cpu.cc (8921:e53972f72165) cpu.cc (8948:e95ee70f876c)
1/*
1/*
2 * Copyright (c) 2011 ARM Limited
2 * Copyright (c) 2011-2012 ARM Limited
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2004-2006 The Regents of The University of Michigan
15 * Copyright (c) 2011 Regents of the University of California
16 * All rights reserved.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 *
41 * Authors: Kevin Lim
42 * Korey Sewell
43 * Rick Strong
44 */
45
46#include "arch/kernel_stats.hh"
47#include "config/the_isa.hh"
48#include "cpu/checker/cpu.hh"
49#include "cpu/checker/thread_context.hh"
50#include "cpu/o3/cpu.hh"
51#include "cpu/o3/isa_specific.hh"
52#include "cpu/o3/thread_context.hh"
53#include "cpu/activity.hh"
54#include "cpu/quiesce_event.hh"
55#include "cpu/simple_thread.hh"
56#include "cpu/thread_context.hh"
57#include "debug/Activity.hh"
58#include "debug/O3CPU.hh"
59#include "debug/Quiesce.hh"
60#include "enums/MemoryMode.hh"
61#include "sim/core.hh"
62#include "sim/full_system.hh"
63#include "sim/process.hh"
64#include "sim/stat_control.hh"
65#include "sim/system.hh"
66
67#if THE_ISA == ALPHA_ISA
68#include "arch/alpha/osfpal.hh"
69#include "debug/Activity.hh"
70#endif
71
72struct BaseCPUParams;
73
74using namespace TheISA;
75using namespace std;
76
77BaseO3CPU::BaseO3CPU(BaseCPUParams *params)
78 : BaseCPU(params)
79{
80}
81
82void
83BaseO3CPU::regStats()
84{
85 BaseCPU::regStats();
86}
87
88template<class Impl>
89bool
90FullO3CPU<Impl>::IcachePort::recvTiming(PacketPtr pkt)
91{
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2004-2006 The Regents of The University of Michigan
15 * Copyright (c) 2011 Regents of the University of California
16 * All rights reserved.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 *
41 * Authors: Kevin Lim
42 * Korey Sewell
43 * Rick Strong
44 */
45
46#include "arch/kernel_stats.hh"
47#include "config/the_isa.hh"
48#include "cpu/checker/cpu.hh"
49#include "cpu/checker/thread_context.hh"
50#include "cpu/o3/cpu.hh"
51#include "cpu/o3/isa_specific.hh"
52#include "cpu/o3/thread_context.hh"
53#include "cpu/activity.hh"
54#include "cpu/quiesce_event.hh"
55#include "cpu/simple_thread.hh"
56#include "cpu/thread_context.hh"
57#include "debug/Activity.hh"
58#include "debug/O3CPU.hh"
59#include "debug/Quiesce.hh"
60#include "enums/MemoryMode.hh"
61#include "sim/core.hh"
62#include "sim/full_system.hh"
63#include "sim/process.hh"
64#include "sim/stat_control.hh"
65#include "sim/system.hh"
66
67#if THE_ISA == ALPHA_ISA
68#include "arch/alpha/osfpal.hh"
69#include "debug/Activity.hh"
70#endif
71
72struct BaseCPUParams;
73
74using namespace TheISA;
75using namespace std;
76
77BaseO3CPU::BaseO3CPU(BaseCPUParams *params)
78 : BaseCPU(params)
79{
80}
81
82void
83BaseO3CPU::regStats()
84{
85 BaseCPU::regStats();
86}
87
88template<class Impl>
89bool
90FullO3CPU<Impl>::IcachePort::recvTiming(PacketPtr pkt)
91{
92 assert(pkt->isResponse());
92 DPRINTF(O3CPU, "Fetch unit received timing\n");
93 DPRINTF(O3CPU, "Fetch unit received timing\n");
93 if (pkt->isResponse()) {
94 // We shouldn't ever get a block in ownership state
95 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted()));
94 // We shouldn't ever get a block in ownership state
95 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted()));
96 fetch->processCacheCompletion(pkt);
96
97
97 fetch->processCacheCompletion(pkt);
98 }
99 //else Snooped a coherence request, just return
100 return true;
101}
102
103template<class Impl>
104void
105FullO3CPU<Impl>::IcachePort::recvRetry()
106{
107 fetch->recvRetry();
108}
109
110template <class Impl>
111bool
112FullO3CPU<Impl>::DcachePort::recvTiming(PacketPtr pkt)
113{
98 return true;
99}
100
101template<class Impl>
102void
103FullO3CPU<Impl>::IcachePort::recvRetry()
104{
105 fetch->recvRetry();
106}
107
108template <class Impl>
109bool
110FullO3CPU<Impl>::DcachePort::recvTiming(PacketPtr pkt)
111{
112 assert(pkt->isResponse());
114 return lsq->recvTiming(pkt);
115}
116
117template <class Impl>
113 return lsq->recvTiming(pkt);
114}
115
116template <class Impl>
117bool
118FullO3CPU<Impl>::DcachePort::recvTimingSnoop(PacketPtr pkt)
119{
120 assert(pkt->isRequest());
121 return lsq->recvTimingSnoop(pkt);
122}
123
124template <class Impl>
118void
119FullO3CPU<Impl>::DcachePort::recvRetry()
120{
121 lsq->recvRetry();
122}
123
124template <class Impl>
125FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c)
126 : Event(CPU_Tick_Pri), cpu(c)
127{
128}
129
130template <class Impl>
131void
132FullO3CPU<Impl>::TickEvent::process()
133{
134 cpu->tick();
135}
136
137template <class Impl>
138const char *
139FullO3CPU<Impl>::TickEvent::description() const
140{
141 return "FullO3CPU tick";
142}
143
144template <class Impl>
145FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent()
146 : Event(CPU_Switch_Pri)
147{
148}
149
150template <class Impl>
151void
152FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num,
153 FullO3CPU<Impl> *thread_cpu)
154{
155 tid = thread_num;
156 cpu = thread_cpu;
157}
158
159template <class Impl>
160void
161FullO3CPU<Impl>::ActivateThreadEvent::process()
162{
163 cpu->activateThread(tid);
164}
165
166template <class Impl>
167const char *
168FullO3CPU<Impl>::ActivateThreadEvent::description() const
169{
170 return "FullO3CPU \"Activate Thread\"";
171}
172
173template <class Impl>
174FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent()
175 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL)
176{
177}
178
179template <class Impl>
180void
181FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num,
182 FullO3CPU<Impl> *thread_cpu)
183{
184 tid = thread_num;
185 cpu = thread_cpu;
186 remove = false;
187}
188
189template <class Impl>
190void
191FullO3CPU<Impl>::DeallocateContextEvent::process()
192{
193 cpu->deactivateThread(tid);
194 if (remove)
195 cpu->removeThread(tid);
196}
197
198template <class Impl>
199const char *
200FullO3CPU<Impl>::DeallocateContextEvent::description() const
201{
202 return "FullO3CPU \"Deallocate Context\"";
203}
204
205template <class Impl>
206FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params)
207 : BaseO3CPU(params),
208 itb(params->itb),
209 dtb(params->dtb),
210 tickEvent(this),
211#ifndef NDEBUG
212 instcount(0),
213#endif
214 removeInstsThisCycle(false),
215 fetch(this, params),
216 decode(this, params),
217 rename(this, params),
218 iew(this, params),
219 commit(this, params),
220
221 regFile(this, params->numPhysIntRegs,
222 params->numPhysFloatRegs),
223
224 freeList(params->numThreads,
225 TheISA::NumIntRegs, params->numPhysIntRegs,
226 TheISA::NumFloatRegs, params->numPhysFloatRegs),
227
228 rob(this,
229 params->numROBEntries, params->squashWidth,
230 params->smtROBPolicy, params->smtROBThreshold,
231 params->numThreads),
232
233 scoreboard(params->numThreads,
234 TheISA::NumIntRegs, params->numPhysIntRegs,
235 TheISA::NumFloatRegs, params->numPhysFloatRegs,
236 TheISA::NumMiscRegs * numThreads,
237 TheISA::ZeroReg),
238
239 icachePort(&fetch, this),
240 dcachePort(&iew.ldstQueue, this),
241
242 timeBuffer(params->backComSize, params->forwardComSize),
243 fetchQueue(params->backComSize, params->forwardComSize),
244 decodeQueue(params->backComSize, params->forwardComSize),
245 renameQueue(params->backComSize, params->forwardComSize),
246 iewQueue(params->backComSize, params->forwardComSize),
247 activityRec(name(), NumStages,
248 params->backComSize + params->forwardComSize,
249 params->activity),
250
251 globalSeqNum(1),
252 system(params->system),
253 drainCount(0),
254 deferRegistration(params->defer_registration)
255{
256 if (!deferRegistration) {
257 _status = Running;
258 } else {
259 _status = Idle;
260 }
261
262 if (params->checker) {
263 BaseCPU *temp_checker = params->checker;
264 checker = dynamic_cast<Checker<Impl> *>(temp_checker);
265 checker->setIcachePort(&icachePort);
266 checker->setSystem(params->system);
267 } else {
268 checker = NULL;
269 }
270
271 if (!FullSystem) {
272 thread.resize(numThreads);
273 tids.resize(numThreads);
274 }
275
276 // The stages also need their CPU pointer setup. However this
277 // must be done at the upper level CPU because they have pointers
278 // to the upper level CPU, and not this FullO3CPU.
279
280 // Set up Pointers to the activeThreads list for each stage
281 fetch.setActiveThreads(&activeThreads);
282 decode.setActiveThreads(&activeThreads);
283 rename.setActiveThreads(&activeThreads);
284 iew.setActiveThreads(&activeThreads);
285 commit.setActiveThreads(&activeThreads);
286
287 // Give each of the stages the time buffer they will use.
288 fetch.setTimeBuffer(&timeBuffer);
289 decode.setTimeBuffer(&timeBuffer);
290 rename.setTimeBuffer(&timeBuffer);
291 iew.setTimeBuffer(&timeBuffer);
292 commit.setTimeBuffer(&timeBuffer);
293
294 // Also setup each of the stages' queues.
295 fetch.setFetchQueue(&fetchQueue);
296 decode.setFetchQueue(&fetchQueue);
297 commit.setFetchQueue(&fetchQueue);
298 decode.setDecodeQueue(&decodeQueue);
299 rename.setDecodeQueue(&decodeQueue);
300 rename.setRenameQueue(&renameQueue);
301 iew.setRenameQueue(&renameQueue);
302 iew.setIEWQueue(&iewQueue);
303 commit.setIEWQueue(&iewQueue);
304 commit.setRenameQueue(&renameQueue);
305
306 commit.setIEWStage(&iew);
307 rename.setIEWStage(&iew);
308 rename.setCommitStage(&commit);
309
310 ThreadID active_threads;
311 if (FullSystem) {
312 active_threads = 1;
313 } else {
314 active_threads = params->workload.size();
315
316 if (active_threads > Impl::MaxThreads) {
317 panic("Workload Size too large. Increase the 'MaxThreads' "
318 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
319 "or edit your workload size.");
320 }
321 }
322
323 //Make Sure That this a Valid Architeture
324 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
325 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
326
327 rename.setScoreboard(&scoreboard);
328 iew.setScoreboard(&scoreboard);
329
330 // Setup the rename map for whichever stages need it.
331 PhysRegIndex lreg_idx = 0;
332 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
333
334 for (ThreadID tid = 0; tid < numThreads; tid++) {
335 bool bindRegs = (tid <= active_threads - 1);
336
337 commitRenameMap[tid].init(TheISA::NumIntRegs,
338 params->numPhysIntRegs,
339 lreg_idx, //Index for Logical. Regs
340
341 TheISA::NumFloatRegs,
342 params->numPhysFloatRegs,
343 freg_idx, //Index for Float Regs
344
345 TheISA::NumMiscRegs,
346
347 TheISA::ZeroReg,
348 TheISA::ZeroReg,
349
350 tid,
351 false);
352
353 renameMap[tid].init(TheISA::NumIntRegs,
354 params->numPhysIntRegs,
355 lreg_idx, //Index for Logical. Regs
356
357 TheISA::NumFloatRegs,
358 params->numPhysFloatRegs,
359 freg_idx, //Index for Float Regs
360
361 TheISA::NumMiscRegs,
362
363 TheISA::ZeroReg,
364 TheISA::ZeroReg,
365
366 tid,
367 bindRegs);
368
369 activateThreadEvent[tid].init(tid, this);
370 deallocateContextEvent[tid].init(tid, this);
371 }
372
373 rename.setRenameMap(renameMap);
374 commit.setRenameMap(commitRenameMap);
375
376 // Give renameMap & rename stage access to the freeList;
377 for (ThreadID tid = 0; tid < numThreads; tid++)
378 renameMap[tid].setFreeList(&freeList);
379 rename.setFreeList(&freeList);
380
381 // Setup the ROB for whichever stages need it.
382 commit.setROB(&rob);
383
384 lastRunningCycle = curTick();
385
386 lastActivatedCycle = -1;
387#if 0
388 // Give renameMap & rename stage access to the freeList;
389 for (ThreadID tid = 0; tid < numThreads; tid++)
390 globalSeqNum[tid] = 1;
391#endif
392
393 contextSwitch = false;
394 DPRINTF(O3CPU, "Creating O3CPU object.\n");
395
396 // Setup any thread state.
397 this->thread.resize(this->numThreads);
398
399 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
400 if (FullSystem) {
401 // SMT is not supported in FS mode yet.
402 assert(this->numThreads == 1);
403 this->thread[tid] = new Thread(this, 0, NULL);
404 } else {
405 if (tid < params->workload.size()) {
406 DPRINTF(O3CPU, "Workload[%i] process is %#x",
407 tid, this->thread[tid]);
408 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
409 (typename Impl::O3CPU *)(this),
410 tid, params->workload[tid]);
411
412 //usedTids[tid] = true;
413 //threadMap[tid] = tid;
414 } else {
415 //Allocate Empty thread so M5 can use later
416 //when scheduling threads to CPU
417 Process* dummy_proc = NULL;
418
419 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
420 (typename Impl::O3CPU *)(this),
421 tid, dummy_proc);
422 //usedTids[tid] = false;
423 }
424 }
425
426 ThreadContext *tc;
427
428 // Setup the TC that will serve as the interface to the threads/CPU.
429 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
430
431 tc = o3_tc;
432
433 // If we're using a checker, then the TC should be the
434 // CheckerThreadContext.
435 if (params->checker) {
436 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
437 o3_tc, this->checker);
438 }
439
440 o3_tc->cpu = (typename Impl::O3CPU *)(this);
441 assert(o3_tc->cpu);
442 o3_tc->thread = this->thread[tid];
443
444 if (FullSystem) {
445 // Setup quiesce event.
446 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
447 }
448 // Give the thread the TC.
449 this->thread[tid]->tc = tc;
450
451 // Add the TC to the CPU's list of TC's.
452 this->threadContexts.push_back(tc);
453 }
454
455 // FullO3CPU always requires an interrupt controller.
456 if (!params->defer_registration && !interrupts) {
457 fatal("FullO3CPU %s has no interrupt controller.\n"
458 "Ensure createInterruptController() is called.\n", name());
459 }
460
461 for (ThreadID tid = 0; tid < this->numThreads; tid++)
462 this->thread[tid]->setFuncExeInst(0);
463
464 lockAddr = 0;
465 lockFlag = false;
466}
467
468template <class Impl>
469FullO3CPU<Impl>::~FullO3CPU()
470{
471}
472
473template <class Impl>
474void
475FullO3CPU<Impl>::regStats()
476{
477 BaseO3CPU::regStats();
478
479 // Register any of the O3CPU's stats here.
480 timesIdled
481 .name(name() + ".timesIdled")
482 .desc("Number of times that the entire CPU went into an idle state and"
483 " unscheduled itself")
484 .prereq(timesIdled);
485
486 idleCycles
487 .name(name() + ".idleCycles")
488 .desc("Total number of cycles that the CPU has spent unscheduled due "
489 "to idling")
490 .prereq(idleCycles);
491
492 quiesceCycles
493 .name(name() + ".quiesceCycles")
494 .desc("Total number of cycles that CPU has spent quiesced or waiting "
495 "for an interrupt")
496 .prereq(quiesceCycles);
497
498 // Number of Instructions simulated
499 // --------------------------------
500 // Should probably be in Base CPU but need templated
501 // MaxThreads so put in here instead
502 committedInsts
503 .init(numThreads)
504 .name(name() + ".committedInsts")
505 .desc("Number of Instructions Simulated");
506
507 committedOps
508 .init(numThreads)
509 .name(name() + ".committedOps")
510 .desc("Number of Ops (including micro ops) Simulated");
511
512 totalCommittedInsts
513 .name(name() + ".committedInsts_total")
514 .desc("Number of Instructions Simulated");
515
516 cpi
517 .name(name() + ".cpi")
518 .desc("CPI: Cycles Per Instruction")
519 .precision(6);
520 cpi = numCycles / committedInsts;
521
522 totalCpi
523 .name(name() + ".cpi_total")
524 .desc("CPI: Total CPI of All Threads")
525 .precision(6);
526 totalCpi = numCycles / totalCommittedInsts;
527
528 ipc
529 .name(name() + ".ipc")
530 .desc("IPC: Instructions Per Cycle")
531 .precision(6);
532 ipc = committedInsts / numCycles;
533
534 totalIpc
535 .name(name() + ".ipc_total")
536 .desc("IPC: Total IPC of All Threads")
537 .precision(6);
538 totalIpc = totalCommittedInsts / numCycles;
539
540 this->fetch.regStats();
541 this->decode.regStats();
542 this->rename.regStats();
543 this->iew.regStats();
544 this->commit.regStats();
545 this->rob.regStats();
546
547 intRegfileReads
548 .name(name() + ".int_regfile_reads")
549 .desc("number of integer regfile reads")
550 .prereq(intRegfileReads);
551
552 intRegfileWrites
553 .name(name() + ".int_regfile_writes")
554 .desc("number of integer regfile writes")
555 .prereq(intRegfileWrites);
556
557 fpRegfileReads
558 .name(name() + ".fp_regfile_reads")
559 .desc("number of floating regfile reads")
560 .prereq(fpRegfileReads);
561
562 fpRegfileWrites
563 .name(name() + ".fp_regfile_writes")
564 .desc("number of floating regfile writes")
565 .prereq(fpRegfileWrites);
566
567 miscRegfileReads
568 .name(name() + ".misc_regfile_reads")
569 .desc("number of misc regfile reads")
570 .prereq(miscRegfileReads);
571
572 miscRegfileWrites
573 .name(name() + ".misc_regfile_writes")
574 .desc("number of misc regfile writes")
575 .prereq(miscRegfileWrites);
576}
577
578template <class Impl>
579void
580FullO3CPU<Impl>::tick()
581{
582 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
583
584 ++numCycles;
585
586// activity = false;
587
588 //Tick each of the stages
589 fetch.tick();
590
591 decode.tick();
592
593 rename.tick();
594
595 iew.tick();
596
597 commit.tick();
598
599 if (!FullSystem)
600 doContextSwitch();
601
602 // Now advance the time buffers
603 timeBuffer.advance();
604
605 fetchQueue.advance();
606 decodeQueue.advance();
607 renameQueue.advance();
608 iewQueue.advance();
609
610 activityRec.advance();
611
612 if (removeInstsThisCycle) {
613 cleanUpRemovedInsts();
614 }
615
616 if (!tickEvent.scheduled()) {
617 if (_status == SwitchedOut ||
618 getState() == SimObject::Drained) {
619 DPRINTF(O3CPU, "Switched out!\n");
620 // increment stat
621 lastRunningCycle = curTick();
622 } else if (!activityRec.active() || _status == Idle) {
623 DPRINTF(O3CPU, "Idle!\n");
624 lastRunningCycle = curTick();
625 timesIdled++;
626 } else {
627 schedule(tickEvent, nextCycle(curTick() + ticks(1)));
628 DPRINTF(O3CPU, "Scheduling next tick!\n");
629 }
630 }
631
632 if (!FullSystem)
633 updateThreadPriority();
634}
635
636template <class Impl>
637void
638FullO3CPU<Impl>::init()
639{
640 BaseCPU::init();
641
642 for (ThreadID tid = 0; tid < numThreads; ++tid) {
643 // Set inSyscall so that the CPU doesn't squash when initially
644 // setting up registers.
645 thread[tid]->inSyscall = true;
646 // Initialise the ThreadContext's memory proxies
647 thread[tid]->initMemProxies(thread[tid]->getTC());
648 }
649
650 // this CPU could still be unconnected if we are restoring from a
651 // checkpoint and this CPU is to be switched in, thus we can only
652 // do this here if the instruction port is actually connected, if
653 // not we have to do it as part of takeOverFrom
654 if (icachePort.isConnected())
655 fetch.setIcache();
656
657 if (FullSystem && !params()->defer_registration) {
658 for (ThreadID tid = 0; tid < numThreads; tid++) {
659 ThreadContext *src_tc = threadContexts[tid];
660 TheISA::initCPU(src_tc, src_tc->contextId());
661 }
662 }
663
664 // Clear inSyscall.
665 for (int tid = 0; tid < numThreads; ++tid)
666 thread[tid]->inSyscall = false;
667
668 // Initialize stages.
669 fetch.initStage();
670 iew.initStage();
671 rename.initStage();
672 commit.initStage();
673
674 commit.setThreads(thread);
675}
676
677template <class Impl>
678void
679FullO3CPU<Impl>::activateThread(ThreadID tid)
680{
681 list<ThreadID>::iterator isActive =
682 std::find(activeThreads.begin(), activeThreads.end(), tid);
683
684 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid);
685
686 if (isActive == activeThreads.end()) {
687 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
688 tid);
689
690 activeThreads.push_back(tid);
691 }
692}
693
694template <class Impl>
695void
696FullO3CPU<Impl>::deactivateThread(ThreadID tid)
697{
698 //Remove From Active List, if Active
699 list<ThreadID>::iterator thread_it =
700 std::find(activeThreads.begin(), activeThreads.end(), tid);
701
702 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
703
704 if (thread_it != activeThreads.end()) {
705 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
706 tid);
707 activeThreads.erase(thread_it);
708 }
709}
710
711template <class Impl>
712Counter
713FullO3CPU<Impl>::totalInsts() const
714{
715 Counter total(0);
716
717 ThreadID size = thread.size();
718 for (ThreadID i = 0; i < size; i++)
719 total += thread[i]->numInst;
720
721 return total;
722}
723
724template <class Impl>
725Counter
726FullO3CPU<Impl>::totalOps() const
727{
728 Counter total(0);
729
730 ThreadID size = thread.size();
731 for (ThreadID i = 0; i < size; i++)
732 total += thread[i]->numOp;
733
734 return total;
735}
736
737template <class Impl>
738void
739FullO3CPU<Impl>::activateContext(ThreadID tid, int delay)
740{
741 // Needs to set each stage to running as well.
742 if (delay){
743 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
744 "on cycle %d\n", tid, curTick() + ticks(delay));
745 scheduleActivateThreadEvent(tid, delay);
746 } else {
747 activateThread(tid);
748 }
749
750 if (lastActivatedCycle < curTick()) {
751 scheduleTickEvent(delay);
752
753 // Be sure to signal that there's some activity so the CPU doesn't
754 // deschedule itself.
755 activityRec.activity();
756 fetch.wakeFromQuiesce();
757
758 quiesceCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
759
760 lastActivatedCycle = curTick();
761
762 _status = Running;
763 }
764}
765
766template <class Impl>
767bool
768FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove,
769 int delay)
770{
771 // Schedule removal of thread data from CPU
772 if (delay){
773 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
774 "on cycle %d\n", tid, curTick() + ticks(delay));
775 scheduleDeallocateContextEvent(tid, remove, delay);
776 return false;
777 } else {
778 deactivateThread(tid);
779 if (remove)
780 removeThread(tid);
781 return true;
782 }
783}
784
785template <class Impl>
786void
787FullO3CPU<Impl>::suspendContext(ThreadID tid)
788{
789 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
790 bool deallocated = scheduleDeallocateContext(tid, false, 1);
791 // If this was the last thread then unschedule the tick event.
792 if ((activeThreads.size() == 1 && !deallocated) ||
793 activeThreads.size() == 0)
794 unscheduleTickEvent();
795
796 DPRINTF(Quiesce, "Suspending Context\n");
797 lastRunningCycle = curTick();
798 _status = Idle;
799}
800
801template <class Impl>
802void
803FullO3CPU<Impl>::haltContext(ThreadID tid)
804{
805 //For now, this is the same as deallocate
806 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
807 scheduleDeallocateContext(tid, true, 1);
808}
809
810template <class Impl>
811void
812FullO3CPU<Impl>::insertThread(ThreadID tid)
813{
814 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
815 // Will change now that the PC and thread state is internal to the CPU
816 // and not in the ThreadContext.
817 ThreadContext *src_tc;
818 if (FullSystem)
819 src_tc = system->threadContexts[tid];
820 else
821 src_tc = tcBase(tid);
822
823 //Bind Int Regs to Rename Map
824 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
825 PhysRegIndex phys_reg = freeList.getIntReg();
826
827 renameMap[tid].setEntry(ireg,phys_reg);
828 scoreboard.setReg(phys_reg);
829 }
830
831 //Bind Float Regs to Rename Map
832 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
833 PhysRegIndex phys_reg = freeList.getFloatReg();
834
835 renameMap[tid].setEntry(freg,phys_reg);
836 scoreboard.setReg(phys_reg);
837 }
838
839 //Copy Thread Data Into RegFile
840 //this->copyFromTC(tid);
841
842 //Set PC/NPC/NNPC
843 pcState(src_tc->pcState(), tid);
844
845 src_tc->setStatus(ThreadContext::Active);
846
847 activateContext(tid,1);
848
849 //Reset ROB/IQ/LSQ Entries
850 commit.rob->resetEntries();
851 iew.resetEntries();
852}
853
854template <class Impl>
855void
856FullO3CPU<Impl>::removeThread(ThreadID tid)
857{
858 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
859
860 // Copy Thread Data From RegFile
861 // If thread is suspended, it might be re-allocated
862 // this->copyToTC(tid);
863
864
865 // @todo: 2-27-2008: Fix how we free up rename mappings
866 // here to alleviate the case for double-freeing registers
867 // in SMT workloads.
868
869 // Unbind Int Regs from Rename Map
870 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
871 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
872
873 scoreboard.unsetReg(phys_reg);
874 freeList.addReg(phys_reg);
875 }
876
877 // Unbind Float Regs from Rename Map
878 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
879 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
880
881 scoreboard.unsetReg(phys_reg);
882 freeList.addReg(phys_reg);
883 }
884
885 // Squash Throughout Pipeline
886 DynInstPtr inst = commit.rob->readHeadInst(tid);
887 InstSeqNum squash_seq_num = inst->seqNum;
888 fetch.squash(0, squash_seq_num, inst, tid);
889 decode.squash(tid);
890 rename.squash(squash_seq_num, tid);
891 iew.squash(tid);
892 iew.ldstQueue.squash(squash_seq_num, tid);
893 commit.rob->squash(squash_seq_num, tid);
894
895
896 assert(iew.instQueue.getCount(tid) == 0);
897 assert(iew.ldstQueue.getCount(tid) == 0);
898
899 // Reset ROB/IQ/LSQ Entries
900
901 // Commented out for now. This should be possible to do by
902 // telling all the pipeline stages to drain first, and then
903 // checking until the drain completes. Once the pipeline is
904 // drained, call resetEntries(). - 10-09-06 ktlim
905/*
906 if (activeThreads.size() >= 1) {
907 commit.rob->resetEntries();
908 iew.resetEntries();
909 }
910*/
911}
912
913
914template <class Impl>
915void
916FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
917{
918 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
919 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
920 tid);
921
922 bool ready = true;
923
924 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
925 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
926 "Phys. Int. Regs.\n",
927 tid);
928 ready = false;
929 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
930 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
931 "Phys. Float. Regs.\n",
932 tid);
933 ready = false;
934 } else if (commit.rob->numFreeEntries() >=
935 commit.rob->entryAmount(activeThreads.size() + 1)) {
936 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
937 "ROB entries.\n",
938 tid);
939 ready = false;
940 } else if (iew.instQueue.numFreeEntries() >=
941 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
942 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
943 "IQ entries.\n",
944 tid);
945 ready = false;
946 } else if (iew.ldstQueue.numFreeEntries() >=
947 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
948 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
949 "LSQ entries.\n",
950 tid);
951 ready = false;
952 }
953
954 if (ready) {
955 insertThread(tid);
956
957 contextSwitch = false;
958
959 cpuWaitList.remove(tid);
960 } else {
961 suspendContext(tid);
962
963 //blocks fetch
964 contextSwitch = true;
965
966 //@todo: dont always add to waitlist
967 //do waitlist
968 cpuWaitList.push_back(tid);
969 }
970}
971
972template <class Impl>
973Fault
974FullO3CPU<Impl>::hwrei(ThreadID tid)
975{
976#if THE_ISA == ALPHA_ISA
977 // Need to clear the lock flag upon returning from an interrupt.
978 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
979
980 this->thread[tid]->kernelStats->hwrei();
981
982 // FIXME: XXX check for interrupts? XXX
983#endif
984 return NoFault;
985}
986
987template <class Impl>
988bool
989FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
990{
991#if THE_ISA == ALPHA_ISA
992 if (this->thread[tid]->kernelStats)
993 this->thread[tid]->kernelStats->callpal(palFunc,
994 this->threadContexts[tid]);
995
996 switch (palFunc) {
997 case PAL::halt:
998 halt();
999 if (--System::numSystemsRunning == 0)
1000 exitSimLoop("all cpus halted");
1001 break;
1002
1003 case PAL::bpt:
1004 case PAL::bugchk:
1005 if (this->system->breakpoint())
1006 return false;
1007 break;
1008 }
1009#endif
1010 return true;
1011}
1012
1013template <class Impl>
1014Fault
1015FullO3CPU<Impl>::getInterrupts()
1016{
1017 // Check if there are any outstanding interrupts
1018 return this->interrupts->getInterrupt(this->threadContexts[0]);
1019}
1020
1021template <class Impl>
1022void
1023FullO3CPU<Impl>::processInterrupts(Fault interrupt)
1024{
1025 // Check for interrupts here. For now can copy the code that
1026 // exists within isa_fullsys_traits.hh. Also assume that thread 0
1027 // is the one that handles the interrupts.
1028 // @todo: Possibly consolidate the interrupt checking code.
1029 // @todo: Allow other threads to handle interrupts.
1030
1031 assert(interrupt != NoFault);
1032 this->interrupts->updateIntrInfo(this->threadContexts[0]);
1033
1034 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
1035 this->trap(interrupt, 0, NULL);
1036}
1037
1038template <class Impl>
1039void
1040FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
1041{
1042 // Pass the thread's TC into the invoke method.
1043 fault->invoke(this->threadContexts[tid], inst);
1044}
1045
1046template <class Impl>
1047void
1048FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
1049{
1050 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
1051
1052 DPRINTF(Activity,"Activity: syscall() called.\n");
1053
1054 // Temporarily increase this by one to account for the syscall
1055 // instruction.
1056 ++(this->thread[tid]->funcExeInst);
1057
1058 // Execute the actual syscall.
1059 this->thread[tid]->syscall(callnum);
1060
1061 // Decrease funcExeInst by one as the normal commit will handle
1062 // incrementing it.
1063 --(this->thread[tid]->funcExeInst);
1064}
1065
1066template <class Impl>
1067void
1068FullO3CPU<Impl>::serialize(std::ostream &os)
1069{
1070 SimObject::State so_state = SimObject::getState();
1071 SERIALIZE_ENUM(so_state);
1072 BaseCPU::serialize(os);
1073 nameOut(os, csprintf("%s.tickEvent", name()));
1074 tickEvent.serialize(os);
1075
1076 // Use SimpleThread's ability to checkpoint to make it easier to
1077 // write out the registers. Also make this static so it doesn't
1078 // get instantiated multiple times (causes a panic in statistics).
1079 static SimpleThread temp;
1080
1081 ThreadID size = thread.size();
1082 for (ThreadID i = 0; i < size; i++) {
1083 nameOut(os, csprintf("%s.xc.%i", name(), i));
1084 temp.copyTC(thread[i]->getTC());
1085 temp.serialize(os);
1086 }
1087}
1088
1089template <class Impl>
1090void
1091FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string &section)
1092{
1093 SimObject::State so_state;
1094 UNSERIALIZE_ENUM(so_state);
1095 BaseCPU::unserialize(cp, section);
1096 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
1097
1098 // Use SimpleThread's ability to checkpoint to make it easier to
1099 // read in the registers. Also make this static so it doesn't
1100 // get instantiated multiple times (causes a panic in statistics).
1101 static SimpleThread temp;
1102
1103 ThreadID size = thread.size();
1104 for (ThreadID i = 0; i < size; i++) {
1105 temp.copyTC(thread[i]->getTC());
1106 temp.unserialize(cp, csprintf("%s.xc.%i", section, i));
1107 thread[i]->getTC()->copyArchRegs(temp.getTC());
1108 }
1109}
1110
1111template <class Impl>
1112unsigned int
1113FullO3CPU<Impl>::drain(Event *drain_event)
1114{
1115 DPRINTF(O3CPU, "Switching out\n");
1116
1117 // If the CPU isn't doing anything, then return immediately.
1118 if (_status == Idle || _status == SwitchedOut) {
1119 return 0;
1120 }
1121
1122 drainCount = 0;
1123 fetch.drain();
1124 decode.drain();
1125 rename.drain();
1126 iew.drain();
1127 commit.drain();
1128
1129 // Wake the CPU and record activity so everything can drain out if
1130 // the CPU was not able to immediately drain.
1131 if (getState() != SimObject::Drained) {
1132 // A bit of a hack...set the drainEvent after all the drain()
1133 // calls have been made, that way if all of the stages drain
1134 // immediately, the signalDrained() function knows not to call
1135 // process on the drain event.
1136 drainEvent = drain_event;
1137
1138 wakeCPU();
1139 activityRec.activity();
1140
1141 return 1;
1142 } else {
1143 return 0;
1144 }
1145}
1146
1147template <class Impl>
1148void
1149FullO3CPU<Impl>::resume()
1150{
1151 fetch.resume();
1152 decode.resume();
1153 rename.resume();
1154 iew.resume();
1155 commit.resume();
1156
1157 changeState(SimObject::Running);
1158
1159 if (_status == SwitchedOut || _status == Idle)
1160 return;
1161
1162 assert(system->getMemoryMode() == Enums::timing);
1163
1164 if (!tickEvent.scheduled())
1165 schedule(tickEvent, nextCycle());
1166 _status = Running;
1167}
1168
1169template <class Impl>
1170void
1171FullO3CPU<Impl>::signalDrained()
1172{
1173 if (++drainCount == NumStages) {
1174 if (tickEvent.scheduled())
1175 tickEvent.squash();
1176
1177 changeState(SimObject::Drained);
1178
1179 BaseCPU::switchOut();
1180
1181 if (drainEvent) {
1182 drainEvent->process();
1183 drainEvent = NULL;
1184 }
1185 }
1186 assert(drainCount <= 5);
1187}
1188
1189template <class Impl>
1190void
1191FullO3CPU<Impl>::switchOut()
1192{
1193 fetch.switchOut();
1194 rename.switchOut();
1195 iew.switchOut();
1196 commit.switchOut();
1197 instList.clear();
1198 while (!removeList.empty()) {
1199 removeList.pop();
1200 }
1201
1202 _status = SwitchedOut;
1203
1204 if (checker)
1205 checker->switchOut();
1206
1207 if (tickEvent.scheduled())
1208 tickEvent.squash();
1209}
1210
1211template <class Impl>
1212void
1213FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1214{
1215 // Flush out any old data from the time buffers.
1216 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1217 timeBuffer.advance();
1218 fetchQueue.advance();
1219 decodeQueue.advance();
1220 renameQueue.advance();
1221 iewQueue.advance();
1222 }
1223
1224 activityRec.reset();
1225
1226 BaseCPU::takeOverFrom(oldCPU);
1227
1228 fetch.takeOverFrom();
1229 decode.takeOverFrom();
1230 rename.takeOverFrom();
1231 iew.takeOverFrom();
1232 commit.takeOverFrom();
1233
1234 assert(!tickEvent.scheduled() || tickEvent.squashed());
1235
1236 // @todo: Figure out how to properly select the tid to put onto
1237 // the active threads list.
1238 ThreadID tid = 0;
1239
1240 list<ThreadID>::iterator isActive =
1241 std::find(activeThreads.begin(), activeThreads.end(), tid);
1242
1243 if (isActive == activeThreads.end()) {
1244 //May Need to Re-code this if the delay variable is the delay
1245 //needed for thread to activate
1246 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n",
1247 tid);
1248
1249 activeThreads.push_back(tid);
1250 }
1251
1252 // Set all statuses to active, schedule the CPU's tick event.
1253 // @todo: Fix up statuses so this is handled properly
1254 ThreadID size = threadContexts.size();
1255 for (ThreadID i = 0; i < size; ++i) {
1256 ThreadContext *tc = threadContexts[i];
1257 if (tc->status() == ThreadContext::Active && _status != Running) {
1258 _status = Running;
1259 reschedule(tickEvent, nextCycle(), true);
1260 }
1261 }
1262 if (!tickEvent.scheduled())
1263 schedule(tickEvent, nextCycle());
1264
1265 lastRunningCycle = curTick();
1266}
1267
1268template <class Impl>
1269TheISA::MiscReg
1270FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1271{
1272 return this->isa[tid].readMiscRegNoEffect(misc_reg);
1273}
1274
1275template <class Impl>
1276TheISA::MiscReg
1277FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1278{
1279 miscRegfileReads++;
1280 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid));
1281}
1282
1283template <class Impl>
1284void
1285FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1286 const TheISA::MiscReg &val, ThreadID tid)
1287{
1288 this->isa[tid].setMiscRegNoEffect(misc_reg, val);
1289}
1290
1291template <class Impl>
1292void
1293FullO3CPU<Impl>::setMiscReg(int misc_reg,
1294 const TheISA::MiscReg &val, ThreadID tid)
1295{
1296 miscRegfileWrites++;
1297 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid));
1298}
1299
1300template <class Impl>
1301uint64_t
1302FullO3CPU<Impl>::readIntReg(int reg_idx)
1303{
1304 intRegfileReads++;
1305 return regFile.readIntReg(reg_idx);
1306}
1307
1308template <class Impl>
1309FloatReg
1310FullO3CPU<Impl>::readFloatReg(int reg_idx)
1311{
1312 fpRegfileReads++;
1313 return regFile.readFloatReg(reg_idx);
1314}
1315
1316template <class Impl>
1317FloatRegBits
1318FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1319{
1320 fpRegfileReads++;
1321 return regFile.readFloatRegBits(reg_idx);
1322}
1323
1324template <class Impl>
1325void
1326FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1327{
1328 intRegfileWrites++;
1329 regFile.setIntReg(reg_idx, val);
1330}
1331
1332template <class Impl>
1333void
1334FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1335{
1336 fpRegfileWrites++;
1337 regFile.setFloatReg(reg_idx, val);
1338}
1339
1340template <class Impl>
1341void
1342FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1343{
1344 fpRegfileWrites++;
1345 regFile.setFloatRegBits(reg_idx, val);
1346}
1347
1348template <class Impl>
1349uint64_t
1350FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1351{
1352 intRegfileReads++;
1353 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1354
1355 return regFile.readIntReg(phys_reg);
1356}
1357
1358template <class Impl>
1359float
1360FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1361{
1362 fpRegfileReads++;
1363 int idx = reg_idx + TheISA::NumIntRegs;
1364 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1365
1366 return regFile.readFloatReg(phys_reg);
1367}
1368
1369template <class Impl>
1370uint64_t
1371FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1372{
1373 fpRegfileReads++;
1374 int idx = reg_idx + TheISA::NumIntRegs;
1375 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1376
1377 return regFile.readFloatRegBits(phys_reg);
1378}
1379
1380template <class Impl>
1381void
1382FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1383{
1384 intRegfileWrites++;
1385 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1386
1387 regFile.setIntReg(phys_reg, val);
1388}
1389
1390template <class Impl>
1391void
1392FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1393{
1394 fpRegfileWrites++;
1395 int idx = reg_idx + TheISA::NumIntRegs;
1396 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1397
1398 regFile.setFloatReg(phys_reg, val);
1399}
1400
1401template <class Impl>
1402void
1403FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1404{
1405 fpRegfileWrites++;
1406 int idx = reg_idx + TheISA::NumIntRegs;
1407 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1408
1409 regFile.setFloatRegBits(phys_reg, val);
1410}
1411
1412template <class Impl>
1413TheISA::PCState
1414FullO3CPU<Impl>::pcState(ThreadID tid)
1415{
1416 return commit.pcState(tid);
1417}
1418
1419template <class Impl>
1420void
1421FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1422{
1423 commit.pcState(val, tid);
1424}
1425
1426template <class Impl>
1427Addr
1428FullO3CPU<Impl>::instAddr(ThreadID tid)
1429{
1430 return commit.instAddr(tid);
1431}
1432
1433template <class Impl>
1434Addr
1435FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1436{
1437 return commit.nextInstAddr(tid);
1438}
1439
1440template <class Impl>
1441MicroPC
1442FullO3CPU<Impl>::microPC(ThreadID tid)
1443{
1444 return commit.microPC(tid);
1445}
1446
1447template <class Impl>
1448void
1449FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1450{
1451 this->thread[tid]->inSyscall = true;
1452 this->commit.generateTCEvent(tid);
1453}
1454
1455template <class Impl>
1456typename FullO3CPU<Impl>::ListIt
1457FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1458{
1459 instList.push_back(inst);
1460
1461 return --(instList.end());
1462}
1463
1464template <class Impl>
1465void
1466FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst)
1467{
1468 // Keep an instruction count.
1469 if (!inst->isMicroop() || inst->isLastMicroop()) {
1470 thread[tid]->numInst++;
1471 thread[tid]->numInsts++;
1472 committedInsts[tid]++;
1473 totalCommittedInsts++;
1474 }
1475 thread[tid]->numOp++;
1476 thread[tid]->numOps++;
1477 committedOps[tid]++;
1478
1479 system->totalNumInsts++;
1480 // Check for instruction-count-based events.
1481 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1482 system->instEventQueue.serviceEvents(system->totalNumInsts);
1483}
1484
1485template <class Impl>
1486void
1487FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1488{
1489 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1490 "[sn:%lli]\n",
1491 inst->threadNumber, inst->pcState(), inst->seqNum);
1492
1493 removeInstsThisCycle = true;
1494
1495 // Remove the front instruction.
1496 removeList.push(inst->getInstListIt());
1497}
1498
1499template <class Impl>
1500void
1501FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1502{
1503 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1504 " list.\n", tid);
1505
1506 ListIt end_it;
1507
1508 bool rob_empty = false;
1509
1510 if (instList.empty()) {
1511 return;
1512 } else if (rob.isEmpty(/*tid*/)) {
1513 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1514 end_it = instList.begin();
1515 rob_empty = true;
1516 } else {
1517 end_it = (rob.readTailInst(tid))->getInstListIt();
1518 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1519 }
1520
1521 removeInstsThisCycle = true;
1522
1523 ListIt inst_it = instList.end();
1524
1525 inst_it--;
1526
1527 // Walk through the instruction list, removing any instructions
1528 // that were inserted after the given instruction iterator, end_it.
1529 while (inst_it != end_it) {
1530 assert(!instList.empty());
1531
1532 squashInstIt(inst_it, tid);
1533
1534 inst_it--;
1535 }
1536
1537 // If the ROB was empty, then we actually need to remove the first
1538 // instruction as well.
1539 if (rob_empty) {
1540 squashInstIt(inst_it, tid);
1541 }
1542}
1543
1544template <class Impl>
1545void
1546FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1547{
1548 assert(!instList.empty());
1549
1550 removeInstsThisCycle = true;
1551
1552 ListIt inst_iter = instList.end();
1553
1554 inst_iter--;
1555
1556 DPRINTF(O3CPU, "Deleting instructions from instruction "
1557 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1558 tid, seq_num, (*inst_iter)->seqNum);
1559
1560 while ((*inst_iter)->seqNum > seq_num) {
1561
1562 bool break_loop = (inst_iter == instList.begin());
1563
1564 squashInstIt(inst_iter, tid);
1565
1566 inst_iter--;
1567
1568 if (break_loop)
1569 break;
1570 }
1571}
1572
1573template <class Impl>
1574inline void
1575FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1576{
1577 if ((*instIt)->threadNumber == tid) {
1578 DPRINTF(O3CPU, "Squashing instruction, "
1579 "[tid:%i] [sn:%lli] PC %s\n",
1580 (*instIt)->threadNumber,
1581 (*instIt)->seqNum,
1582 (*instIt)->pcState());
1583
1584 // Mark it as squashed.
1585 (*instIt)->setSquashed();
1586
1587 // @todo: Formulate a consistent method for deleting
1588 // instructions from the instruction list
1589 // Remove the instruction from the list.
1590 removeList.push(instIt);
1591 }
1592}
1593
1594template <class Impl>
1595void
1596FullO3CPU<Impl>::cleanUpRemovedInsts()
1597{
1598 while (!removeList.empty()) {
1599 DPRINTF(O3CPU, "Removing instruction, "
1600 "[tid:%i] [sn:%lli] PC %s\n",
1601 (*removeList.front())->threadNumber,
1602 (*removeList.front())->seqNum,
1603 (*removeList.front())->pcState());
1604
1605 instList.erase(removeList.front());
1606
1607 removeList.pop();
1608 }
1609
1610 removeInstsThisCycle = false;
1611}
1612/*
1613template <class Impl>
1614void
1615FullO3CPU<Impl>::removeAllInsts()
1616{
1617 instList.clear();
1618}
1619*/
1620template <class Impl>
1621void
1622FullO3CPU<Impl>::dumpInsts()
1623{
1624 int num = 0;
1625
1626 ListIt inst_list_it = instList.begin();
1627
1628 cprintf("Dumping Instruction List\n");
1629
1630 while (inst_list_it != instList.end()) {
1631 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1632 "Squashed:%i\n\n",
1633 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1634 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1635 (*inst_list_it)->isSquashed());
1636 inst_list_it++;
1637 ++num;
1638 }
1639}
1640/*
1641template <class Impl>
1642void
1643FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1644{
1645 iew.wakeDependents(inst);
1646}
1647*/
1648template <class Impl>
1649void
1650FullO3CPU<Impl>::wakeCPU()
1651{
1652 if (activityRec.active() || tickEvent.scheduled()) {
1653 DPRINTF(Activity, "CPU already running.\n");
1654 return;
1655 }
1656
1657 DPRINTF(Activity, "Waking up CPU\n");
1658
1659 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1660 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1661
1662 schedule(tickEvent, nextCycle());
1663}
1664
1665template <class Impl>
1666void
1667FullO3CPU<Impl>::wakeup()
1668{
1669 if (this->thread[0]->status() != ThreadContext::Suspended)
1670 return;
1671
1672 this->wakeCPU();
1673
1674 DPRINTF(Quiesce, "Suspended Processor woken\n");
1675 this->threadContexts[0]->activate();
1676}
1677
1678template <class Impl>
1679ThreadID
1680FullO3CPU<Impl>::getFreeTid()
1681{
1682 for (ThreadID tid = 0; tid < numThreads; tid++) {
1683 if (!tids[tid]) {
1684 tids[tid] = true;
1685 return tid;
1686 }
1687 }
1688
1689 return InvalidThreadID;
1690}
1691
1692template <class Impl>
1693void
1694FullO3CPU<Impl>::doContextSwitch()
1695{
1696 if (contextSwitch) {
1697
1698 //ADD CODE TO DEACTIVE THREAD HERE (???)
1699
1700 ThreadID size = cpuWaitList.size();
1701 for (ThreadID tid = 0; tid < size; tid++) {
1702 activateWhenReady(tid);
1703 }
1704
1705 if (cpuWaitList.size() == 0)
1706 contextSwitch = true;
1707 }
1708}
1709
1710template <class Impl>
1711void
1712FullO3CPU<Impl>::updateThreadPriority()
1713{
1714 if (activeThreads.size() > 1) {
1715 //DEFAULT TO ROUND ROBIN SCHEME
1716 //e.g. Move highest priority to end of thread list
1717 list<ThreadID>::iterator list_begin = activeThreads.begin();
1718
1719 unsigned high_thread = *list_begin;
1720
1721 activeThreads.erase(list_begin);
1722
1723 activeThreads.push_back(high_thread);
1724 }
1725}
1726
1727// Forward declaration of FullO3CPU.
1728template class FullO3CPU<O3CPUImpl>;
125void
126FullO3CPU<Impl>::DcachePort::recvRetry()
127{
128 lsq->recvRetry();
129}
130
131template <class Impl>
132FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c)
133 : Event(CPU_Tick_Pri), cpu(c)
134{
135}
136
137template <class Impl>
138void
139FullO3CPU<Impl>::TickEvent::process()
140{
141 cpu->tick();
142}
143
144template <class Impl>
145const char *
146FullO3CPU<Impl>::TickEvent::description() const
147{
148 return "FullO3CPU tick";
149}
150
151template <class Impl>
152FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent()
153 : Event(CPU_Switch_Pri)
154{
155}
156
157template <class Impl>
158void
159FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num,
160 FullO3CPU<Impl> *thread_cpu)
161{
162 tid = thread_num;
163 cpu = thread_cpu;
164}
165
166template <class Impl>
167void
168FullO3CPU<Impl>::ActivateThreadEvent::process()
169{
170 cpu->activateThread(tid);
171}
172
173template <class Impl>
174const char *
175FullO3CPU<Impl>::ActivateThreadEvent::description() const
176{
177 return "FullO3CPU \"Activate Thread\"";
178}
179
180template <class Impl>
181FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent()
182 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL)
183{
184}
185
186template <class Impl>
187void
188FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num,
189 FullO3CPU<Impl> *thread_cpu)
190{
191 tid = thread_num;
192 cpu = thread_cpu;
193 remove = false;
194}
195
196template <class Impl>
197void
198FullO3CPU<Impl>::DeallocateContextEvent::process()
199{
200 cpu->deactivateThread(tid);
201 if (remove)
202 cpu->removeThread(tid);
203}
204
205template <class Impl>
206const char *
207FullO3CPU<Impl>::DeallocateContextEvent::description() const
208{
209 return "FullO3CPU \"Deallocate Context\"";
210}
211
212template <class Impl>
213FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params)
214 : BaseO3CPU(params),
215 itb(params->itb),
216 dtb(params->dtb),
217 tickEvent(this),
218#ifndef NDEBUG
219 instcount(0),
220#endif
221 removeInstsThisCycle(false),
222 fetch(this, params),
223 decode(this, params),
224 rename(this, params),
225 iew(this, params),
226 commit(this, params),
227
228 regFile(this, params->numPhysIntRegs,
229 params->numPhysFloatRegs),
230
231 freeList(params->numThreads,
232 TheISA::NumIntRegs, params->numPhysIntRegs,
233 TheISA::NumFloatRegs, params->numPhysFloatRegs),
234
235 rob(this,
236 params->numROBEntries, params->squashWidth,
237 params->smtROBPolicy, params->smtROBThreshold,
238 params->numThreads),
239
240 scoreboard(params->numThreads,
241 TheISA::NumIntRegs, params->numPhysIntRegs,
242 TheISA::NumFloatRegs, params->numPhysFloatRegs,
243 TheISA::NumMiscRegs * numThreads,
244 TheISA::ZeroReg),
245
246 icachePort(&fetch, this),
247 dcachePort(&iew.ldstQueue, this),
248
249 timeBuffer(params->backComSize, params->forwardComSize),
250 fetchQueue(params->backComSize, params->forwardComSize),
251 decodeQueue(params->backComSize, params->forwardComSize),
252 renameQueue(params->backComSize, params->forwardComSize),
253 iewQueue(params->backComSize, params->forwardComSize),
254 activityRec(name(), NumStages,
255 params->backComSize + params->forwardComSize,
256 params->activity),
257
258 globalSeqNum(1),
259 system(params->system),
260 drainCount(0),
261 deferRegistration(params->defer_registration)
262{
263 if (!deferRegistration) {
264 _status = Running;
265 } else {
266 _status = Idle;
267 }
268
269 if (params->checker) {
270 BaseCPU *temp_checker = params->checker;
271 checker = dynamic_cast<Checker<Impl> *>(temp_checker);
272 checker->setIcachePort(&icachePort);
273 checker->setSystem(params->system);
274 } else {
275 checker = NULL;
276 }
277
278 if (!FullSystem) {
279 thread.resize(numThreads);
280 tids.resize(numThreads);
281 }
282
283 // The stages also need their CPU pointer setup. However this
284 // must be done at the upper level CPU because they have pointers
285 // to the upper level CPU, and not this FullO3CPU.
286
287 // Set up Pointers to the activeThreads list for each stage
288 fetch.setActiveThreads(&activeThreads);
289 decode.setActiveThreads(&activeThreads);
290 rename.setActiveThreads(&activeThreads);
291 iew.setActiveThreads(&activeThreads);
292 commit.setActiveThreads(&activeThreads);
293
294 // Give each of the stages the time buffer they will use.
295 fetch.setTimeBuffer(&timeBuffer);
296 decode.setTimeBuffer(&timeBuffer);
297 rename.setTimeBuffer(&timeBuffer);
298 iew.setTimeBuffer(&timeBuffer);
299 commit.setTimeBuffer(&timeBuffer);
300
301 // Also setup each of the stages' queues.
302 fetch.setFetchQueue(&fetchQueue);
303 decode.setFetchQueue(&fetchQueue);
304 commit.setFetchQueue(&fetchQueue);
305 decode.setDecodeQueue(&decodeQueue);
306 rename.setDecodeQueue(&decodeQueue);
307 rename.setRenameQueue(&renameQueue);
308 iew.setRenameQueue(&renameQueue);
309 iew.setIEWQueue(&iewQueue);
310 commit.setIEWQueue(&iewQueue);
311 commit.setRenameQueue(&renameQueue);
312
313 commit.setIEWStage(&iew);
314 rename.setIEWStage(&iew);
315 rename.setCommitStage(&commit);
316
317 ThreadID active_threads;
318 if (FullSystem) {
319 active_threads = 1;
320 } else {
321 active_threads = params->workload.size();
322
323 if (active_threads > Impl::MaxThreads) {
324 panic("Workload Size too large. Increase the 'MaxThreads' "
325 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
326 "or edit your workload size.");
327 }
328 }
329
330 //Make Sure That this a Valid Architeture
331 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
332 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
333
334 rename.setScoreboard(&scoreboard);
335 iew.setScoreboard(&scoreboard);
336
337 // Setup the rename map for whichever stages need it.
338 PhysRegIndex lreg_idx = 0;
339 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
340
341 for (ThreadID tid = 0; tid < numThreads; tid++) {
342 bool bindRegs = (tid <= active_threads - 1);
343
344 commitRenameMap[tid].init(TheISA::NumIntRegs,
345 params->numPhysIntRegs,
346 lreg_idx, //Index for Logical. Regs
347
348 TheISA::NumFloatRegs,
349 params->numPhysFloatRegs,
350 freg_idx, //Index for Float Regs
351
352 TheISA::NumMiscRegs,
353
354 TheISA::ZeroReg,
355 TheISA::ZeroReg,
356
357 tid,
358 false);
359
360 renameMap[tid].init(TheISA::NumIntRegs,
361 params->numPhysIntRegs,
362 lreg_idx, //Index for Logical. Regs
363
364 TheISA::NumFloatRegs,
365 params->numPhysFloatRegs,
366 freg_idx, //Index for Float Regs
367
368 TheISA::NumMiscRegs,
369
370 TheISA::ZeroReg,
371 TheISA::ZeroReg,
372
373 tid,
374 bindRegs);
375
376 activateThreadEvent[tid].init(tid, this);
377 deallocateContextEvent[tid].init(tid, this);
378 }
379
380 rename.setRenameMap(renameMap);
381 commit.setRenameMap(commitRenameMap);
382
383 // Give renameMap & rename stage access to the freeList;
384 for (ThreadID tid = 0; tid < numThreads; tid++)
385 renameMap[tid].setFreeList(&freeList);
386 rename.setFreeList(&freeList);
387
388 // Setup the ROB for whichever stages need it.
389 commit.setROB(&rob);
390
391 lastRunningCycle = curTick();
392
393 lastActivatedCycle = -1;
394#if 0
395 // Give renameMap & rename stage access to the freeList;
396 for (ThreadID tid = 0; tid < numThreads; tid++)
397 globalSeqNum[tid] = 1;
398#endif
399
400 contextSwitch = false;
401 DPRINTF(O3CPU, "Creating O3CPU object.\n");
402
403 // Setup any thread state.
404 this->thread.resize(this->numThreads);
405
406 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
407 if (FullSystem) {
408 // SMT is not supported in FS mode yet.
409 assert(this->numThreads == 1);
410 this->thread[tid] = new Thread(this, 0, NULL);
411 } else {
412 if (tid < params->workload.size()) {
413 DPRINTF(O3CPU, "Workload[%i] process is %#x",
414 tid, this->thread[tid]);
415 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
416 (typename Impl::O3CPU *)(this),
417 tid, params->workload[tid]);
418
419 //usedTids[tid] = true;
420 //threadMap[tid] = tid;
421 } else {
422 //Allocate Empty thread so M5 can use later
423 //when scheduling threads to CPU
424 Process* dummy_proc = NULL;
425
426 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
427 (typename Impl::O3CPU *)(this),
428 tid, dummy_proc);
429 //usedTids[tid] = false;
430 }
431 }
432
433 ThreadContext *tc;
434
435 // Setup the TC that will serve as the interface to the threads/CPU.
436 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
437
438 tc = o3_tc;
439
440 // If we're using a checker, then the TC should be the
441 // CheckerThreadContext.
442 if (params->checker) {
443 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
444 o3_tc, this->checker);
445 }
446
447 o3_tc->cpu = (typename Impl::O3CPU *)(this);
448 assert(o3_tc->cpu);
449 o3_tc->thread = this->thread[tid];
450
451 if (FullSystem) {
452 // Setup quiesce event.
453 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
454 }
455 // Give the thread the TC.
456 this->thread[tid]->tc = tc;
457
458 // Add the TC to the CPU's list of TC's.
459 this->threadContexts.push_back(tc);
460 }
461
462 // FullO3CPU always requires an interrupt controller.
463 if (!params->defer_registration && !interrupts) {
464 fatal("FullO3CPU %s has no interrupt controller.\n"
465 "Ensure createInterruptController() is called.\n", name());
466 }
467
468 for (ThreadID tid = 0; tid < this->numThreads; tid++)
469 this->thread[tid]->setFuncExeInst(0);
470
471 lockAddr = 0;
472 lockFlag = false;
473}
474
475template <class Impl>
476FullO3CPU<Impl>::~FullO3CPU()
477{
478}
479
480template <class Impl>
481void
482FullO3CPU<Impl>::regStats()
483{
484 BaseO3CPU::regStats();
485
486 // Register any of the O3CPU's stats here.
487 timesIdled
488 .name(name() + ".timesIdled")
489 .desc("Number of times that the entire CPU went into an idle state and"
490 " unscheduled itself")
491 .prereq(timesIdled);
492
493 idleCycles
494 .name(name() + ".idleCycles")
495 .desc("Total number of cycles that the CPU has spent unscheduled due "
496 "to idling")
497 .prereq(idleCycles);
498
499 quiesceCycles
500 .name(name() + ".quiesceCycles")
501 .desc("Total number of cycles that CPU has spent quiesced or waiting "
502 "for an interrupt")
503 .prereq(quiesceCycles);
504
505 // Number of Instructions simulated
506 // --------------------------------
507 // Should probably be in Base CPU but need templated
508 // MaxThreads so put in here instead
509 committedInsts
510 .init(numThreads)
511 .name(name() + ".committedInsts")
512 .desc("Number of Instructions Simulated");
513
514 committedOps
515 .init(numThreads)
516 .name(name() + ".committedOps")
517 .desc("Number of Ops (including micro ops) Simulated");
518
519 totalCommittedInsts
520 .name(name() + ".committedInsts_total")
521 .desc("Number of Instructions Simulated");
522
523 cpi
524 .name(name() + ".cpi")
525 .desc("CPI: Cycles Per Instruction")
526 .precision(6);
527 cpi = numCycles / committedInsts;
528
529 totalCpi
530 .name(name() + ".cpi_total")
531 .desc("CPI: Total CPI of All Threads")
532 .precision(6);
533 totalCpi = numCycles / totalCommittedInsts;
534
535 ipc
536 .name(name() + ".ipc")
537 .desc("IPC: Instructions Per Cycle")
538 .precision(6);
539 ipc = committedInsts / numCycles;
540
541 totalIpc
542 .name(name() + ".ipc_total")
543 .desc("IPC: Total IPC of All Threads")
544 .precision(6);
545 totalIpc = totalCommittedInsts / numCycles;
546
547 this->fetch.regStats();
548 this->decode.regStats();
549 this->rename.regStats();
550 this->iew.regStats();
551 this->commit.regStats();
552 this->rob.regStats();
553
554 intRegfileReads
555 .name(name() + ".int_regfile_reads")
556 .desc("number of integer regfile reads")
557 .prereq(intRegfileReads);
558
559 intRegfileWrites
560 .name(name() + ".int_regfile_writes")
561 .desc("number of integer regfile writes")
562 .prereq(intRegfileWrites);
563
564 fpRegfileReads
565 .name(name() + ".fp_regfile_reads")
566 .desc("number of floating regfile reads")
567 .prereq(fpRegfileReads);
568
569 fpRegfileWrites
570 .name(name() + ".fp_regfile_writes")
571 .desc("number of floating regfile writes")
572 .prereq(fpRegfileWrites);
573
574 miscRegfileReads
575 .name(name() + ".misc_regfile_reads")
576 .desc("number of misc regfile reads")
577 .prereq(miscRegfileReads);
578
579 miscRegfileWrites
580 .name(name() + ".misc_regfile_writes")
581 .desc("number of misc regfile writes")
582 .prereq(miscRegfileWrites);
583}
584
585template <class Impl>
586void
587FullO3CPU<Impl>::tick()
588{
589 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
590
591 ++numCycles;
592
593// activity = false;
594
595 //Tick each of the stages
596 fetch.tick();
597
598 decode.tick();
599
600 rename.tick();
601
602 iew.tick();
603
604 commit.tick();
605
606 if (!FullSystem)
607 doContextSwitch();
608
609 // Now advance the time buffers
610 timeBuffer.advance();
611
612 fetchQueue.advance();
613 decodeQueue.advance();
614 renameQueue.advance();
615 iewQueue.advance();
616
617 activityRec.advance();
618
619 if (removeInstsThisCycle) {
620 cleanUpRemovedInsts();
621 }
622
623 if (!tickEvent.scheduled()) {
624 if (_status == SwitchedOut ||
625 getState() == SimObject::Drained) {
626 DPRINTF(O3CPU, "Switched out!\n");
627 // increment stat
628 lastRunningCycle = curTick();
629 } else if (!activityRec.active() || _status == Idle) {
630 DPRINTF(O3CPU, "Idle!\n");
631 lastRunningCycle = curTick();
632 timesIdled++;
633 } else {
634 schedule(tickEvent, nextCycle(curTick() + ticks(1)));
635 DPRINTF(O3CPU, "Scheduling next tick!\n");
636 }
637 }
638
639 if (!FullSystem)
640 updateThreadPriority();
641}
642
643template <class Impl>
644void
645FullO3CPU<Impl>::init()
646{
647 BaseCPU::init();
648
649 for (ThreadID tid = 0; tid < numThreads; ++tid) {
650 // Set inSyscall so that the CPU doesn't squash when initially
651 // setting up registers.
652 thread[tid]->inSyscall = true;
653 // Initialise the ThreadContext's memory proxies
654 thread[tid]->initMemProxies(thread[tid]->getTC());
655 }
656
657 // this CPU could still be unconnected if we are restoring from a
658 // checkpoint and this CPU is to be switched in, thus we can only
659 // do this here if the instruction port is actually connected, if
660 // not we have to do it as part of takeOverFrom
661 if (icachePort.isConnected())
662 fetch.setIcache();
663
664 if (FullSystem && !params()->defer_registration) {
665 for (ThreadID tid = 0; tid < numThreads; tid++) {
666 ThreadContext *src_tc = threadContexts[tid];
667 TheISA::initCPU(src_tc, src_tc->contextId());
668 }
669 }
670
671 // Clear inSyscall.
672 for (int tid = 0; tid < numThreads; ++tid)
673 thread[tid]->inSyscall = false;
674
675 // Initialize stages.
676 fetch.initStage();
677 iew.initStage();
678 rename.initStage();
679 commit.initStage();
680
681 commit.setThreads(thread);
682}
683
684template <class Impl>
685void
686FullO3CPU<Impl>::activateThread(ThreadID tid)
687{
688 list<ThreadID>::iterator isActive =
689 std::find(activeThreads.begin(), activeThreads.end(), tid);
690
691 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid);
692
693 if (isActive == activeThreads.end()) {
694 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
695 tid);
696
697 activeThreads.push_back(tid);
698 }
699}
700
701template <class Impl>
702void
703FullO3CPU<Impl>::deactivateThread(ThreadID tid)
704{
705 //Remove From Active List, if Active
706 list<ThreadID>::iterator thread_it =
707 std::find(activeThreads.begin(), activeThreads.end(), tid);
708
709 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
710
711 if (thread_it != activeThreads.end()) {
712 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
713 tid);
714 activeThreads.erase(thread_it);
715 }
716}
717
718template <class Impl>
719Counter
720FullO3CPU<Impl>::totalInsts() const
721{
722 Counter total(0);
723
724 ThreadID size = thread.size();
725 for (ThreadID i = 0; i < size; i++)
726 total += thread[i]->numInst;
727
728 return total;
729}
730
731template <class Impl>
732Counter
733FullO3CPU<Impl>::totalOps() const
734{
735 Counter total(0);
736
737 ThreadID size = thread.size();
738 for (ThreadID i = 0; i < size; i++)
739 total += thread[i]->numOp;
740
741 return total;
742}
743
744template <class Impl>
745void
746FullO3CPU<Impl>::activateContext(ThreadID tid, int delay)
747{
748 // Needs to set each stage to running as well.
749 if (delay){
750 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
751 "on cycle %d\n", tid, curTick() + ticks(delay));
752 scheduleActivateThreadEvent(tid, delay);
753 } else {
754 activateThread(tid);
755 }
756
757 if (lastActivatedCycle < curTick()) {
758 scheduleTickEvent(delay);
759
760 // Be sure to signal that there's some activity so the CPU doesn't
761 // deschedule itself.
762 activityRec.activity();
763 fetch.wakeFromQuiesce();
764
765 quiesceCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
766
767 lastActivatedCycle = curTick();
768
769 _status = Running;
770 }
771}
772
773template <class Impl>
774bool
775FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove,
776 int delay)
777{
778 // Schedule removal of thread data from CPU
779 if (delay){
780 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
781 "on cycle %d\n", tid, curTick() + ticks(delay));
782 scheduleDeallocateContextEvent(tid, remove, delay);
783 return false;
784 } else {
785 deactivateThread(tid);
786 if (remove)
787 removeThread(tid);
788 return true;
789 }
790}
791
792template <class Impl>
793void
794FullO3CPU<Impl>::suspendContext(ThreadID tid)
795{
796 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
797 bool deallocated = scheduleDeallocateContext(tid, false, 1);
798 // If this was the last thread then unschedule the tick event.
799 if ((activeThreads.size() == 1 && !deallocated) ||
800 activeThreads.size() == 0)
801 unscheduleTickEvent();
802
803 DPRINTF(Quiesce, "Suspending Context\n");
804 lastRunningCycle = curTick();
805 _status = Idle;
806}
807
808template <class Impl>
809void
810FullO3CPU<Impl>::haltContext(ThreadID tid)
811{
812 //For now, this is the same as deallocate
813 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
814 scheduleDeallocateContext(tid, true, 1);
815}
816
817template <class Impl>
818void
819FullO3CPU<Impl>::insertThread(ThreadID tid)
820{
821 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
822 // Will change now that the PC and thread state is internal to the CPU
823 // and not in the ThreadContext.
824 ThreadContext *src_tc;
825 if (FullSystem)
826 src_tc = system->threadContexts[tid];
827 else
828 src_tc = tcBase(tid);
829
830 //Bind Int Regs to Rename Map
831 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
832 PhysRegIndex phys_reg = freeList.getIntReg();
833
834 renameMap[tid].setEntry(ireg,phys_reg);
835 scoreboard.setReg(phys_reg);
836 }
837
838 //Bind Float Regs to Rename Map
839 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
840 PhysRegIndex phys_reg = freeList.getFloatReg();
841
842 renameMap[tid].setEntry(freg,phys_reg);
843 scoreboard.setReg(phys_reg);
844 }
845
846 //Copy Thread Data Into RegFile
847 //this->copyFromTC(tid);
848
849 //Set PC/NPC/NNPC
850 pcState(src_tc->pcState(), tid);
851
852 src_tc->setStatus(ThreadContext::Active);
853
854 activateContext(tid,1);
855
856 //Reset ROB/IQ/LSQ Entries
857 commit.rob->resetEntries();
858 iew.resetEntries();
859}
860
861template <class Impl>
862void
863FullO3CPU<Impl>::removeThread(ThreadID tid)
864{
865 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
866
867 // Copy Thread Data From RegFile
868 // If thread is suspended, it might be re-allocated
869 // this->copyToTC(tid);
870
871
872 // @todo: 2-27-2008: Fix how we free up rename mappings
873 // here to alleviate the case for double-freeing registers
874 // in SMT workloads.
875
876 // Unbind Int Regs from Rename Map
877 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
878 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
879
880 scoreboard.unsetReg(phys_reg);
881 freeList.addReg(phys_reg);
882 }
883
884 // Unbind Float Regs from Rename Map
885 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
886 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
887
888 scoreboard.unsetReg(phys_reg);
889 freeList.addReg(phys_reg);
890 }
891
892 // Squash Throughout Pipeline
893 DynInstPtr inst = commit.rob->readHeadInst(tid);
894 InstSeqNum squash_seq_num = inst->seqNum;
895 fetch.squash(0, squash_seq_num, inst, tid);
896 decode.squash(tid);
897 rename.squash(squash_seq_num, tid);
898 iew.squash(tid);
899 iew.ldstQueue.squash(squash_seq_num, tid);
900 commit.rob->squash(squash_seq_num, tid);
901
902
903 assert(iew.instQueue.getCount(tid) == 0);
904 assert(iew.ldstQueue.getCount(tid) == 0);
905
906 // Reset ROB/IQ/LSQ Entries
907
908 // Commented out for now. This should be possible to do by
909 // telling all the pipeline stages to drain first, and then
910 // checking until the drain completes. Once the pipeline is
911 // drained, call resetEntries(). - 10-09-06 ktlim
912/*
913 if (activeThreads.size() >= 1) {
914 commit.rob->resetEntries();
915 iew.resetEntries();
916 }
917*/
918}
919
920
921template <class Impl>
922void
923FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
924{
925 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
926 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
927 tid);
928
929 bool ready = true;
930
931 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
932 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
933 "Phys. Int. Regs.\n",
934 tid);
935 ready = false;
936 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
937 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
938 "Phys. Float. Regs.\n",
939 tid);
940 ready = false;
941 } else if (commit.rob->numFreeEntries() >=
942 commit.rob->entryAmount(activeThreads.size() + 1)) {
943 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
944 "ROB entries.\n",
945 tid);
946 ready = false;
947 } else if (iew.instQueue.numFreeEntries() >=
948 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
949 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
950 "IQ entries.\n",
951 tid);
952 ready = false;
953 } else if (iew.ldstQueue.numFreeEntries() >=
954 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
955 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
956 "LSQ entries.\n",
957 tid);
958 ready = false;
959 }
960
961 if (ready) {
962 insertThread(tid);
963
964 contextSwitch = false;
965
966 cpuWaitList.remove(tid);
967 } else {
968 suspendContext(tid);
969
970 //blocks fetch
971 contextSwitch = true;
972
973 //@todo: dont always add to waitlist
974 //do waitlist
975 cpuWaitList.push_back(tid);
976 }
977}
978
979template <class Impl>
980Fault
981FullO3CPU<Impl>::hwrei(ThreadID tid)
982{
983#if THE_ISA == ALPHA_ISA
984 // Need to clear the lock flag upon returning from an interrupt.
985 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
986
987 this->thread[tid]->kernelStats->hwrei();
988
989 // FIXME: XXX check for interrupts? XXX
990#endif
991 return NoFault;
992}
993
994template <class Impl>
995bool
996FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
997{
998#if THE_ISA == ALPHA_ISA
999 if (this->thread[tid]->kernelStats)
1000 this->thread[tid]->kernelStats->callpal(palFunc,
1001 this->threadContexts[tid]);
1002
1003 switch (palFunc) {
1004 case PAL::halt:
1005 halt();
1006 if (--System::numSystemsRunning == 0)
1007 exitSimLoop("all cpus halted");
1008 break;
1009
1010 case PAL::bpt:
1011 case PAL::bugchk:
1012 if (this->system->breakpoint())
1013 return false;
1014 break;
1015 }
1016#endif
1017 return true;
1018}
1019
1020template <class Impl>
1021Fault
1022FullO3CPU<Impl>::getInterrupts()
1023{
1024 // Check if there are any outstanding interrupts
1025 return this->interrupts->getInterrupt(this->threadContexts[0]);
1026}
1027
1028template <class Impl>
1029void
1030FullO3CPU<Impl>::processInterrupts(Fault interrupt)
1031{
1032 // Check for interrupts here. For now can copy the code that
1033 // exists within isa_fullsys_traits.hh. Also assume that thread 0
1034 // is the one that handles the interrupts.
1035 // @todo: Possibly consolidate the interrupt checking code.
1036 // @todo: Allow other threads to handle interrupts.
1037
1038 assert(interrupt != NoFault);
1039 this->interrupts->updateIntrInfo(this->threadContexts[0]);
1040
1041 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
1042 this->trap(interrupt, 0, NULL);
1043}
1044
1045template <class Impl>
1046void
1047FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
1048{
1049 // Pass the thread's TC into the invoke method.
1050 fault->invoke(this->threadContexts[tid], inst);
1051}
1052
1053template <class Impl>
1054void
1055FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
1056{
1057 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
1058
1059 DPRINTF(Activity,"Activity: syscall() called.\n");
1060
1061 // Temporarily increase this by one to account for the syscall
1062 // instruction.
1063 ++(this->thread[tid]->funcExeInst);
1064
1065 // Execute the actual syscall.
1066 this->thread[tid]->syscall(callnum);
1067
1068 // Decrease funcExeInst by one as the normal commit will handle
1069 // incrementing it.
1070 --(this->thread[tid]->funcExeInst);
1071}
1072
1073template <class Impl>
1074void
1075FullO3CPU<Impl>::serialize(std::ostream &os)
1076{
1077 SimObject::State so_state = SimObject::getState();
1078 SERIALIZE_ENUM(so_state);
1079 BaseCPU::serialize(os);
1080 nameOut(os, csprintf("%s.tickEvent", name()));
1081 tickEvent.serialize(os);
1082
1083 // Use SimpleThread's ability to checkpoint to make it easier to
1084 // write out the registers. Also make this static so it doesn't
1085 // get instantiated multiple times (causes a panic in statistics).
1086 static SimpleThread temp;
1087
1088 ThreadID size = thread.size();
1089 for (ThreadID i = 0; i < size; i++) {
1090 nameOut(os, csprintf("%s.xc.%i", name(), i));
1091 temp.copyTC(thread[i]->getTC());
1092 temp.serialize(os);
1093 }
1094}
1095
1096template <class Impl>
1097void
1098FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string &section)
1099{
1100 SimObject::State so_state;
1101 UNSERIALIZE_ENUM(so_state);
1102 BaseCPU::unserialize(cp, section);
1103 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
1104
1105 // Use SimpleThread's ability to checkpoint to make it easier to
1106 // read in the registers. Also make this static so it doesn't
1107 // get instantiated multiple times (causes a panic in statistics).
1108 static SimpleThread temp;
1109
1110 ThreadID size = thread.size();
1111 for (ThreadID i = 0; i < size; i++) {
1112 temp.copyTC(thread[i]->getTC());
1113 temp.unserialize(cp, csprintf("%s.xc.%i", section, i));
1114 thread[i]->getTC()->copyArchRegs(temp.getTC());
1115 }
1116}
1117
1118template <class Impl>
1119unsigned int
1120FullO3CPU<Impl>::drain(Event *drain_event)
1121{
1122 DPRINTF(O3CPU, "Switching out\n");
1123
1124 // If the CPU isn't doing anything, then return immediately.
1125 if (_status == Idle || _status == SwitchedOut) {
1126 return 0;
1127 }
1128
1129 drainCount = 0;
1130 fetch.drain();
1131 decode.drain();
1132 rename.drain();
1133 iew.drain();
1134 commit.drain();
1135
1136 // Wake the CPU and record activity so everything can drain out if
1137 // the CPU was not able to immediately drain.
1138 if (getState() != SimObject::Drained) {
1139 // A bit of a hack...set the drainEvent after all the drain()
1140 // calls have been made, that way if all of the stages drain
1141 // immediately, the signalDrained() function knows not to call
1142 // process on the drain event.
1143 drainEvent = drain_event;
1144
1145 wakeCPU();
1146 activityRec.activity();
1147
1148 return 1;
1149 } else {
1150 return 0;
1151 }
1152}
1153
1154template <class Impl>
1155void
1156FullO3CPU<Impl>::resume()
1157{
1158 fetch.resume();
1159 decode.resume();
1160 rename.resume();
1161 iew.resume();
1162 commit.resume();
1163
1164 changeState(SimObject::Running);
1165
1166 if (_status == SwitchedOut || _status == Idle)
1167 return;
1168
1169 assert(system->getMemoryMode() == Enums::timing);
1170
1171 if (!tickEvent.scheduled())
1172 schedule(tickEvent, nextCycle());
1173 _status = Running;
1174}
1175
1176template <class Impl>
1177void
1178FullO3CPU<Impl>::signalDrained()
1179{
1180 if (++drainCount == NumStages) {
1181 if (tickEvent.scheduled())
1182 tickEvent.squash();
1183
1184 changeState(SimObject::Drained);
1185
1186 BaseCPU::switchOut();
1187
1188 if (drainEvent) {
1189 drainEvent->process();
1190 drainEvent = NULL;
1191 }
1192 }
1193 assert(drainCount <= 5);
1194}
1195
1196template <class Impl>
1197void
1198FullO3CPU<Impl>::switchOut()
1199{
1200 fetch.switchOut();
1201 rename.switchOut();
1202 iew.switchOut();
1203 commit.switchOut();
1204 instList.clear();
1205 while (!removeList.empty()) {
1206 removeList.pop();
1207 }
1208
1209 _status = SwitchedOut;
1210
1211 if (checker)
1212 checker->switchOut();
1213
1214 if (tickEvent.scheduled())
1215 tickEvent.squash();
1216}
1217
1218template <class Impl>
1219void
1220FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1221{
1222 // Flush out any old data from the time buffers.
1223 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1224 timeBuffer.advance();
1225 fetchQueue.advance();
1226 decodeQueue.advance();
1227 renameQueue.advance();
1228 iewQueue.advance();
1229 }
1230
1231 activityRec.reset();
1232
1233 BaseCPU::takeOverFrom(oldCPU);
1234
1235 fetch.takeOverFrom();
1236 decode.takeOverFrom();
1237 rename.takeOverFrom();
1238 iew.takeOverFrom();
1239 commit.takeOverFrom();
1240
1241 assert(!tickEvent.scheduled() || tickEvent.squashed());
1242
1243 // @todo: Figure out how to properly select the tid to put onto
1244 // the active threads list.
1245 ThreadID tid = 0;
1246
1247 list<ThreadID>::iterator isActive =
1248 std::find(activeThreads.begin(), activeThreads.end(), tid);
1249
1250 if (isActive == activeThreads.end()) {
1251 //May Need to Re-code this if the delay variable is the delay
1252 //needed for thread to activate
1253 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n",
1254 tid);
1255
1256 activeThreads.push_back(tid);
1257 }
1258
1259 // Set all statuses to active, schedule the CPU's tick event.
1260 // @todo: Fix up statuses so this is handled properly
1261 ThreadID size = threadContexts.size();
1262 for (ThreadID i = 0; i < size; ++i) {
1263 ThreadContext *tc = threadContexts[i];
1264 if (tc->status() == ThreadContext::Active && _status != Running) {
1265 _status = Running;
1266 reschedule(tickEvent, nextCycle(), true);
1267 }
1268 }
1269 if (!tickEvent.scheduled())
1270 schedule(tickEvent, nextCycle());
1271
1272 lastRunningCycle = curTick();
1273}
1274
1275template <class Impl>
1276TheISA::MiscReg
1277FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1278{
1279 return this->isa[tid].readMiscRegNoEffect(misc_reg);
1280}
1281
1282template <class Impl>
1283TheISA::MiscReg
1284FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1285{
1286 miscRegfileReads++;
1287 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid));
1288}
1289
1290template <class Impl>
1291void
1292FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1293 const TheISA::MiscReg &val, ThreadID tid)
1294{
1295 this->isa[tid].setMiscRegNoEffect(misc_reg, val);
1296}
1297
1298template <class Impl>
1299void
1300FullO3CPU<Impl>::setMiscReg(int misc_reg,
1301 const TheISA::MiscReg &val, ThreadID tid)
1302{
1303 miscRegfileWrites++;
1304 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid));
1305}
1306
1307template <class Impl>
1308uint64_t
1309FullO3CPU<Impl>::readIntReg(int reg_idx)
1310{
1311 intRegfileReads++;
1312 return regFile.readIntReg(reg_idx);
1313}
1314
1315template <class Impl>
1316FloatReg
1317FullO3CPU<Impl>::readFloatReg(int reg_idx)
1318{
1319 fpRegfileReads++;
1320 return regFile.readFloatReg(reg_idx);
1321}
1322
1323template <class Impl>
1324FloatRegBits
1325FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1326{
1327 fpRegfileReads++;
1328 return regFile.readFloatRegBits(reg_idx);
1329}
1330
1331template <class Impl>
1332void
1333FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1334{
1335 intRegfileWrites++;
1336 regFile.setIntReg(reg_idx, val);
1337}
1338
1339template <class Impl>
1340void
1341FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1342{
1343 fpRegfileWrites++;
1344 regFile.setFloatReg(reg_idx, val);
1345}
1346
1347template <class Impl>
1348void
1349FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1350{
1351 fpRegfileWrites++;
1352 regFile.setFloatRegBits(reg_idx, val);
1353}
1354
1355template <class Impl>
1356uint64_t
1357FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1358{
1359 intRegfileReads++;
1360 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1361
1362 return regFile.readIntReg(phys_reg);
1363}
1364
1365template <class Impl>
1366float
1367FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1368{
1369 fpRegfileReads++;
1370 int idx = reg_idx + TheISA::NumIntRegs;
1371 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1372
1373 return regFile.readFloatReg(phys_reg);
1374}
1375
1376template <class Impl>
1377uint64_t
1378FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1379{
1380 fpRegfileReads++;
1381 int idx = reg_idx + TheISA::NumIntRegs;
1382 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1383
1384 return regFile.readFloatRegBits(phys_reg);
1385}
1386
1387template <class Impl>
1388void
1389FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1390{
1391 intRegfileWrites++;
1392 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1393
1394 regFile.setIntReg(phys_reg, val);
1395}
1396
1397template <class Impl>
1398void
1399FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1400{
1401 fpRegfileWrites++;
1402 int idx = reg_idx + TheISA::NumIntRegs;
1403 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1404
1405 regFile.setFloatReg(phys_reg, val);
1406}
1407
1408template <class Impl>
1409void
1410FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1411{
1412 fpRegfileWrites++;
1413 int idx = reg_idx + TheISA::NumIntRegs;
1414 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1415
1416 regFile.setFloatRegBits(phys_reg, val);
1417}
1418
1419template <class Impl>
1420TheISA::PCState
1421FullO3CPU<Impl>::pcState(ThreadID tid)
1422{
1423 return commit.pcState(tid);
1424}
1425
1426template <class Impl>
1427void
1428FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1429{
1430 commit.pcState(val, tid);
1431}
1432
1433template <class Impl>
1434Addr
1435FullO3CPU<Impl>::instAddr(ThreadID tid)
1436{
1437 return commit.instAddr(tid);
1438}
1439
1440template <class Impl>
1441Addr
1442FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1443{
1444 return commit.nextInstAddr(tid);
1445}
1446
1447template <class Impl>
1448MicroPC
1449FullO3CPU<Impl>::microPC(ThreadID tid)
1450{
1451 return commit.microPC(tid);
1452}
1453
1454template <class Impl>
1455void
1456FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1457{
1458 this->thread[tid]->inSyscall = true;
1459 this->commit.generateTCEvent(tid);
1460}
1461
1462template <class Impl>
1463typename FullO3CPU<Impl>::ListIt
1464FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1465{
1466 instList.push_back(inst);
1467
1468 return --(instList.end());
1469}
1470
1471template <class Impl>
1472void
1473FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst)
1474{
1475 // Keep an instruction count.
1476 if (!inst->isMicroop() || inst->isLastMicroop()) {
1477 thread[tid]->numInst++;
1478 thread[tid]->numInsts++;
1479 committedInsts[tid]++;
1480 totalCommittedInsts++;
1481 }
1482 thread[tid]->numOp++;
1483 thread[tid]->numOps++;
1484 committedOps[tid]++;
1485
1486 system->totalNumInsts++;
1487 // Check for instruction-count-based events.
1488 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1489 system->instEventQueue.serviceEvents(system->totalNumInsts);
1490}
1491
1492template <class Impl>
1493void
1494FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1495{
1496 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1497 "[sn:%lli]\n",
1498 inst->threadNumber, inst->pcState(), inst->seqNum);
1499
1500 removeInstsThisCycle = true;
1501
1502 // Remove the front instruction.
1503 removeList.push(inst->getInstListIt());
1504}
1505
1506template <class Impl>
1507void
1508FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1509{
1510 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1511 " list.\n", tid);
1512
1513 ListIt end_it;
1514
1515 bool rob_empty = false;
1516
1517 if (instList.empty()) {
1518 return;
1519 } else if (rob.isEmpty(/*tid*/)) {
1520 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1521 end_it = instList.begin();
1522 rob_empty = true;
1523 } else {
1524 end_it = (rob.readTailInst(tid))->getInstListIt();
1525 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1526 }
1527
1528 removeInstsThisCycle = true;
1529
1530 ListIt inst_it = instList.end();
1531
1532 inst_it--;
1533
1534 // Walk through the instruction list, removing any instructions
1535 // that were inserted after the given instruction iterator, end_it.
1536 while (inst_it != end_it) {
1537 assert(!instList.empty());
1538
1539 squashInstIt(inst_it, tid);
1540
1541 inst_it--;
1542 }
1543
1544 // If the ROB was empty, then we actually need to remove the first
1545 // instruction as well.
1546 if (rob_empty) {
1547 squashInstIt(inst_it, tid);
1548 }
1549}
1550
1551template <class Impl>
1552void
1553FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1554{
1555 assert(!instList.empty());
1556
1557 removeInstsThisCycle = true;
1558
1559 ListIt inst_iter = instList.end();
1560
1561 inst_iter--;
1562
1563 DPRINTF(O3CPU, "Deleting instructions from instruction "
1564 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1565 tid, seq_num, (*inst_iter)->seqNum);
1566
1567 while ((*inst_iter)->seqNum > seq_num) {
1568
1569 bool break_loop = (inst_iter == instList.begin());
1570
1571 squashInstIt(inst_iter, tid);
1572
1573 inst_iter--;
1574
1575 if (break_loop)
1576 break;
1577 }
1578}
1579
1580template <class Impl>
1581inline void
1582FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1583{
1584 if ((*instIt)->threadNumber == tid) {
1585 DPRINTF(O3CPU, "Squashing instruction, "
1586 "[tid:%i] [sn:%lli] PC %s\n",
1587 (*instIt)->threadNumber,
1588 (*instIt)->seqNum,
1589 (*instIt)->pcState());
1590
1591 // Mark it as squashed.
1592 (*instIt)->setSquashed();
1593
1594 // @todo: Formulate a consistent method for deleting
1595 // instructions from the instruction list
1596 // Remove the instruction from the list.
1597 removeList.push(instIt);
1598 }
1599}
1600
1601template <class Impl>
1602void
1603FullO3CPU<Impl>::cleanUpRemovedInsts()
1604{
1605 while (!removeList.empty()) {
1606 DPRINTF(O3CPU, "Removing instruction, "
1607 "[tid:%i] [sn:%lli] PC %s\n",
1608 (*removeList.front())->threadNumber,
1609 (*removeList.front())->seqNum,
1610 (*removeList.front())->pcState());
1611
1612 instList.erase(removeList.front());
1613
1614 removeList.pop();
1615 }
1616
1617 removeInstsThisCycle = false;
1618}
1619/*
1620template <class Impl>
1621void
1622FullO3CPU<Impl>::removeAllInsts()
1623{
1624 instList.clear();
1625}
1626*/
1627template <class Impl>
1628void
1629FullO3CPU<Impl>::dumpInsts()
1630{
1631 int num = 0;
1632
1633 ListIt inst_list_it = instList.begin();
1634
1635 cprintf("Dumping Instruction List\n");
1636
1637 while (inst_list_it != instList.end()) {
1638 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1639 "Squashed:%i\n\n",
1640 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1641 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1642 (*inst_list_it)->isSquashed());
1643 inst_list_it++;
1644 ++num;
1645 }
1646}
1647/*
1648template <class Impl>
1649void
1650FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1651{
1652 iew.wakeDependents(inst);
1653}
1654*/
1655template <class Impl>
1656void
1657FullO3CPU<Impl>::wakeCPU()
1658{
1659 if (activityRec.active() || tickEvent.scheduled()) {
1660 DPRINTF(Activity, "CPU already running.\n");
1661 return;
1662 }
1663
1664 DPRINTF(Activity, "Waking up CPU\n");
1665
1666 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1667 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1668
1669 schedule(tickEvent, nextCycle());
1670}
1671
1672template <class Impl>
1673void
1674FullO3CPU<Impl>::wakeup()
1675{
1676 if (this->thread[0]->status() != ThreadContext::Suspended)
1677 return;
1678
1679 this->wakeCPU();
1680
1681 DPRINTF(Quiesce, "Suspended Processor woken\n");
1682 this->threadContexts[0]->activate();
1683}
1684
1685template <class Impl>
1686ThreadID
1687FullO3CPU<Impl>::getFreeTid()
1688{
1689 for (ThreadID tid = 0; tid < numThreads; tid++) {
1690 if (!tids[tid]) {
1691 tids[tid] = true;
1692 return tid;
1693 }
1694 }
1695
1696 return InvalidThreadID;
1697}
1698
1699template <class Impl>
1700void
1701FullO3CPU<Impl>::doContextSwitch()
1702{
1703 if (contextSwitch) {
1704
1705 //ADD CODE TO DEACTIVE THREAD HERE (???)
1706
1707 ThreadID size = cpuWaitList.size();
1708 for (ThreadID tid = 0; tid < size; tid++) {
1709 activateWhenReady(tid);
1710 }
1711
1712 if (cpuWaitList.size() == 0)
1713 contextSwitch = true;
1714 }
1715}
1716
1717template <class Impl>
1718void
1719FullO3CPU<Impl>::updateThreadPriority()
1720{
1721 if (activeThreads.size() > 1) {
1722 //DEFAULT TO ROUND ROBIN SCHEME
1723 //e.g. Move highest priority to end of thread list
1724 list<ThreadID>::iterator list_begin = activeThreads.begin();
1725
1726 unsigned high_thread = *list_begin;
1727
1728 activeThreads.erase(list_begin);
1729
1730 activeThreads.push_back(high_thread);
1731 }
1732}
1733
1734// Forward declaration of FullO3CPU.
1735template class FullO3CPU<O3CPUImpl>;