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 isa
[tid
].expandForMultithreading(numThreads
, 1/*numVirtProcs*/);
320 // Define dummy instructions and resource requests to be used.
321 dummyInst
[tid
] = new InOrderDynInst(this,
327 dummyReq
[tid
] = new ResourceRequest(resPool
->getResource(0),
335 dummyReqInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
336 dummyReqInst
->setSquashed();
338 dummyBufferInst
= new InOrderDynInst(this, NULL
, 0, 0, 0);
339 dummyBufferInst
->setSquashed();
341 lastRunningCycle
= curTick();
343 // Reset CPU to reset state.
345 Fault resetFault
= new ResetFault();
346 resetFault
->invoke(tcBase());
351 dummyBufferInst
->resetInstCount();
353 // Schedule First Tick Event, CPU will reschedule itself from here on out.
354 scheduleTickEvent(0);
357 InOrderCPU::~InOrderCPU()
364 InOrderCPU::regStats()
366 /* Register the Resource Pool's stats here.*/
369 /* Register for each Pipeline Stage */
370 for (int stage_num
=0; stage_num
< ThePipeline::NumStages
; stage_num
++) {
371 pipelineStage
[stage_num
]->regStats();
374 /* Register any of the InOrderCPU's stats here.*/
376 .name(name() + ".instsPerContextSwitch")
377 .desc("Instructions Committed Per Context Switch")
378 .prereq(instsPerCtxtSwitch
);
381 .name(name() + ".contextSwitches")
382 .desc("Number of context switches");
385 .name(name() + ".comLoads")
386 .desc("Number of Load instructions committed");
389 .name(name() + ".comStores")
390 .desc("Number of Store instructions committed");
393 .name(name() + ".comBranches")
394 .desc("Number of Branches instructions committed");
397 .name(name() + ".comNops")
398 .desc("Number of Nop instructions committed");
401 .name(name() + ".comNonSpec")
402 .desc("Number of Non-Speculative instructions committed");
405 .name(name() + ".comInts")
406 .desc("Number of Integer instructions committed");
409 .name(name() + ".comFloats")
410 .desc("Number of Floating Point instructions committed");
413 .name(name() + ".timesIdled")
414 .desc("Number of times that the entire CPU went into an idle state and"
415 " unscheduled itself")
419 .name(name() + ".idleCycles")
420 .desc("Number of cycles cpu's stages were not processed");
423 .name(name() + ".runCycles")
424 .desc("Number of cycles cpu stages are processed.");
427 .name(name() + ".activity")
428 .desc("Percentage of cycles cpu is active")
430 activity
= (runCycles
/ numCycles
) * 100;
434 .name(name() + ".threadCycles")
435 .desc("Total Number of Cycles A Thread Was Active in CPU (Per-Thread)");
438 .name(name() + ".smtCycles")
439 .desc("Total number of cycles that the CPU was in SMT-mode");
443 .name(name() + ".committedInsts")
444 .desc("Number of Instructions Simulated (Per-Thread)");
448 .name(name() + ".smtCommittedInsts")
449 .desc("Number of SMT Instructions Simulated (Per-Thread)");
452 .name(name() + ".committedInsts_total")
453 .desc("Number of Instructions Simulated (Total)");
456 .name(name() + ".cpi")
457 .desc("CPI: Cycles Per Instruction (Per-Thread)")
459 cpi
= numCycles
/ committedInsts
;
462 .name(name() + ".smt_cpi")
463 .desc("CPI: Total SMT-CPI")
465 smtCpi
= smtCycles
/ smtCommittedInsts
;
468 .name(name() + ".cpi_total")
469 .desc("CPI: Total CPI of All Threads")
471 totalCpi
= numCycles
/ totalCommittedInsts
;
474 .name(name() + ".ipc")
475 .desc("IPC: Instructions Per Cycle (Per-Thread)")
477 ipc
= committedInsts
/ numCycles
;
480 .name(name() + ".smt_ipc")
481 .desc("IPC: Total SMT-IPC")
483 smtIpc
= smtCommittedInsts
/ smtCycles
;
486 .name(name() + ".ipc_total")
487 .desc("IPC: Total IPC of All Threads")
489 totalIpc
= totalCommittedInsts
/ numCycles
;
498 DPRINTF(InOrderCPU
, "\n\nInOrderCPU: Ticking main, InOrderCPU.\n");
502 bool pipes_idle
= true;
504 //Tick each of the stages
505 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
506 pipelineStage
[stNum
]->tick();
508 pipes_idle
= pipes_idle
&& pipelineStage
[stNum
]->idle
;
516 // Now advance the time buffers one tick
517 timeBuffer
.advance();
518 for (int sqNum
=0; sqNum
< NumStages
- 1; sqNum
++) {
519 stageQueue
[sqNum
]->advance();
521 activityRec
.advance();
523 // Any squashed requests, events, or insts then remove them now
524 cleanUpRemovedReqs();
525 cleanUpRemovedEvents();
526 cleanUpRemovedInsts();
528 // Re-schedule CPU for this cycle
529 if (!tickEvent
.scheduled()) {
530 if (_status
== SwitchedOut
) {
532 lastRunningCycle
= curTick();
533 } else if (!activityRec
.active()) {
534 DPRINTF(InOrderCPU
, "sleeping CPU.\n");
535 lastRunningCycle
= curTick();
538 //Tick next_tick = curTick() + cycles(1);
539 //tickEvent.schedule(next_tick);
540 schedule(&tickEvent
, nextCycle(curTick() + 1));
541 DPRINTF(InOrderCPU
, "Scheduled CPU for next tick @ %i.\n",
542 nextCycle(curTick() + 1));
547 updateThreadPriority();
554 if (!deferRegistration
) {
555 registerThreadContexts();
558 // Set inSyscall so that the CPU doesn't squash when initially
559 // setting up registers.
560 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
561 thread
[tid
]->inSyscall
= true;
564 for (ThreadID tid
= 0; tid
< numThreads
; tid
++) {
565 ThreadContext
*src_tc
= threadContexts
[tid
];
566 TheISA::initCPU(src_tc
, src_tc
->contextId());
571 for (ThreadID tid
= 0; tid
< numThreads
; ++tid
)
572 thread
[tid
]->inSyscall
= false;
574 // Call Initializiation Routine for Resource Pool
581 for (int i
= 0; i
< numThreads
; i
++) {
582 isa
[i
].reset(coreType
, numThreads
,
583 1/*numVirtProcs*/, dynamic_cast<BaseCPU
*>(this));
588 InOrderCPU::getPort(const std::string
&if_name
, int idx
)
590 return resPool
->getPort(if_name
, idx
);
595 InOrderCPU::hwrei(ThreadID tid
)
597 panic("hwrei: Unimplemented");
604 InOrderCPU::simPalCheck(int palFunc
, ThreadID tid
)
606 panic("simPalCheck: Unimplemented");
613 InOrderCPU::getInterrupts()
615 // Check if there are any outstanding interrupts
616 return interrupts
->getInterrupt(threadContexts
[0]);
621 InOrderCPU::processInterrupts(Fault interrupt
)
623 // Check for interrupts here. For now can copy the code that
624 // exists within isa_fullsys_traits.hh. Also assume that thread 0
625 // is the one that handles the interrupts.
626 // @todo: Possibly consolidate the interrupt checking code.
627 // @todo: Allow other threads to handle interrupts.
629 assert(interrupt
!= NoFault
);
630 interrupts
->updateIntrInfo(threadContexts
[0]);
632 DPRINTF(InOrderCPU
, "Interrupt %s being handled\n", interrupt
->name());
634 // Note: Context ID ok here? Impl. of FS mode needs to revisit this
635 trap(interrupt
, threadContexts
[0]->contextId(), dummyBufferInst
);
640 InOrderCPU::updateMemPorts()
642 // Update all ThreadContext's memory ports (Functional/Virtual
644 ThreadID size
= thread
.size();
645 for (ThreadID i
= 0; i
< size
; ++i
)
646 thread
[i
]->connectMemPorts(thread
[i
]->getTC());
651 InOrderCPU::trap(Fault fault
, ThreadID tid
, DynInstPtr inst
, int delay
)
653 //@ Squash Pipeline during TRAP
654 scheduleCpuEvent(Trap
, fault
, tid
, inst
, delay
);
658 InOrderCPU::trapCPU(Fault fault
, ThreadID tid
, DynInstPtr inst
)
660 fault
->invoke(tcBase(tid
), inst
->staticInst
);
664 InOrderCPU::squashFromMemStall(DynInstPtr inst
, ThreadID tid
, int delay
)
666 scheduleCpuEvent(SquashFromMemStall
, NoFault
, tid
, inst
, delay
);
671 InOrderCPU::squashDueToMemStall(int stage_num
, InstSeqNum seq_num
,
674 DPRINTF(InOrderCPU
, "Squashing Pipeline Stages Due to Memory Stall...\n");
676 // Squash all instructions in each stage including
677 // instruction that caused the squash (seq_num - 1)
678 // NOTE: The stage bandwidth needs to be cleared so thats why
679 // the stalling instruction is squashed as well. The stalled
680 // instruction is previously placed in another intermediate buffer
681 // while it's stall is being handled.
682 InstSeqNum squash_seq_num
= seq_num
- 1;
684 for (int stNum
=stage_num
; stNum
>= 0 ; stNum
--) {
685 pipelineStage
[stNum
]->squashDueToMemStall(squash_seq_num
, tid
);
690 InOrderCPU::scheduleCpuEvent(CPUEventType c_event
, Fault fault
,
691 ThreadID tid
, DynInstPtr inst
,
692 unsigned delay
, unsigned event_pri_offset
)
694 CPUEvent
*cpu_event
= new CPUEvent(this, c_event
, fault
, tid
, inst
,
697 Tick sked_tick
= nextCycle(curTick() + ticks(delay
));
699 DPRINTF(InOrderCPU
, "Scheduling CPU Event (%s) for cycle %i, [tid:%i].\n",
700 eventNames
[c_event
], curTick() + delay
, tid
);
701 schedule(cpu_event
, sked_tick
);
703 cpu_event
->process();
704 cpuEventRemoveList
.push(cpu_event
);
707 // Broadcast event to the Resource Pool
708 // Need to reset tid just in case this is a dummy instruction
710 resPool
->scheduleEvent(c_event
, inst
, 0, 0, tid
);
714 InOrderCPU::isThreadActive(ThreadID tid
)
716 list
<ThreadID
>::iterator isActive
=
717 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
719 return (isActive
!= activeThreads
.end());
723 InOrderCPU::isThreadReady(ThreadID tid
)
725 list
<ThreadID
>::iterator isReady
=
726 std::find(readyThreads
.begin(), readyThreads
.end(), tid
);
728 return (isReady
!= readyThreads
.end());
732 InOrderCPU::isThreadSuspended(ThreadID tid
)
734 list
<ThreadID
>::iterator isSuspended
=
735 std::find(suspendedThreads
.begin(), suspendedThreads
.end(), tid
);
737 return (isSuspended
!= suspendedThreads
.end());
741 InOrderCPU::activateNextReadyThread()
743 if (readyThreads
.size() >= 1) {
744 ThreadID ready_tid
= readyThreads
.front();
746 // Activate in Pipeline
747 activateThread(ready_tid
);
749 // Activate in Resource Pool
750 resPool
->activateAll(ready_tid
);
752 list
<ThreadID
>::iterator ready_it
=
753 std::find(readyThreads
.begin(), readyThreads
.end(), ready_tid
);
754 readyThreads
.erase(ready_it
);
757 "Attempting to activate new thread, but No Ready Threads to"
760 "Unable to switch to next active thread.\n");
765 InOrderCPU::activateThread(ThreadID tid
)
767 if (isThreadSuspended(tid
)) {
769 "Removing [tid:%i] from suspended threads list.\n", tid
);
771 list
<ThreadID
>::iterator susp_it
=
772 std::find(suspendedThreads
.begin(), suspendedThreads
.end(),
774 suspendedThreads
.erase(susp_it
);
777 if (threadModel
== SwitchOnCacheMiss
&&
778 numActiveThreads() == 1) {
780 "Ignoring activation of [tid:%i], since [tid:%i] is "
781 "already running.\n", tid
, activeThreadId());
783 DPRINTF(InOrderCPU
,"Placing [tid:%i] on ready threads list\n",
786 readyThreads
.push_back(tid
);
788 } else if (!isThreadActive(tid
)) {
790 "Adding [tid:%i] to active threads list.\n", tid
);
791 activeThreads
.push_back(tid
);
793 activateThreadInPipeline(tid
);
795 thread
[tid
]->lastActivate
= curTick();
797 tcBase(tid
)->setStatus(ThreadContext::Active
);
806 InOrderCPU::activateThreadInPipeline(ThreadID tid
)
808 for (int stNum
=0; stNum
< NumStages
; stNum
++) {
809 pipelineStage
[stNum
]->activateThread(tid
);
814 InOrderCPU::deactivateContext(ThreadID tid
, int delay
)
816 DPRINTF(InOrderCPU
,"[tid:%i]: Deactivating ...\n", tid
);
818 scheduleCpuEvent(DeactivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
820 // Be sure to signal that there's some activity so the CPU doesn't
821 // deschedule itself.
822 activityRec
.activity();
828 InOrderCPU::deactivateThread(ThreadID tid
)
830 DPRINTF(InOrderCPU
, "[tid:%i]: Calling deactivate thread.\n", tid
);
832 if (isThreadActive(tid
)) {
833 DPRINTF(InOrderCPU
,"[tid:%i]: Removing from active threads list\n",
835 list
<ThreadID
>::iterator thread_it
=
836 std::find(activeThreads
.begin(), activeThreads
.end(), tid
);
838 removePipelineStalls(*thread_it
);
840 activeThreads
.erase(thread_it
);
842 // Ideally, this should be triggered from the
843 // suspendContext/Thread functions
844 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
847 assert(!isThreadActive(tid
));
851 InOrderCPU::removePipelineStalls(ThreadID tid
)
853 DPRINTF(InOrderCPU
,"[tid:%i]: Removing all pipeline stalls\n",
856 for (int stNum
= 0; stNum
< NumStages
; stNum
++) {
857 pipelineStage
[stNum
]->removeStalls(tid
);
863 InOrderCPU::updateThreadPriority()
865 if (activeThreads
.size() > 1)
867 //DEFAULT TO ROUND ROBIN SCHEME
868 //e.g. Move highest priority to end of thread list
869 list
<ThreadID
>::iterator list_begin
= activeThreads
.begin();
870 list
<ThreadID
>::iterator list_end
= activeThreads
.end();
872 unsigned high_thread
= *list_begin
;
874 activeThreads
.erase(list_begin
);
876 activeThreads
.push_back(high_thread
);
881 InOrderCPU::tickThreadStats()
883 /** Keep track of cycles that each thread is active */
884 list
<ThreadID
>::iterator thread_it
= activeThreads
.begin();
885 while (thread_it
!= activeThreads
.end()) {
886 threadCycles
[*thread_it
]++;
890 // Keep track of cycles where SMT is active
891 if (activeThreads
.size() > 1) {
897 InOrderCPU::activateContext(ThreadID tid
, int delay
)
899 DPRINTF(InOrderCPU
,"[tid:%i]: Activating ...\n", tid
);
902 scheduleCpuEvent(ActivateThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
904 // Be sure to signal that there's some activity so the CPU doesn't
905 // deschedule itself.
906 activityRec
.activity();
912 InOrderCPU::activateNextReadyContext(int delay
)
914 DPRINTF(InOrderCPU
,"Activating next ready thread\n");
916 // NOTE: Add 5 to the event priority so that we always activate
917 // threads after we've finished deactivating, squashing,etc.
919 scheduleCpuEvent(ActivateNextReadyThread
, NoFault
, 0/*tid*/, dummyInst
[0],
922 // Be sure to signal that there's some activity so the CPU doesn't
923 // deschedule itself.
924 activityRec
.activity();
930 InOrderCPU::haltContext(ThreadID tid
, int delay
)
932 DPRINTF(InOrderCPU
, "[tid:%i]: Calling Halt Context...\n", tid
);
934 scheduleCpuEvent(HaltThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
936 activityRec
.activity();
940 InOrderCPU::haltThread(ThreadID tid
)
942 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Halted Threads List...\n", tid
);
943 deactivateThread(tid
);
944 squashThreadInPipeline(tid
);
945 haltedThreads
.push_back(tid
);
947 tcBase(tid
)->setStatus(ThreadContext::Halted
);
949 if (threadModel
== SwitchOnCacheMiss
) {
950 activateNextReadyContext();
955 InOrderCPU::suspendContext(ThreadID tid
, int delay
)
957 scheduleCpuEvent(SuspendThread
, NoFault
, tid
, dummyInst
[tid
], delay
);
961 InOrderCPU::suspendThread(ThreadID tid
)
963 DPRINTF(InOrderCPU
, "[tid:%i]: Placing on Suspended Threads List...\n",
965 deactivateThread(tid
);
966 suspendedThreads
.push_back(tid
);
967 thread
[tid
]->lastSuspend
= curTick();
969 tcBase(tid
)->setStatus(ThreadContext::Suspended
);
973 InOrderCPU::squashThreadInPipeline(ThreadID tid
)
975 //Squash all instructions in each stage
976 for (int stNum
=NumStages
- 1; stNum
>= 0 ; stNum
--) {
977 pipelineStage
[stNum
]->squash(0 /*seq_num*/, tid
);
982 InOrderCPU::getPipeStage(int stage_num
)
984 return pipelineStage
[stage_num
];
988 InOrderCPU::readIntReg(int reg_idx
, ThreadID tid
)
990 return intRegs
[tid
][reg_idx
];
994 InOrderCPU::readFloatReg(int reg_idx
, ThreadID tid
)
996 return floatRegs
.f
[tid
][reg_idx
];
1000 InOrderCPU::readFloatRegBits(int reg_idx
, ThreadID tid
)
1002 return floatRegs
.i
[tid
][reg_idx
];
1006 InOrderCPU::setIntReg(int reg_idx
, uint64_t val
, ThreadID tid
)
1008 intRegs
[tid
][reg_idx
] = val
;
1013 InOrderCPU::setFloatReg(int reg_idx
, FloatReg val
, ThreadID tid
)
1015 floatRegs
.f
[tid
][reg_idx
] = val
;
1020 InOrderCPU::setFloatRegBits(int reg_idx
, FloatRegBits val
, ThreadID tid
)
1022 floatRegs
.i
[tid
][reg_idx
] = val
;
1026 InOrderCPU::readRegOtherThread(unsigned reg_idx
, ThreadID tid
)
1028 // If Default value is set, then retrieve target thread
1029 if (tid
== InvalidThreadID
) {
1030 tid
= TheISA::getTargetThread(tcBase(tid
));
1033 if (reg_idx
< FP_Base_DepTag
) {
1034 // Integer Register File
1035 return readIntReg(reg_idx
, tid
);
1036 } else if (reg_idx
< Ctrl_Base_DepTag
) {
1037 // Float Register File
1038 reg_idx
-= FP_Base_DepTag
;
1039 return readFloatRegBits(reg_idx
, tid
);
1041 reg_idx
-= Ctrl_Base_DepTag
;
1042 return readMiscReg(reg_idx
, tid
); // Misc. Register File
1046 InOrderCPU::setRegOtherThread(unsigned reg_idx
, const MiscReg
&val
,
1049 // If Default value is set, then retrieve target thread
1050 if (tid
== InvalidThreadID
) {
1051 tid
= TheISA::getTargetThread(tcBase(tid
));
1054 if (reg_idx
< FP_Base_DepTag
) { // Integer Register File
1055 setIntReg(reg_idx
, val
, tid
);
1056 } else if (reg_idx
< Ctrl_Base_DepTag
) { // Float Register File
1057 reg_idx
-= FP_Base_DepTag
;
1058 setFloatRegBits(reg_idx
, val
, tid
);
1060 reg_idx
-= Ctrl_Base_DepTag
;
1061 setMiscReg(reg_idx
, val
, tid
); // Misc. Register File
1066 InOrderCPU::readMiscRegNoEffect(int misc_reg
, ThreadID tid
)
1068 return isa
[tid
].readMiscRegNoEffect(misc_reg
);
1072 InOrderCPU::readMiscReg(int misc_reg
, ThreadID tid
)
1074 return isa
[tid
].readMiscReg(misc_reg
, tcBase(tid
));
1078 InOrderCPU::setMiscRegNoEffect(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1080 isa
[tid
].setMiscRegNoEffect(misc_reg
, val
);
1084 InOrderCPU::setMiscReg(int misc_reg
, const MiscReg
&val
, ThreadID tid
)
1086 isa
[tid
].setMiscReg(misc_reg
, val
, tcBase(tid
));
1091 InOrderCPU::addInst(DynInstPtr
&inst
)
1093 ThreadID tid
= inst
->readTid();
1095 instList
[tid
].push_back(inst
);
1097 return --(instList
[tid
].end());
1101 InOrderCPU::updateContextSwitchStats()
1103 // Set Average Stat Here, then reset to 0
1104 instsPerCtxtSwitch
= instsPerSwitch
;
1110 InOrderCPU::instDone(DynInstPtr inst
, ThreadID tid
)
1112 // Set the CPU's PCs - This contributes to the precise state of the CPU
1113 // which can be used when restoring a thread to the CPU after after any
1114 // type of context switching activity (fork, exception, etc.)
1115 pcState(inst
->pcState(), tid
);
1117 if (inst
->isControl()) {
1118 thread
[tid
]->lastGradIsBranch
= true;
1119 thread
[tid
]->lastBranchPC
= inst
->pcState();
1120 TheISA::advancePC(thread
[tid
]->lastBranchPC
, inst
->staticInst
);
1122 thread
[tid
]->lastGradIsBranch
= false;
1126 // Finalize Trace Data For Instruction
1127 if (inst
->traceData
) {
1128 //inst->traceData->setCycle(curTick());
1129 inst
->traceData
->setFetchSeq(inst
->seqNum
);
1130 //inst->traceData->setCPSeq(cpu->tcBase(tid)->numInst);
1131 inst
->traceData
->dump();
1132 delete inst
->traceData
;
1133 inst
->traceData
= NULL
;
1136 // Increment active thread's instruction count
1139 // Increment thread-state's instruction count
1140 thread
[tid
]->numInst
++;
1142 // Increment thread-state's instruction stats
1143 thread
[tid
]->numInsts
++;
1145 // Count committed insts per thread stats
1146 committedInsts
[tid
]++;
1148 // Count total insts committed stat
1149 totalCommittedInsts
++;
1151 // Count SMT-committed insts per thread stat
1152 if (numActiveThreads() > 1) {
1153 smtCommittedInsts
[tid
]++;
1156 // Instruction-Mix Stats
1157 if (inst
->isLoad()) {
1159 } else if (inst
->isStore()) {
1161 } else if (inst
->isControl()) {
1163 } else if (inst
->isNop()) {
1165 } else if (inst
->isNonSpeculative()) {
1167 } else if (inst
->isInteger()) {
1169 } else if (inst
->isFloating()) {
1173 // Check for instruction-count-based events.
1174 comInstEventQueue
[tid
]->serviceEvents(thread
[tid
]->numInst
);
1176 // Broadcast to other resources an instruction
1177 // has been completed
1178 resPool
->scheduleEvent((CPUEventType
)ResourcePool::InstGraduated
, inst
,
1181 // Finally, remove instruction from CPU
1185 // currently unused function, but substitute repetitive code w/this function
1188 InOrderCPU::addToRemoveList(DynInstPtr
&inst
)
1190 removeInstsThisCycle
= true;
1191 if (!inst
->isRemoveList()) {
1192 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1193 "[sn:%lli] to remove list\n",
1194 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1195 inst
->setRemoveList();
1196 removeList
.push(inst
->getInstListIt());
1198 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1199 "[sn:%lli], already remove list\n",
1200 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1206 InOrderCPU::removeInst(DynInstPtr
&inst
)
1208 DPRINTF(InOrderCPU
, "Removing graduated instruction [tid:%i] PC %s "
1210 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1212 removeInstsThisCycle
= true;
1214 // Remove the instruction.
1215 if (!inst
->isRemoveList()) {
1216 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1217 "[sn:%lli] to remove list\n",
1218 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1219 inst
->setRemoveList();
1220 removeList
.push(inst
->getInstListIt());
1222 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i] PC %s "
1223 "[sn:%lli], already on remove list\n",
1224 inst
->threadNumber
, inst
->pcState(), inst
->seqNum
);
1230 InOrderCPU::removeInstsUntil(const InstSeqNum
&seq_num
, ThreadID tid
)
1232 //assert(!instList[tid].empty());
1234 removeInstsThisCycle
= true;
1236 ListIt inst_iter
= instList
[tid
].end();
1240 DPRINTF(InOrderCPU
, "Squashing instructions from CPU instruction "
1241 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1242 tid
, seq_num
, (*inst_iter
)->seqNum
);
1244 while ((*inst_iter
)->seqNum
> seq_num
) {
1246 bool break_loop
= (inst_iter
== instList
[tid
].begin());
1248 squashInstIt(inst_iter
, tid
);
1259 InOrderCPU::squashInstIt(const ListIt
&instIt
, ThreadID tid
)
1261 if ((*instIt
)->threadNumber
== tid
) {
1262 DPRINTF(InOrderCPU
, "Squashing instruction, "
1263 "[tid:%i] [sn:%lli] PC %s\n",
1264 (*instIt
)->threadNumber
,
1266 (*instIt
)->pcState());
1268 (*instIt
)->setSquashed();
1270 if (!(*instIt
)->isRemoveList()) {
1271 DPRINTF(InOrderCPU
, "Pushing instruction [tid:%i] PC %s "
1272 "[sn:%lli] to remove list\n",
1273 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1275 (*instIt
)->setRemoveList();
1276 removeList
.push(instIt
);
1278 DPRINTF(InOrderCPU
, "Ignoring instruction removal for [tid:%i]"
1279 " PC %s [sn:%lli], already on remove list\n",
1280 (*instIt
)->threadNumber
, (*instIt
)->pcState(),
1290 InOrderCPU::cleanUpRemovedInsts()
1292 while (!removeList
.empty()) {
1293 DPRINTF(InOrderCPU
, "Removing instruction, "
1294 "[tid:%i] [sn:%lli] PC %s\n",
1295 (*removeList
.front())->threadNumber
,
1296 (*removeList
.front())->seqNum
,
1297 (*removeList
.front())->pcState());
1299 DynInstPtr inst
= *removeList
.front();
1300 ThreadID tid
= inst
->threadNumber
;
1302 // Make Sure Resource Schedule Is Emptied Out
1303 ThePipeline::ResSchedule
*inst_sched
= &inst
->resSched
;
1304 while (!inst_sched
->empty()) {
1305 ScheduleEntry
* sch_entry
= inst_sched
->top();
1310 // Remove From Register Dependency Map, If Necessary
1311 archRegDepMap
[(*removeList
.front())->threadNumber
].
1312 remove((*removeList
.front()));
1315 // Clear if Non-Speculative
1316 if (inst
->staticInst
&&
1317 inst
->seqNum
== nonSpecSeqNum
[tid
] &&
1318 nonSpecInstActive
[tid
] == true) {
1319 nonSpecInstActive
[tid
] = false;
1322 instList
[tid
].erase(removeList
.front());
1327 removeInstsThisCycle
= false;
1331 InOrderCPU::cleanUpRemovedReqs()
1333 while (!reqRemoveList
.empty()) {
1334 ResourceRequest
*res_req
= reqRemoveList
.front();
1336 DPRINTF(RefCount
, "[tid:%i] [sn:%lli]: Removing Request "
1337 "[stage_num:%i] [res:%s] [slot:%i] [completed:%i].\n",
1338 res_req
->inst
->threadNumber
,
1339 res_req
->inst
->seqNum
,
1340 res_req
->getStageNum(),
1341 res_req
->res
->name(),
1342 (res_req
->isCompleted()) ?
1343 res_req
->getComplSlot() : res_req
->getSlot(),
1344 res_req
->isCompleted());
1346 reqRemoveList
.pop();
1353 InOrderCPU::cleanUpRemovedEvents()
1355 while (!cpuEventRemoveList
.empty()) {
1356 Event
*cpu_event
= cpuEventRemoveList
.front();
1357 cpuEventRemoveList
.pop();
1364 InOrderCPU::dumpInsts()
1368 ListIt inst_list_it
= instList
[0].begin();
1370 cprintf("Dumping Instruction List\n");
1372 while (inst_list_it
!= instList
[0].end()) {
1373 cprintf("Instruction:%i\nPC:%s\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1375 num
, (*inst_list_it
)->pcState(),
1376 (*inst_list_it
)->threadNumber
,
1377 (*inst_list_it
)->seqNum
, (*inst_list_it
)->isIssued(),
1378 (*inst_list_it
)->isSquashed());
1385 InOrderCPU::wakeCPU()
1387 if (/*activityRec.active() || */tickEvent
.scheduled()) {
1388 DPRINTF(Activity
, "CPU already running.\n");
1392 DPRINTF(Activity
, "Waking up CPU\n");
1394 Tick extra_cycles
= tickToCycles((curTick() - 1) - lastRunningCycle
);
1396 idleCycles
+= extra_cycles
;
1397 for (int stage_num
= 0; stage_num
< NumStages
; stage_num
++) {
1398 pipelineStage
[stage_num
]->idleCycles
+= extra_cycles
;
1401 numCycles
+= extra_cycles
;
1403 schedule(&tickEvent
, nextCycle(curTick()));
1409 InOrderCPU::wakeup()
1411 if (thread
[0]->status() != ThreadContext::Suspended
)
1416 DPRINTF(Quiesce
, "Suspended Processor woken\n");
1417 threadContexts
[0]->activate();
1423 InOrderCPU::syscall(int64_t callnum
, ThreadID tid
)
1425 DPRINTF(InOrderCPU
, "[tid:%i] Executing syscall().\n\n", tid
);
1427 DPRINTF(Activity
,"Activity: syscall() called.\n");
1429 // Temporarily increase this by one to account for the syscall
1431 ++(this->thread
[tid
]->funcExeInst
);
1433 // Execute the actual syscall.
1434 this->thread
[tid
]->syscall(callnum
);
1436 // Decrease funcExeInst by one as the normal commit will handle
1438 --(this->thread
[tid
]->funcExeInst
);
1440 // Clear Non-Speculative Block Variable
1441 nonSpecInstActive
[tid
] = false;
1446 InOrderCPU::getITBPtr()
1448 CacheUnit
*itb_res
=
1449 dynamic_cast<CacheUnit
*>(resPool
->getResource(fetchPortIdx
));
1450 return itb_res
->tlb();
1455 InOrderCPU::getDTBPtr()
1457 CacheUnit
*dtb_res
=
1458 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1459 return dtb_res
->tlb();
1463 InOrderCPU::read(DynInstPtr inst
, Addr addr
,
1464 uint8_t *data
, unsigned size
, unsigned flags
)
1466 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1467 // you want to run w/out caches?
1468 CacheUnit
*cache_res
=
1469 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1471 return cache_res
->read(inst
, addr
, data
, size
, flags
);
1475 InOrderCPU::write(DynInstPtr inst
, uint8_t *data
, unsigned size
,
1476 Addr addr
, unsigned flags
, uint64_t *write_res
)
1478 //@TODO: Generalize name "CacheUnit" to "MemUnit" just in case
1479 // you want to run w/out caches?
1480 CacheUnit
*cache_res
=
1481 dynamic_cast<CacheUnit
*>(resPool
->getResource(dataPortIdx
));
1482 return cache_res
->write(inst
, data
, size
, addr
, flags
, write_res
);