setup the TC that will serve as the interface to the threads/CPU.
269 InOrderThreadContext
*tc
= new InOrderThreadContext
;
271 tc
->thread
= thread
[tid
];
273 // Give the thread the TC.
274 thread
[tid
]->tc
= tc
;
275 thread
[tid
]->setFuncExeInst(0);
276 globalSeqNum
[tid
] = 1;
278 // Add the TC to the CPU's list of TC's.
279 this->threadContexts
.push_back(tc
);
282 // Initialize TimeBuffer Stage Queues
283 for (int stNum
=0; stNum
< NumStages
- 1; stNum
++) {
284 stageQueue
[stNum
] = new StageQueue(NumStages
, NumStages
);
285 stageQueue
[stNum
]->id(stNum
);
289 // Set Up Pipeline Stages
290 for (int stNum
=0; stNum
< NumStages
; stNum
++) {
292 pipelineStage
[stNum
] = new FirstStage(params
, stNum
);
294 pipelineStage
[stNum
] = new PipelineStage(params
, stNum
);
296 pipelineStage
[stNum
]->setCPU(this);
297 pipelineStage
[stNum
]->setActiveThreads(&activeThreads
);
298 pipelineStage
[stNum
]->setTimeBuffer(&timeBuffer
);
300 // Take Care of 1st/Nth stages
302 pipelineStage
[stNum
]->setPrevStageQueue(stageQueue
[stNum
- 1]);
303 if (stNum
< NumStages
- 1)
304 pipelineStage
[stNum
]->setNextStageQueue(stageQueue
[stNum
]);
307 // Initialize thread specific variables
308 for (ThreadID tid
= 0; tid
< numThreads
; tid
++) {
309 archRegDepMap
[tid
].setCPU(this);
311 nonSpecInstActive
[tid
] = false;
312 nonSpecSeqNum
[tid
] = 0;
314 squashSeqNum
[tid
] = MaxAddr
;
315 lastSquashCycle
[tid
] = 0;
317 memset(intRegs
[tid
], 0, sizeof(intRegs
[tid
]));
318 memset(floatRegs
.i
[tid
], 0, sizeof(floatRegs
.i
[tid
]));
321 isa
[tid
].expandForMultithreading(numThreads
, 1/*numVirtProcs*/);
323 // Define dummy instructions and resource requests to be used.
324 dummyInst
[tid
] = new InOrderDynInst(this,
330 dummyReq
[tid
] = new ResourceRequest(resPool
->getResource(0),
338 dummyReqInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
339 dummyReqInst
->setSquashed();
341 dummyBufferInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
342 dummyBufferInst
->setSquashed();
344 lastRunningCycle
= curTick
;
346 // Reset CPU to reset state.
348 Fault resetFault
= new ResetFault();
349 resetFault
->invoke(tcBase());
354 dummyBufferInst
->resetInstCount();
356 // Schedule First Tick Event, CPU will reschedule itself from here on out.
357 scheduleTickEvent(0);
360 InOrderCPU::~InOrderCPU()
367 InOrderCPU::regStats()
369 /* Register the Resource Pool's stats here.*/
372 /* Register for each Pipeline Stage */
373 for (int stage_num
=0; stage_num
< ThePipeline::NumStages
; stage_num
++) {
374 pipelineStage
[stage_num
]->regStats();
377 /* Register any of the InOrderCPU's stats here.*/
379 .name(name() + ".instsPerContextSwitch")
380 .desc("Instructions Committed Per Context Switch")
381 .prereq(instsPerCtxtSwitch
);
384 .name(name() + ".contextSwitches")
385 .desc("Number of context switches");
388 .name(name() + ".comLoads")
389 .desc("Number of Load instructions committed");
392 .name(name() + ".comStores")
393 .desc("Number of Store instructions committed");
396 .name(name() + ".comBranches")
397 .desc("Number of Branches instructions committed");
400 .name(name() + ".comNops")
401 .desc("Number of Nop instructions committed");
404 .name(name() + ".comNonSpec")
405 .desc("Number of Non-Speculative instructions committed");
408 .name(name() + ".comInts")
409 .desc("Number of Integer instructions committed");
412 .name(name() + ".comFloats")
413 .desc("Number of Floating Point instructions committed");
416 .name(name() + ".timesIdled")
417 .desc("Number of times that the entire CPU went into an idle state and"
418 " unscheduled itself")
422 .name(name() + ".idleCycles")
423 .desc("Number of cycles cpu's stages were not processed");
426 .name(name() + ".runCycles")
427 .desc("Number of cycles cpu stages are processed.");
430 .name(name() + ".activity")
431 .desc("Percentage of cycles cpu is active")
433 activity
= (runCycles
/ numCycles
) * 100;
437 .name(name() + ".threadCycles")
438 .desc("Total Number of Cycles A Thread Was Active in CPU (Per-Thread)");
441 .name(name() + ".smtCycles")
442 .desc("Total number of cycles that the CPU was in SMT-mode");
446 .name(name() + ".committedInsts")
447 .desc("Number of Instructions Simulated (Per-Thread)");
451 .name(name() + ".smtCommittedInsts")
452 .desc("Number of SMT Instructions Simulated (Per-Thread)");
455 .name(name() + ".committedInsts_total")
456 .desc("Number of Instructions Simulated (Total)");
459 .name(name() + ".cpi")
460 .desc("CPI: Cycles Per Instruction (Per-Thread)")
462 cpi
= numCycles
/ committedInsts
;
465 .name(name() + ".smt_cpi")
466 .desc("CPI: Total SMT-CPI")
468 smtCpi
= smtCycles
/ smtCommittedInsts
;
471 .name(name() + ".cpi_total")
472 .desc("CPI: Total CPI of All Threads")
474 totalCpi
= numCycles
/ totalCommittedInsts
;
477 .name(name() + ".ipc")
478 .desc("IPC: Instructions Per Cycle (Per-Thread)")
480 ipc
= committedInsts
/ numCycles
;
483 .name(name() + ".smt_ipc")
484 .desc("IPC: Total SMT-IPC")
486 smtIpc
= smtCommittedInsts
/ smtCycles
;
489 .name(name() + ".ipc_total")
490 .desc("IPC: Total IPC of All Threads")
492 totalIpc
= totalCommittedInsts
/ numCycles
;
501 DPRINTF(InOrderCPU
, "\n\nInOrderCPU: Ticking main, InOrderCPU.\n");
505 bool pipes_idle
= true;
507 //Tick each of the stages
508 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
509 pipelineStage
[stNum
]->tick();
511 pipes_idle
= pipes_idle
&& pipelineStage
[stNum
]->idle
;
519 // Now advance the time buffers one tick
520 timeBuffer
.advance();
521 for (int sqNum
=0; sqNum
< NumStages
- 1; sqNum
++) {
522 stageQueue
[sqNum
]->advance();
524 activityRec
.advance();
526 // Any squashed requests, events, or insts then remove them now
527 cleanUpRemovedReqs();
528 cleanUpRemovedEvents();
529 cleanUpRemovedInsts();
531 // Re-schedule CPU for this cycle
532 if (!tickEvent
.scheduled()) {
533 if (_status
== SwitchedOut
) {
535 lastRunningCycle
= curTick
;
536 } else if (!activityRec
.active()) {
537 DPRINTF(InOrderCPU
, "sleeping CPU.\n");
538 lastRunningCycle
= curTick
;
541 //Tick next_tick = curTick + cycles(1);
542 //tickEvent.schedule(next_tick);
543 mainEventQueue
.schedule(&tickEvent
, nextCycle(curTick
+ 1));
544 DPRINTF(InOrderCPU
, "Scheduled CPU for next tick @ %i.\n",
545 nextCycle(curTick
+ 1));
550 updateThreadPriority();
557 if (!deferRegistration
) {
558 registerThreadContexts();
561 // Set inSyscall so that the CPU doesn't squash when initially
562 // setting up registers.
563 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
564 thread
[tid
]->inSyscall
= true;
567 for (ThreadID tid
= 0; tid
< numThreads
; tid
++) {
568 ThreadContext
*src_tc
= threadContexts
[tid
];
569 TheISA::initCPU(src_tc
, src_tc
->contextId());
574 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
575 thread
[tid
]->inSyscall
= false;
577 // Call Initializiation Routine for Resource Pool
584 for (int i
= 0; i
< numThreads
; i
++) {
585 isa
[i
].reset(coreType
, numThreads
,
586 1/*numVirtProcs*/, dynamic_cast<BaseCPU
*>(this));
591 InOrderCPU::getPort(const std::string
&if_name
, int idx
)
593 return resPool
->getPort(if_name
, idx
);
598 InOrderCPU::hwrei(ThreadID tid
)
600 panic("hwrei: Unimplemented");
607 InOrderCPU::simPalCheck(int palFunc
, ThreadID tid
)
609 panic("simPalCheck: Unimplemented");
616 InOrderCPU::getInterrupts()
618 // Check if there are any outstanding interrupts
619 return this->interrupts
->getInterrupt(this->threadContexts
[0]);
624 InOrderCPU::processInterrupts(Fault interrupt
)
626 // Check for interrupts here. For now can copy the code that
627 // exists within isa_fullsys_traits.hh. Also assume that thread 0
628 // is the one that handles the interrupts.
629 // @todo: Possibly consolidate the interrupt checking code.
630 // @todo: Allow other threads to handle interrupts.
632 assert(interrupt
!= NoFault
);
633 this->interrupts
->updateIntrInfo(this->threadContexts
[0]);
635 DPRINTF(InOrderCPU
, "Interrupt %s being handled\n", interrupt
->name());
636 static StaticInstPtr
dummyStatic(TheISA::NoopMachInst
, 0);
637 static DynInstPtr dummyDyn
= new Impl::DynInst(dummyStatic
);
638 this->trap(interrupt
, dummyDyn
);
643 InOrderCPU::updateMemPorts()
645 // Update all ThreadContext's memory ports (Functional/Virtual
647 ThreadID size
= thread
.size();
648 for (ThreadID i
= 0; i
< size
; ++i
)
649 thread
[i
]->connectMemPorts(thread
[i
]->getTC());
654 InOrderCPU::trap(Fault fault
, ThreadID tid
, DynInstPtr inst
, int delay
)
656 //@ Squash Pipeline during TRAP
657 scheduleCpuEvent(Trap
, fault
, tid
, inst
, delay
);
661 InOrderCPU::trapCPU(Fault fault
, ThreadID tid
, DynInstPtr inst
)
663 fault
->invoke(tcBase(tid
), inst
->staticInst
);
667 InOrderCPU::squashFromMemStall(DynInstPtr inst
, ThreadID tid
, int delay
)
669 scheduleCpuEvent(SquashFromMemStall
, NoFault
, tid
, inst
, delay
);
674 InOrderCPU::squashDueToMemStall(int stage_num
, InstSeqNum seq_num
,
677 DPRINTF(InOrderCPU
, "Squashing Pipeline Stages Due to Memory Stall...\n");
679 // Squash all instructions in each stage including
680 // instruction that caused the squash (seq_num - 1)
681 // NOTE: The stage bandwidth needs to be cleared so thats why
682 // the stalling instruction is squashed as well. The stalled
683 // instruction is previously placed in another intermediate buffer
684 // while it's stall is being handled.
685 InstSeqNum squash_seq_num
= seq_num
- 1;
687 for (int stNum
=stage_num
; stNum
>= 0 ; stNum
--) {
688 pipelineStage
[stNum
]->squashDueToMemStall(squash_seq_num
, tid
);
693 InOrderCPU::scheduleCpuEvent(CPUEventType c_event
, Fault fault
,
694 ThreadID tid
, DynInstPtr inst
,
695 unsigned delay
, unsigned event_pri_offset
)
697 CPUEvent
*cpu_event
= new CPUEvent(this, c_event
, fault
, tid
, inst
,
700 Tick sked_tick
= nextCycle(curTick
+ ticks(delay
));
702 DPRINTF(InOrderCPU
, "Scheduling CPU Event (%s) for cycle %i, [tid:%i].\n",
703 eventNames
[c_event
], curTick
+ delay
, tid
);
704 mainEventQueue
.schedule(cpu_event
, sked_tick
);
706 cpu_event
->process();
707 cpuEventRemoveList
.push(cpu_event
);
710 // Broadcast event to the Resource Pool
711 // Need to reset tid just in case this is a dummy instruction
713 resPool
->scheduleEvent(c_event
, inst
, 0, 0, tid
);
717 InOrderCPU::isThreadActive(ThreadID tid
)
719 list
<ThreadID
>::iterator isActive
=
720 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
722 return (isActive
!= activeThreads
.end());
726 InOrderCPU::isThreadReady(ThreadID tid
)
728 list
<ThreadID
>::iterator isReady
=
729 std::find(readyThreads
.begin(), readyThreads
.end(), tid
);
731 return (isReady
!= readyThreads
.end());
735 InOrderCPU::isThreadSuspended(ThreadID tid
)
737 list
<ThreadID
>::iterator isSuspended
=
738 std::find(suspendedThreads
.begin(), suspendedThreads
.end(), tid
);
740 return (isSuspended
!= suspendedThreads
.end());
744 InOrderCPU::activateNextReadyThread()
746 if (readyThreads
.size() >= 1) {
747 ThreadID ready_tid
= readyThreads
.front();
749 // Activate in Pipeline
750 activateThread(ready_tid
);
752 // Activate in Resource Pool
753 resPool
->activateAll(ready_tid
);
755 list
<ThreadID
>::iterator ready_it
=
756 std::find(readyThreads
.begin(), readyThreads
.end(), ready_tid
);
757 readyThreads
.erase(ready_it
);
760 "Attempting to activate new thread, but No Ready Threads to"
763 "Unable to switch to next active thread.\n");
768 InOrderCPU::activateThread(ThreadID tid
)
770 if (isThreadSuspended(tid
)) {
772 "Removing [tid:%i] from suspended threads list.\n", tid
);
774 list
<ThreadID
>::iterator susp_it
=
775 std::find(suspendedThreads
.begin(), suspendedThreads
.end(),
777 suspendedThreads
.erase(susp_it
);
780 if (threadModel
== SwitchOnCacheMiss
&&
781 numActiveThreads() == 1) {
783 "Ignoring activation of [tid:%i], since [tid:%i] is "
784 "already running.\n", tid
, activeThreadId());
786 DPRINTF(InOrderCPU
,"Placing [tid:%i] on ready threads list\n",
789 readyThreads
.push_back(tid
);
791 } else if (!isThreadActive(tid
)) {
793 "Adding [tid:%i] to active threads list.\n", tid
);
794 activeThreads
.push_back(tid
);
796 activateThreadInPipeline(tid
);
798 thread
[tid
]->lastActivate
= curTick
;
800 tcBase(tid
)->setStatus(ThreadContext::Active
);
809 InOrderCPU::activateThreadInPipeline(ThreadID tid
)
811 for (int stNum
=0; stNum
< NumStages
; stNum
++) {
812 pipelineStage
[stNum
]->activateThread(tid
);
817 InOrderCPU::deactivateContext(ThreadID tid
, int delay
)
819 DPRINTF(InOrderCPU
,"[tid:%i]: Deactivating ...\n", tid
);
821 scheduleCpuEvent(DeactivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
823 // Be sure to signal that there's some activity so the CPU doesn't
824 // deschedule itself.
825 activityRec
.activity();
831 InOrderCPU::deactivateThread(ThreadID tid
)
833 DPRINTF(InOrderCPU
, "[tid:%i]: Calling deactivate thread.\n", tid
);
835 if (isThreadActive(tid
)) {
836 DPRINTF(InOrderCPU
,"[tid:%i]: Removing from active threads list\n",
838 list
<ThreadID
>::iterator thread_it
=
839 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
841 removePipelineStalls(*thread_it
);
843 activeThreads
.erase(thread_it
);
845 // Ideally, this should be triggered from the
846 // suspendContext/Thread functions
847 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
850 assert(!isThreadActive(tid
));
854 InOrderCPU::removePipelineStalls(ThreadID tid
)
856 DPRINTF(InOrderCPU
,"[tid:%i]: Removing all pipeline stalls\n",
859 for (int stNum
= 0; stNum
< NumStages
; stNum
++) {
860 pipelineStage
[stNum
]->removeStalls(tid
);
866 InOrderCPU::updateThreadPriority()
868 if (activeThreads
.size() > 1)
870 //DEFAULT TO ROUND ROBIN SCHEME
871 //e.g. Move highest priority to end of thread list
872 list
<ThreadID
>::iterator list_begin
= activeThreads
.begin();
873 list
<ThreadID
>::iterator list_end
= activeThreads
.end();
875 unsigned high_thread
= *list_begin
;
877 activeThreads
.erase(list_begin
);
879 activeThreads
.push_back(high_thread
);
884 InOrderCPU::tickThreadStats()
886 /** Keep track of cycles that each thread is active */
887 list
<ThreadID
>::iterator thread_it
= activeThreads
.begin();
888 while (thread_it
!= activeThreads
.end()) {
889 threadCycles
[*thread_it
]++;
893 // Keep track of cycles where SMT is active
894 if (activeThreads
.size() > 1) {
900 InOrderCPU::activateContext(ThreadID tid
, int delay
)
902 DPRINTF(InOrderCPU
,"[tid:%i]: Activating ...\n", tid
);
905 scheduleCpuEvent(ActivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
907 // Be sure to signal that there's some activity so the CPU doesn't
908 // deschedule itself.
909 activityRec
.activity();
915 InOrderCPU::activateNextReadyContext(int delay
)
917 DPRINTF(InOrderCPU
,"Activating next ready thread\n");
919 // NOTE: Add 5 to the event priority so that we always activate
920 // threads after we've finished deactivating, squashing,etc.
922 scheduleCpuEvent(ActivateNextReadyThread
, NoFault
, 0/*tid*/, dummyInst
[0],
925 // Be sure to signal that there's some activity so the CPU doesn't
926 // deschedule itself.
927 activityRec
.activity();
933 InOrderCPU::haltContext(ThreadID tid
, int delay
)
935 DPRINTF(InOrderCPU
, "[tid:%i]: Calling Halt Context...\n", tid
);
937 scheduleCpuEvent(HaltThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
939 activityRec
.activity();
943 InOrderCPU::haltThread(ThreadID tid
)
945 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Halted Threads List...\n", tid
);
946 deactivateThread(tid
);
947 squashThreadInPipeline(tid
);
948 haltedThreads
.push_back(tid
);
950 tcBase(tid
)->setStatus(ThreadContext::Halted
);
952 if (threadModel
== SwitchOnCacheMiss
) {
953 activateNextReadyContext();
958 InOrderCPU::suspendContext(ThreadID tid
, int delay
)
960 scheduleCpuEvent(SuspendThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
964 InOrderCPU::suspendThread(ThreadID tid
)
966 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Suspended Threads List...\n",
968 deactivateThread(tid
);
969 suspendedThreads
.push_back(tid
);
970 thread
[tid
]->lastSuspend
= curTick
;
972 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
976 InOrderCPU::squashThreadInPipeline(ThreadID tid
)
978 //Squash all instructions in each stage
979 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
980 pipelineStage
[stNum
]->squash(0 /*seq_num*/, tid
);
985 InOrderCPU::getPipeStage(int stage_num
)
987 return pipelineStage
[stage_num
];
991 InOrderCPU::readIntReg(int reg_idx
, ThreadID tid
)
993 return intRegs
[tid
][reg_idx
];
997 InOrderCPU::readFloatReg(int reg_idx
, ThreadID tid
)
999 return floatRegs
.f
[tid
][reg_idx
];
1003 InOrderCPU::readFloatRegBits(int reg_idx
, ThreadID tid
)
1005 return floatRegs
.i
[tid
][reg_idx
];
1009 InOrderCPU::setIntReg(int reg_idx
, uint64_t val
, ThreadID tid
)
1011 intRegs
[tid
][reg_idx
] = val
;
1016 InOrderCPU::setFloatReg(int reg_idx
, FloatReg val
, ThreadID tid
)
1018 floatRegs
.f
[tid
][reg_idx
] = val
;
1023 InOrderCPU::setFloatRegBits(int reg_idx
, FloatRegBits val
, ThreadID tid
)
1025 floatRegs
.i
[tid
][reg_idx
] = val
;
1029 InOrderCPU::readRegOtherThread(unsigned reg_idx
, ThreadID tid
)
1031 // If Default value is set, then retrieve target thread
1032 if (tid
== InvalidThreadID
) {
1033 tid
= TheISA::getTargetThread(tcBase(tid
));
1036 if (reg_idx
< FP_Base_DepTag
) {
1037 // Integer Register File
1038 return readIntReg(reg_idx
, tid
);
1039 } else if (reg_idx
< Ctrl_Base_DepTag
) {
1040 // Float Register File
1041 reg_idx
-= FP_Base_DepTag
;
1042 return readFloatRegBits(reg_idx
, tid
);
1044 reg_idx
-= Ctrl_Base_DepTag
;
1045 return readMiscReg(reg_idx
, tid
); // Misc. Register File
1049 InOrderCPU::setRegOtherThread(unsigned reg_idx
, const MiscReg
&val
,
1052 // If Default value is set, then retrieve target thread
1053 if (tid
== InvalidThreadID
) {
1054 tid
= TheISA::getTargetThread(tcBase(tid
));
1057 if (reg_idx
< FP_Base_DepTag
) { // Integer Register File
1058 setIntReg(reg_idx
, val
, tid
);
1059 } else if (reg_idx
< Ctrl_Base_DepTag
) { // Float Register File
1060 reg_idx
-= FP_Base_DepTag
;
1061 setFloatRegBits(reg_idx
, val
, tid
);
1063 reg_idx
-= Ctrl_Base_DepTag
;
1064 setMiscReg(reg_idx
, val
, tid
); // Misc. Register File
1069 InOrderCPU::readMiscRegNoEffect(int misc_reg
, ThreadID tid
)
1071 return isa
[tid
].readMiscRegNoEffect(misc_reg
);
1075 InOrderCPU::readMiscReg(int misc_reg
, ThreadID tid
)
1077 return isa
[tid
].readMiscReg(misc_reg
, tcBase(tid
));
1081 InOrderCPU::setMiscRegNoEffect(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1083 isa
[tid
].setMiscRegNoEffect(misc_reg
, val
);
1087 InOrderCPU::setMiscReg(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1089 isa
[tid
].setMiscReg(misc_reg
, val
, tcBase(tid
));
1094 InOrderCPU::addInst(DynInstPtr
&inst
)
1096 ThreadID tid
= inst
->readTid();
1098 instList
[tid
].push_back(inst
);
1100 return --(instList
[tid
].end());
1104 InOrderCPU::updateContextSwitchStats()
1106 // Set Average Stat Here, then reset to 0
1107 instsPerCtxtSwitch
= instsPerSwitch
;
1113 InOrderCPU::instDone(DynInstPtr inst
, ThreadID tid
)
1115 // Set the CPU's PCs - This contributes to the precise state of the CPU
1116 // which can be used when restoring a thread to the CPU after after any
1117 // type of context switching activity (fork, exception, etc.)
1118 pcState(inst
->pcState(), tid
);
1120 if (inst
->isControl()) {
1121 thread
[tid
]->lastGradIsBranch
= true;
1122 thread
[tid
]->lastBranchPC
= inst
->pcState();
1123 TheISA::advancePC(thread
[tid
]->lastBranchPC
, inst
->staticInst
);
1125 thread
[tid
]->lastGradIsBranch
= false;
1129 // Finalize Trace Data For Instruction
1130 if (inst
->traceData
) {
1131 //inst->traceData->setCycle(curTick);
1132 inst
->traceData
->setFetchSeq(inst
->seqNum
);
1133 //inst->traceData->setCPSeq(cpu->tcBase(tid)->numInst);
1134 inst
->traceData
->dump();
1135 delete inst
->traceData
;
1136 inst
->traceData
= NULL
;
1139 // Increment active thread's instruction count
1142 // Increment thread-state's instruction count
1143 thread
[tid
]->numInst
++;
1145 // Increment thread-state's instruction stats
1146 thread
[tid
]->numInsts
++;
1148 // Count committed insts per thread stats
1149 committedInsts
[tid
]++;
1151 // Count total insts committed stat
1152 totalCommittedInsts
++;
1154 // Count SMT-committed insts per thread stat
1155 if (numActiveThreads() > 1) {
1156 smtCommittedInsts
[tid
]++;
1159 // Instruction-Mix Stats
1160 if (inst
->isLoad()) {
1162 } else if (inst
->isStore()) {
1164 } else if (inst
->isControl()) {
1166 } else if (inst
->isNop()) {
1168 } else if (inst
->isNonSpeculative()) {
1170 } else if (inst
->isInteger()) {
1172 } else if (inst
->isFloating()) {
1176 // Check for instruction-count-based events.
1177 comInstEventQueue
[tid
]->serviceEvents(thread
[tid
]->numInst
);
1179 // Broadcast to other resources an instruction
1180 // has been completed
1181 resPool
->scheduleEvent((CPUEventType
)ResourcePool::InstGraduated
, inst
,
1184 // Finally, remove instruction from CPU
1188 // currently unused function, but substitute repetitive code w/this function
1191 InOrderCPU::addToRemoveList(DynInstPtr
&inst
)
1193 removeInstsThisCycle
= true;
1194 if (!inst
->isRemoveList()) {
1195 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1196 "[sn:%lli] to remove list\n",
1197 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1198 inst
->setRemoveList();
1199 removeList
.push(inst
->getInstListIt());
1201 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1202 "[sn:%lli], already remove list\n",
1203 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1209 InOrderCPU::removeInst(DynInstPtr
&inst
)
1211 DPRINTF(InOrderCPU
, "Removing graduated instruction [tid:%i] PC %s "
1213 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1215 removeInstsThisCycle
= true;
1217 // Remove the instruction.
1218 if (!inst
->isRemoveList()) {
1219 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1220 "[sn:%lli] to remove list\n",
1221 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1222 inst
->setRemoveList();
1223 removeList
.push(inst
->getInstListIt());
1225 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1226 "[sn:%lli], already on remove list\n",
1227 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1233 InOrderCPU::removeInstsUntil(const InstSeqNum
&seq_num
, ThreadID tid
)
1235 //assert(!instList[tid].empty());
1237 removeInstsThisCycle
= true;
1239 ListIt inst_iter
= instList
[tid
].end();
1243 DPRINTF(InOrderCPU
, "Squashing instructions from CPU instruction "
1244 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1245 tid
, seq_num
, (*inst_iter
)->seqNum
);
1247 while ((*inst_iter
)->seqNum
> seq_num
) {
1249 bool break_loop
= (inst_iter
== instList
[tid
].begin());
1251 squashInstIt(inst_iter
, tid
);
1262 InOrderCPU::squashInstIt(const ListIt
&instIt
, ThreadID tid
)
1264 if ((*instIt
)->threadNumber
== tid
) {
1265 DPRINTF(InOrderCPU
, "Squashing instruction, "
1266 "[tid:%i] [sn:%lli] PC %s\n",
1267 (*instIt
)->threadNumber
,
1269 (*instIt
)->pcState());
1271 (*instIt
)->setSquashed();
1273 if (!(*instIt
)->isRemoveList()) {
1274 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1275 "[sn:%lli] to remove list\n",
1276 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1278 (*instIt
)->setRemoveList();
1279 removeList
.push(instIt
);
1281 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i]"
1282 " PC %s [sn:%lli], already on remove list\n",
1283 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1293 InOrderCPU::cleanUpRemovedInsts()
1295 while (!removeList
.empty()) {
1296 DPRINTF(InOrderCPU
, "Removing instruction, "
1297 "[tid:%i] [sn:%lli] PC %s\n",
1298 (*removeList
.front())->threadNumber
,
1299 (*removeList
.front())->seqNum
,
1300 (*removeList
.front())->pcState());
1302 DynInstPtr inst
= *removeList
.front();
1303 ThreadID tid
= inst
->threadNumber
;
1305 // Make Sure Resource Schedule Is Emptied Out
1306 ThePipeline::ResSchedule
*inst_sched
= &inst
->resSched
;
1307 while (!inst_sched
->empty()) {
1308 ScheduleEntry
* sch_entry
= inst_sched
->top();
1313 // Remove From Register Dependency Map, If Necessary
1314 archRegDepMap
[(*removeList
.front())->threadNumber
].
1315 remove((*removeList
.front()));
1318 // Clear if Non-Speculative
1319 if (inst
->staticInst
&&
1320 inst
->seqNum
== nonSpecSeqNum
[tid
] &&
1321 nonSpecInstActive
[tid
] == true) {
1322 nonSpecInstActive
[tid
] = false;
1325 instList
[tid
].erase(removeList
.front());
1330 removeInstsThisCycle
= false;
1334 InOrderCPU::cleanUpRemovedReqs()
1336 while (!reqRemoveList
.empty()) {
1337 ResourceRequest
*res_req
= reqRemoveList
.front();
1339 DPRINTF(RefCount
, "[tid:%i] [sn:%lli]: Removing Request "
1340 "[stage_num:%i] [res:%s] [slot:%i] [completed:%i].\n",
1341 res_req
->inst
->threadNumber
,
1342 res_req
->inst
->seqNum
,
1343 res_req
->getStageNum(),
1344 res_req
->res
->name(),
1345 (res_req
->isCompleted()) ?
1346 res_req
->getComplSlot() : res_req
->getSlot(),
1347 res_req
->isCompleted());
1349 reqRemoveList
.pop();
1356 InOrderCPU::cleanUpRemovedEvents()
1358 while (!cpuEventRemoveList
.empty()) {
1359 Event
*cpu_event
= cpuEventRemoveList
.front();
1360 cpuEventRemoveList
.pop();
1367 InOrderCPU::dumpInsts()
1371 ListIt inst_list_it
= instList
[0].begin();
1373 cprintf("Dumping Instruction List\n");
1375 while (inst_list_it
!= instList
[0].end()) {
1376 cprintf("Instruction:%i\nPC:%s\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1378 num
, (*inst_list_it
)->pcState(),
1379 (*inst_list_it
)->threadNumber
,
1380 (*inst_list_it
)->seqNum
, (*inst_list_it
)->isIssued(),
1381 (*inst_list_it
)->isSquashed());
1388 InOrderCPU::wakeCPU()
1390 if (/*activityRec.active() || */tickEvent
.scheduled()) {
1391 DPRINTF(Activity
, "CPU already running.\n");
1395 DPRINTF(Activity
, "Waking up CPU\n");
1397 Tick extra_cycles
= tickToCycles((curTick
- 1) - lastRunningCycle
);
1399 idleCycles
+= extra_cycles
;
1400 for (int stage_num
= 0; stage_num
< NumStages
; stage_num
++) {
1401 pipelineStage
[stage_num
]->idleCycles
+= extra_cycles
;
1404 numCycles
+= extra_cycles
;
1406 mainEventQueue
.schedule(&tickEvent
, nextCycle(curTick
));
1412 InOrderCPU::wakeup()
1414 if (this->thread
[0]->status() != ThreadContext::Suspended
)
1419 DPRINTF(Quiesce
, "Suspended Processor woken\n");
1420 this->threadContexts
[0]->activate();
1426 InOrderCPU::syscall(int64_t callnum
, ThreadID tid
)
1428 DPRINTF(InOrderCPU
, "[tid:%i] Executing syscall().\n\n", tid
);
1430 DPRINTF(Activity
,"Activity: syscall() called.\n");
1432 // Temporarily increase this by one to account for the syscall
1434 ++(this->thread
[tid
]->funcExeInst
);
1436 // Execute the actual syscall.
1437 this->thread
[tid
]->syscall(callnum
);
1439 // Decrease funcExeInst by one as the normal commit will handle
1441 --(this->thread
[tid
]->funcExeInst
);
1443 // Clear Non-Speculative Block Variable
1444 nonSpecInstActive
[tid
] = false;
1449 InOrderCPU::prefetch(DynInstPtr inst
)
1451 Resource
*mem_res
= resPool
->getResource(dataPortIdx
);
1452 return mem_res
->prefetch(inst
);
1456 InOrderCPU::writeHint(DynInstPtr inst
)
1458 Resource
*mem_res
= resPool
->getResource(dataPortIdx
);
1459 return mem_res
->writeHint(inst
);
1464 InOrderCPU::getITBPtr()
1466 CacheUnit
*itb_res
=
1467 dynamic_cast<CacheUnit
*>(resPool
->getResource(fetchPortIdx
));
1468 return itb_res
->tlb();
1473 InOrderCPU::getDTBPtr()
1475 CacheUnit
*dtb_res
=
1476 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1477 return dtb_res
->tlb();
1481 InOrderCPU::read(DynInstPtr inst
, Addr addr
,
1482 uint8_t *data
, unsigned size
, unsigned flags
)
1484 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1485 // you want to run w/out caches?
1486 CacheUnit
*cache_res
=
1487 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1489 return cache_res
->read(inst
, addr
, data
, size
, flags
);
1493 InOrderCPU::write(DynInstPtr inst
, uint8_t *data
, unsigned size
,
1494 Addr addr
, unsigned flags
, uint64_t *write_res
)
1496 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1497 // you want to run w/out caches?
1498 CacheUnit
*cache_res
=
1499 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1500 return cache_res
->write(inst
, data
, size
, addr
, flags
, write_res
);