X86: Fix argument register indexing.
[gem5.git] / src / cpu / o3 / cpu.cc
1 /*
2 * Copyright (c) 2004-2006 The Regents of The University of Michigan
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Authors: Kevin Lim
29 * Korey Sewell
30 */
31
32 #include "config/full_system.hh"
33 #include "config/use_checker.hh"
34
35 #include "cpu/activity.hh"
36 #include "cpu/simple_thread.hh"
37 #include "cpu/thread_context.hh"
38 #include "cpu/o3/isa_specific.hh"
39 #include "cpu/o3/cpu.hh"
40 #include "enums/MemoryMode.hh"
41 #include "sim/core.hh"
42 #include "sim/stat_control.hh"
43
44 #if FULL_SYSTEM
45 #include "cpu/quiesce_event.hh"
46 #include "sim/system.hh"
47 #else
48 #include "sim/process.hh"
49 #endif
50
51 #if USE_CHECKER
52 #include "cpu/checker/cpu.hh"
53 #endif
54
55 using namespace std;
56 using namespace TheISA;
57
58 BaseO3CPU::BaseO3CPU(Params *params)
59 : BaseCPU(params), cpu_id(0)
60 {
61 }
62
63 void
64 BaseO3CPU::regStats()
65 {
66 BaseCPU::regStats();
67 }
68
69 template <class Impl>
70 FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c)
71 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c)
72 {
73 }
74
75 template <class Impl>
76 void
77 FullO3CPU<Impl>::TickEvent::process()
78 {
79 cpu->tick();
80 }
81
82 template <class Impl>
83 const char *
84 FullO3CPU<Impl>::TickEvent::description()
85 {
86 return "FullO3CPU tick event";
87 }
88
89 template <class Impl>
90 FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent()
91 : Event(&mainEventQueue, CPU_Switch_Pri)
92 {
93 }
94
95 template <class Impl>
96 void
97 FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num,
98 FullO3CPU<Impl> *thread_cpu)
99 {
100 tid = thread_num;
101 cpu = thread_cpu;
102 }
103
104 template <class Impl>
105 void
106 FullO3CPU<Impl>::ActivateThreadEvent::process()
107 {
108 cpu->activateThread(tid);
109 }
110
111 template <class Impl>
112 const char *
113 FullO3CPU<Impl>::ActivateThreadEvent::description()
114 {
115 return "FullO3CPU \"Activate Thread\" event";
116 }
117
118 template <class Impl>
119 FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent()
120 : Event(&mainEventQueue, CPU_Tick_Pri), tid(0), remove(false), cpu(NULL)
121 {
122 }
123
124 template <class Impl>
125 void
126 FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num,
127 FullO3CPU<Impl> *thread_cpu)
128 {
129 tid = thread_num;
130 cpu = thread_cpu;
131 remove = false;
132 }
133
134 template <class Impl>
135 void
136 FullO3CPU<Impl>::DeallocateContextEvent::process()
137 {
138 cpu->deactivateThread(tid);
139 if (remove)
140 cpu->removeThread(tid);
141 }
142
143 template <class Impl>
144 const char *
145 FullO3CPU<Impl>::DeallocateContextEvent::description()
146 {
147 return "FullO3CPU \"Deallocate Context\" event";
148 }
149
150 template <class Impl>
151 FullO3CPU<Impl>::FullO3CPU(O3CPU *o3_cpu, Params *params)
152 : BaseO3CPU(params),
153 #if FULL_SYSTEM
154 itb(params->itb),
155 dtb(params->dtb),
156 #endif
157 tickEvent(this),
158 removeInstsThisCycle(false),
159 fetch(o3_cpu, params),
160 decode(o3_cpu, params),
161 rename(o3_cpu, params),
162 iew(o3_cpu, params),
163 commit(o3_cpu, params),
164
165 regFile(o3_cpu, params->numPhysIntRegs,
166 params->numPhysFloatRegs),
167
168 freeList(params->numberOfThreads,
169 TheISA::NumIntRegs, params->numPhysIntRegs,
170 TheISA::NumFloatRegs, params->numPhysFloatRegs),
171
172 rob(o3_cpu,
173 params->numROBEntries, params->squashWidth,
174 params->smtROBPolicy, params->smtROBThreshold,
175 params->numberOfThreads),
176
177 scoreboard(params->numberOfThreads,
178 TheISA::NumIntRegs, params->numPhysIntRegs,
179 TheISA::NumFloatRegs, params->numPhysFloatRegs,
180 TheISA::NumMiscRegs * number_of_threads,
181 TheISA::ZeroReg),
182
183 timeBuffer(params->backComSize, params->forwardComSize),
184 fetchQueue(params->backComSize, params->forwardComSize),
185 decodeQueue(params->backComSize, params->forwardComSize),
186 renameQueue(params->backComSize, params->forwardComSize),
187 iewQueue(params->backComSize, params->forwardComSize),
188 activityRec(NumStages,
189 params->backComSize + params->forwardComSize,
190 params->activity),
191
192 globalSeqNum(1),
193 #if FULL_SYSTEM
194 system(params->system),
195 physmem(system->physmem),
196 #endif // FULL_SYSTEM
197 drainCount(0),
198 deferRegistration(params->deferRegistration),
199 numThreads(number_of_threads)
200 {
201 if (!deferRegistration) {
202 _status = Running;
203 } else {
204 _status = Idle;
205 }
206
207 #if USE_CHECKER
208 if (params->checker) {
209 BaseCPU *temp_checker = params->checker;
210 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker);
211 #if FULL_SYSTEM
212 checker->setSystem(params->system);
213 #endif
214 } else {
215 checker = NULL;
216 }
217 #endif // USE_CHECKER
218
219 #if !FULL_SYSTEM
220 thread.resize(number_of_threads);
221 tids.resize(number_of_threads);
222 #endif
223
224 // The stages also need their CPU pointer setup. However this
225 // must be done at the upper level CPU because they have pointers
226 // to the upper level CPU, and not this FullO3CPU.
227
228 // Set up Pointers to the activeThreads list for each stage
229 fetch.setActiveThreads(&activeThreads);
230 decode.setActiveThreads(&activeThreads);
231 rename.setActiveThreads(&activeThreads);
232 iew.setActiveThreads(&activeThreads);
233 commit.setActiveThreads(&activeThreads);
234
235 // Give each of the stages the time buffer they will use.
236 fetch.setTimeBuffer(&timeBuffer);
237 decode.setTimeBuffer(&timeBuffer);
238 rename.setTimeBuffer(&timeBuffer);
239 iew.setTimeBuffer(&timeBuffer);
240 commit.setTimeBuffer(&timeBuffer);
241
242 // Also setup each of the stages' queues.
243 fetch.setFetchQueue(&fetchQueue);
244 decode.setFetchQueue(&fetchQueue);
245 commit.setFetchQueue(&fetchQueue);
246 decode.setDecodeQueue(&decodeQueue);
247 rename.setDecodeQueue(&decodeQueue);
248 rename.setRenameQueue(&renameQueue);
249 iew.setRenameQueue(&renameQueue);
250 iew.setIEWQueue(&iewQueue);
251 commit.setIEWQueue(&iewQueue);
252 commit.setRenameQueue(&renameQueue);
253
254 commit.setIEWStage(&iew);
255 rename.setIEWStage(&iew);
256 rename.setCommitStage(&commit);
257
258 #if !FULL_SYSTEM
259 int active_threads = params->workload.size();
260
261 if (active_threads > Impl::MaxThreads) {
262 panic("Workload Size too large. Increase the 'MaxThreads'"
263 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or "
264 "edit your workload size.");
265 }
266 #else
267 int active_threads = 1;
268 #endif
269
270 //Make Sure That this a Valid Architeture
271 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
272 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
273
274 rename.setScoreboard(&scoreboard);
275 iew.setScoreboard(&scoreboard);
276
277 // Setup the rename map for whichever stages need it.
278 PhysRegIndex lreg_idx = 0;
279 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
280
281 for (int tid=0; tid < numThreads; tid++) {
282 bool bindRegs = (tid <= active_threads - 1);
283
284 commitRenameMap[tid].init(TheISA::NumIntRegs,
285 params->numPhysIntRegs,
286 lreg_idx, //Index for Logical. Regs
287
288 TheISA::NumFloatRegs,
289 params->numPhysFloatRegs,
290 freg_idx, //Index for Float Regs
291
292 TheISA::NumMiscRegs,
293
294 TheISA::ZeroReg,
295 TheISA::ZeroReg,
296
297 tid,
298 false);
299
300 renameMap[tid].init(TheISA::NumIntRegs,
301 params->numPhysIntRegs,
302 lreg_idx, //Index for Logical. Regs
303
304 TheISA::NumFloatRegs,
305 params->numPhysFloatRegs,
306 freg_idx, //Index for Float Regs
307
308 TheISA::NumMiscRegs,
309
310 TheISA::ZeroReg,
311 TheISA::ZeroReg,
312
313 tid,
314 bindRegs);
315
316 activateThreadEvent[tid].init(tid, this);
317 deallocateContextEvent[tid].init(tid, this);
318 }
319
320 rename.setRenameMap(renameMap);
321 commit.setRenameMap(commitRenameMap);
322
323 // Give renameMap & rename stage access to the freeList;
324 for (int i=0; i < numThreads; i++) {
325 renameMap[i].setFreeList(&freeList);
326 }
327 rename.setFreeList(&freeList);
328
329 // Setup the ROB for whichever stages need it.
330 commit.setROB(&rob);
331
332 lastRunningCycle = curTick;
333
334 lastActivatedCycle = -1;
335
336 // Give renameMap & rename stage access to the freeList;
337 //for (int i=0; i < numThreads; i++) {
338 //globalSeqNum[i] = 1;
339 //}
340
341 contextSwitch = false;
342 }
343
344 template <class Impl>
345 FullO3CPU<Impl>::~FullO3CPU()
346 {
347 }
348
349 template <class Impl>
350 void
351 FullO3CPU<Impl>::fullCPURegStats()
352 {
353 BaseO3CPU::regStats();
354
355 // Register any of the O3CPU's stats here.
356 timesIdled
357 .name(name() + ".timesIdled")
358 .desc("Number of times that the entire CPU went into an idle state and"
359 " unscheduled itself")
360 .prereq(timesIdled);
361
362 idleCycles
363 .name(name() + ".idleCycles")
364 .desc("Total number of cycles that the CPU has spent unscheduled due "
365 "to idling")
366 .prereq(idleCycles);
367
368 // Number of Instructions simulated
369 // --------------------------------
370 // Should probably be in Base CPU but need templated
371 // MaxThreads so put in here instead
372 committedInsts
373 .init(numThreads)
374 .name(name() + ".committedInsts")
375 .desc("Number of Instructions Simulated");
376
377 totalCommittedInsts
378 .name(name() + ".committedInsts_total")
379 .desc("Number of Instructions Simulated");
380
381 cpi
382 .name(name() + ".cpi")
383 .desc("CPI: Cycles Per Instruction")
384 .precision(6);
385 cpi = numCycles / committedInsts;
386
387 totalCpi
388 .name(name() + ".cpi_total")
389 .desc("CPI: Total CPI of All Threads")
390 .precision(6);
391 totalCpi = numCycles / totalCommittedInsts;
392
393 ipc
394 .name(name() + ".ipc")
395 .desc("IPC: Instructions Per Cycle")
396 .precision(6);
397 ipc = committedInsts / numCycles;
398
399 totalIpc
400 .name(name() + ".ipc_total")
401 .desc("IPC: Total IPC of All Threads")
402 .precision(6);
403 totalIpc = totalCommittedInsts / numCycles;
404
405 }
406
407 template <class Impl>
408 Port *
409 FullO3CPU<Impl>::getPort(const std::string &if_name, int idx)
410 {
411 if (if_name == "dcache_port")
412 return iew.getDcachePort();
413 else if (if_name == "icache_port")
414 return fetch.getIcachePort();
415 else
416 panic("No Such Port\n");
417 }
418
419 template <class Impl>
420 void
421 FullO3CPU<Impl>::tick()
422 {
423 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
424
425 ++numCycles;
426
427 // activity = false;
428
429 //Tick each of the stages
430 fetch.tick();
431
432 decode.tick();
433
434 rename.tick();
435
436 iew.tick();
437
438 commit.tick();
439
440 #if !FULL_SYSTEM
441 doContextSwitch();
442 #endif
443
444 // Now advance the time buffers
445 timeBuffer.advance();
446
447 fetchQueue.advance();
448 decodeQueue.advance();
449 renameQueue.advance();
450 iewQueue.advance();
451
452 activityRec.advance();
453
454 if (removeInstsThisCycle) {
455 cleanUpRemovedInsts();
456 }
457
458 if (!tickEvent.scheduled()) {
459 if (_status == SwitchedOut ||
460 getState() == SimObject::Drained) {
461 DPRINTF(O3CPU, "Switched out!\n");
462 // increment stat
463 lastRunningCycle = curTick;
464 } else if (!activityRec.active() || _status == Idle) {
465 DPRINTF(O3CPU, "Idle!\n");
466 lastRunningCycle = curTick;
467 timesIdled++;
468 } else {
469 tickEvent.schedule(nextCycle(curTick + cycles(1)));
470 DPRINTF(O3CPU, "Scheduling next tick!\n");
471 }
472 }
473
474 #if !FULL_SYSTEM
475 updateThreadPriority();
476 #endif
477
478 }
479
480 template <class Impl>
481 void
482 FullO3CPU<Impl>::init()
483 {
484 if (!deferRegistration) {
485 registerThreadContexts();
486 }
487
488 // Set inSyscall so that the CPU doesn't squash when initially
489 // setting up registers.
490 for (int i = 0; i < number_of_threads; ++i)
491 thread[i]->inSyscall = true;
492
493 for (int tid=0; tid < number_of_threads; tid++) {
494 #if FULL_SYSTEM
495 ThreadContext *src_tc = threadContexts[tid];
496 #else
497 ThreadContext *src_tc = thread[tid]->getTC();
498 #endif
499 // Threads start in the Suspended State
500 if (src_tc->status() != ThreadContext::Suspended) {
501 continue;
502 }
503
504 #if FULL_SYSTEM
505 TheISA::initCPU(src_tc, src_tc->readCpuId());
506 #endif
507 }
508
509 // Clear inSyscall.
510 for (int i = 0; i < number_of_threads; ++i)
511 thread[i]->inSyscall = false;
512
513 // Initialize stages.
514 fetch.initStage();
515 iew.initStage();
516 rename.initStage();
517 commit.initStage();
518
519 commit.setThreads(thread);
520 }
521
522 template <class Impl>
523 void
524 FullO3CPU<Impl>::activateThread(unsigned tid)
525 {
526 list<unsigned>::iterator isActive = find(
527 activeThreads.begin(), activeThreads.end(), tid);
528
529 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid);
530
531 if (isActive == activeThreads.end()) {
532 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
533 tid);
534
535 activeThreads.push_back(tid);
536 }
537 }
538
539 template <class Impl>
540 void
541 FullO3CPU<Impl>::deactivateThread(unsigned tid)
542 {
543 //Remove From Active List, if Active
544 list<unsigned>::iterator thread_it =
545 find(activeThreads.begin(), activeThreads.end(), tid);
546
547 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
548
549 if (thread_it != activeThreads.end()) {
550 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
551 tid);
552 activeThreads.erase(thread_it);
553 }
554 }
555
556 template <class Impl>
557 void
558 FullO3CPU<Impl>::activateContext(int tid, int delay)
559 {
560 // Needs to set each stage to running as well.
561 if (delay){
562 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
563 "on cycle %d\n", tid, curTick + cycles(delay));
564 scheduleActivateThreadEvent(tid, delay);
565 } else {
566 activateThread(tid);
567 }
568
569 if (lastActivatedCycle < curTick) {
570 scheduleTickEvent(delay);
571
572 // Be sure to signal that there's some activity so the CPU doesn't
573 // deschedule itself.
574 activityRec.activity();
575 fetch.wakeFromQuiesce();
576
577 lastActivatedCycle = curTick;
578
579 _status = Running;
580 }
581 }
582
583 template <class Impl>
584 bool
585 FullO3CPU<Impl>::deallocateContext(int tid, bool remove, int delay)
586 {
587 // Schedule removal of thread data from CPU
588 if (delay){
589 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
590 "on cycle %d\n", tid, curTick + cycles(delay));
591 scheduleDeallocateContextEvent(tid, remove, delay);
592 return false;
593 } else {
594 deactivateThread(tid);
595 if (remove)
596 removeThread(tid);
597 return true;
598 }
599 }
600
601 template <class Impl>
602 void
603 FullO3CPU<Impl>::suspendContext(int tid)
604 {
605 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
606 bool deallocated = deallocateContext(tid, false, 1);
607 // If this was the last thread then unschedule the tick event.
608 if (activeThreads.size() == 1 && !deallocated ||
609 activeThreads.size() == 0)
610 unscheduleTickEvent();
611 _status = Idle;
612 }
613
614 template <class Impl>
615 void
616 FullO3CPU<Impl>::haltContext(int tid)
617 {
618 //For now, this is the same as deallocate
619 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
620 deallocateContext(tid, true, 1);
621 }
622
623 template <class Impl>
624 void
625 FullO3CPU<Impl>::insertThread(unsigned tid)
626 {
627 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
628 // Will change now that the PC and thread state is internal to the CPU
629 // and not in the ThreadContext.
630 #if FULL_SYSTEM
631 ThreadContext *src_tc = system->threadContexts[tid];
632 #else
633 ThreadContext *src_tc = tcBase(tid);
634 #endif
635
636 //Bind Int Regs to Rename Map
637 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
638 PhysRegIndex phys_reg = freeList.getIntReg();
639
640 renameMap[tid].setEntry(ireg,phys_reg);
641 scoreboard.setReg(phys_reg);
642 }
643
644 //Bind Float Regs to Rename Map
645 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
646 PhysRegIndex phys_reg = freeList.getFloatReg();
647
648 renameMap[tid].setEntry(freg,phys_reg);
649 scoreboard.setReg(phys_reg);
650 }
651
652 //Copy Thread Data Into RegFile
653 //this->copyFromTC(tid);
654
655 //Set PC/NPC/NNPC
656 setPC(src_tc->readPC(), tid);
657 setNextPC(src_tc->readNextPC(), tid);
658 setNextNPC(src_tc->readNextNPC(), tid);
659
660 src_tc->setStatus(ThreadContext::Active);
661
662 activateContext(tid,1);
663
664 //Reset ROB/IQ/LSQ Entries
665 commit.rob->resetEntries();
666 iew.resetEntries();
667 }
668
669 template <class Impl>
670 void
671 FullO3CPU<Impl>::removeThread(unsigned tid)
672 {
673 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
674
675 // Copy Thread Data From RegFile
676 // If thread is suspended, it might be re-allocated
677 //this->copyToTC(tid);
678
679 // Unbind Int Regs from Rename Map
680 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
681 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
682
683 scoreboard.unsetReg(phys_reg);
684 freeList.addReg(phys_reg);
685 }
686
687 // Unbind Float Regs from Rename Map
688 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
689 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
690
691 scoreboard.unsetReg(phys_reg);
692 freeList.addReg(phys_reg);
693 }
694
695 // Squash Throughout Pipeline
696 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum;
697 fetch.squash(0, sizeof(TheISA::MachInst), 0, squash_seq_num, tid);
698 decode.squash(tid);
699 rename.squash(squash_seq_num, tid);
700 iew.squash(tid);
701 commit.rob->squash(squash_seq_num, tid);
702
703 assert(iew.ldstQueue.getCount(tid) == 0);
704
705 // Reset ROB/IQ/LSQ Entries
706
707 // Commented out for now. This should be possible to do by
708 // telling all the pipeline stages to drain first, and then
709 // checking until the drain completes. Once the pipeline is
710 // drained, call resetEntries(). - 10-09-06 ktlim
711 /*
712 if (activeThreads.size() >= 1) {
713 commit.rob->resetEntries();
714 iew.resetEntries();
715 }
716 */
717 }
718
719
720 template <class Impl>
721 void
722 FullO3CPU<Impl>::activateWhenReady(int tid)
723 {
724 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
725 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
726 tid);
727
728 bool ready = true;
729
730 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
731 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
732 "Phys. Int. Regs.\n",
733 tid);
734 ready = false;
735 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
736 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
737 "Phys. Float. Regs.\n",
738 tid);
739 ready = false;
740 } else if (commit.rob->numFreeEntries() >=
741 commit.rob->entryAmount(activeThreads.size() + 1)) {
742 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
743 "ROB entries.\n",
744 tid);
745 ready = false;
746 } else if (iew.instQueue.numFreeEntries() >=
747 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
748 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
749 "IQ entries.\n",
750 tid);
751 ready = false;
752 } else if (iew.ldstQueue.numFreeEntries() >=
753 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
754 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
755 "LSQ entries.\n",
756 tid);
757 ready = false;
758 }
759
760 if (ready) {
761 insertThread(tid);
762
763 contextSwitch = false;
764
765 cpuWaitList.remove(tid);
766 } else {
767 suspendContext(tid);
768
769 //blocks fetch
770 contextSwitch = true;
771
772 //@todo: dont always add to waitlist
773 //do waitlist
774 cpuWaitList.push_back(tid);
775 }
776 }
777
778 #if FULL_SYSTEM
779 template <class Impl>
780 void
781 FullO3CPU<Impl>::updateMemPorts()
782 {
783 // Update all ThreadContext's memory ports (Functional/Virtual
784 // Ports)
785 for (int i = 0; i < thread.size(); ++i)
786 thread[i]->connectMemPorts();
787 }
788 #endif
789
790 template <class Impl>
791 void
792 FullO3CPU<Impl>::serialize(std::ostream &os)
793 {
794 SimObject::State so_state = SimObject::getState();
795 SERIALIZE_ENUM(so_state);
796 BaseCPU::serialize(os);
797 nameOut(os, csprintf("%s.tickEvent", name()));
798 tickEvent.serialize(os);
799
800 // Use SimpleThread's ability to checkpoint to make it easier to
801 // write out the registers. Also make this static so it doesn't
802 // get instantiated multiple times (causes a panic in statistics).
803 static SimpleThread temp;
804
805 for (int i = 0; i < thread.size(); i++) {
806 nameOut(os, csprintf("%s.xc.%i", name(), i));
807 temp.copyTC(thread[i]->getTC());
808 temp.serialize(os);
809 }
810 }
811
812 template <class Impl>
813 void
814 FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string &section)
815 {
816 SimObject::State so_state;
817 UNSERIALIZE_ENUM(so_state);
818 BaseCPU::unserialize(cp, section);
819 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
820
821 // Use SimpleThread's ability to checkpoint to make it easier to
822 // read in the registers. Also make this static so it doesn't
823 // get instantiated multiple times (causes a panic in statistics).
824 static SimpleThread temp;
825
826 for (int i = 0; i < thread.size(); i++) {
827 temp.copyTC(thread[i]->getTC());
828 temp.unserialize(cp, csprintf("%s.xc.%i", section, i));
829 thread[i]->getTC()->copyArchRegs(temp.getTC());
830 }
831 }
832
833 template <class Impl>
834 unsigned int
835 FullO3CPU<Impl>::drain(Event *drain_event)
836 {
837 DPRINTF(O3CPU, "Switching out\n");
838
839 // If the CPU isn't doing anything, then return immediately.
840 if (_status == Idle || _status == SwitchedOut) {
841 return 0;
842 }
843
844 drainCount = 0;
845 fetch.drain();
846 decode.drain();
847 rename.drain();
848 iew.drain();
849 commit.drain();
850
851 // Wake the CPU and record activity so everything can drain out if
852 // the CPU was not able to immediately drain.
853 if (getState() != SimObject::Drained) {
854 // A bit of a hack...set the drainEvent after all the drain()
855 // calls have been made, that way if all of the stages drain
856 // immediately, the signalDrained() function knows not to call
857 // process on the drain event.
858 drainEvent = drain_event;
859
860 wakeCPU();
861 activityRec.activity();
862
863 return 1;
864 } else {
865 return 0;
866 }
867 }
868
869 template <class Impl>
870 void
871 FullO3CPU<Impl>::resume()
872 {
873 fetch.resume();
874 decode.resume();
875 rename.resume();
876 iew.resume();
877 commit.resume();
878
879 changeState(SimObject::Running);
880
881 if (_status == SwitchedOut || _status == Idle)
882 return;
883
884 #if FULL_SYSTEM
885 assert(system->getMemoryMode() == Enums::timing);
886 #endif
887
888 if (!tickEvent.scheduled())
889 tickEvent.schedule(nextCycle());
890 _status = Running;
891 }
892
893 template <class Impl>
894 void
895 FullO3CPU<Impl>::signalDrained()
896 {
897 if (++drainCount == NumStages) {
898 if (tickEvent.scheduled())
899 tickEvent.squash();
900
901 changeState(SimObject::Drained);
902
903 BaseCPU::switchOut();
904
905 if (drainEvent) {
906 drainEvent->process();
907 drainEvent = NULL;
908 }
909 }
910 assert(drainCount <= 5);
911 }
912
913 template <class Impl>
914 void
915 FullO3CPU<Impl>::switchOut()
916 {
917 fetch.switchOut();
918 rename.switchOut();
919 iew.switchOut();
920 commit.switchOut();
921 instList.clear();
922 while (!removeList.empty()) {
923 removeList.pop();
924 }
925
926 _status = SwitchedOut;
927 #if USE_CHECKER
928 if (checker)
929 checker->switchOut();
930 #endif
931 if (tickEvent.scheduled())
932 tickEvent.squash();
933 }
934
935 template <class Impl>
936 void
937 FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
938 {
939 // Flush out any old data from the time buffers.
940 for (int i = 0; i < timeBuffer.getSize(); ++i) {
941 timeBuffer.advance();
942 fetchQueue.advance();
943 decodeQueue.advance();
944 renameQueue.advance();
945 iewQueue.advance();
946 }
947
948 activityRec.reset();
949
950 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort());
951
952 fetch.takeOverFrom();
953 decode.takeOverFrom();
954 rename.takeOverFrom();
955 iew.takeOverFrom();
956 commit.takeOverFrom();
957
958 assert(!tickEvent.scheduled());
959
960 // @todo: Figure out how to properly select the tid to put onto
961 // the active threads list.
962 int tid = 0;
963
964 list<unsigned>::iterator isActive = find(
965 activeThreads.begin(), activeThreads.end(), tid);
966
967 if (isActive == activeThreads.end()) {
968 //May Need to Re-code this if the delay variable is the delay
969 //needed for thread to activate
970 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n",
971 tid);
972
973 activeThreads.push_back(tid);
974 }
975
976 // Set all statuses to active, schedule the CPU's tick event.
977 // @todo: Fix up statuses so this is handled properly
978 for (int i = 0; i < threadContexts.size(); ++i) {
979 ThreadContext *tc = threadContexts[i];
980 if (tc->status() == ThreadContext::Active && _status != Running) {
981 _status = Running;
982 tickEvent.schedule(nextCycle());
983 }
984 }
985 if (!tickEvent.scheduled())
986 tickEvent.schedule(nextCycle());
987 }
988
989 template <class Impl>
990 uint64_t
991 FullO3CPU<Impl>::readIntReg(int reg_idx)
992 {
993 return regFile.readIntReg(reg_idx);
994 }
995
996 template <class Impl>
997 FloatReg
998 FullO3CPU<Impl>::readFloatReg(int reg_idx, int width)
999 {
1000 return regFile.readFloatReg(reg_idx, width);
1001 }
1002
1003 template <class Impl>
1004 FloatReg
1005 FullO3CPU<Impl>::readFloatReg(int reg_idx)
1006 {
1007 return regFile.readFloatReg(reg_idx);
1008 }
1009
1010 template <class Impl>
1011 FloatRegBits
1012 FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width)
1013 {
1014 return regFile.readFloatRegBits(reg_idx, width);
1015 }
1016
1017 template <class Impl>
1018 FloatRegBits
1019 FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1020 {
1021 return regFile.readFloatRegBits(reg_idx);
1022 }
1023
1024 template <class Impl>
1025 void
1026 FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1027 {
1028 regFile.setIntReg(reg_idx, val);
1029 }
1030
1031 template <class Impl>
1032 void
1033 FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width)
1034 {
1035 regFile.setFloatReg(reg_idx, val, width);
1036 }
1037
1038 template <class Impl>
1039 void
1040 FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1041 {
1042 regFile.setFloatReg(reg_idx, val);
1043 }
1044
1045 template <class Impl>
1046 void
1047 FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width)
1048 {
1049 regFile.setFloatRegBits(reg_idx, val, width);
1050 }
1051
1052 template <class Impl>
1053 void
1054 FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1055 {
1056 regFile.setFloatRegBits(reg_idx, val);
1057 }
1058
1059 template <class Impl>
1060 uint64_t
1061 FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid)
1062 {
1063 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1064
1065 return regFile.readIntReg(phys_reg);
1066 }
1067
1068 template <class Impl>
1069 float
1070 FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid)
1071 {
1072 int idx = reg_idx + TheISA::FP_Base_DepTag;
1073 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1074
1075 return regFile.readFloatReg(phys_reg);
1076 }
1077
1078 template <class Impl>
1079 double
1080 FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid)
1081 {
1082 int idx = reg_idx + TheISA::FP_Base_DepTag;
1083 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1084
1085 return regFile.readFloatReg(phys_reg, 64);
1086 }
1087
1088 template <class Impl>
1089 uint64_t
1090 FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid)
1091 {
1092 int idx = reg_idx + TheISA::FP_Base_DepTag;
1093 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1094
1095 return regFile.readFloatRegBits(phys_reg);
1096 }
1097
1098 template <class Impl>
1099 void
1100 FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid)
1101 {
1102 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1103
1104 regFile.setIntReg(phys_reg, val);
1105 }
1106
1107 template <class Impl>
1108 void
1109 FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid)
1110 {
1111 int idx = reg_idx + TheISA::FP_Base_DepTag;
1112 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1113
1114 regFile.setFloatReg(phys_reg, val);
1115 }
1116
1117 template <class Impl>
1118 void
1119 FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid)
1120 {
1121 int idx = reg_idx + TheISA::FP_Base_DepTag;
1122 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1123
1124 regFile.setFloatReg(phys_reg, val, 64);
1125 }
1126
1127 template <class Impl>
1128 void
1129 FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid)
1130 {
1131 int idx = reg_idx + TheISA::FP_Base_DepTag;
1132 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1133
1134 regFile.setFloatRegBits(phys_reg, val);
1135 }
1136
1137 template <class Impl>
1138 uint64_t
1139 FullO3CPU<Impl>::readPC(unsigned tid)
1140 {
1141 return commit.readPC(tid);
1142 }
1143
1144 template <class Impl>
1145 void
1146 FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid)
1147 {
1148 commit.setPC(new_PC, tid);
1149 }
1150
1151 template <class Impl>
1152 uint64_t
1153 FullO3CPU<Impl>::readMicroPC(unsigned tid)
1154 {
1155 return commit.readMicroPC(tid);
1156 }
1157
1158 template <class Impl>
1159 void
1160 FullO3CPU<Impl>::setMicroPC(Addr new_PC,unsigned tid)
1161 {
1162 commit.setMicroPC(new_PC, tid);
1163 }
1164
1165 template <class Impl>
1166 uint64_t
1167 FullO3CPU<Impl>::readNextPC(unsigned tid)
1168 {
1169 return commit.readNextPC(tid);
1170 }
1171
1172 template <class Impl>
1173 void
1174 FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid)
1175 {
1176 commit.setNextPC(val, tid);
1177 }
1178
1179 template <class Impl>
1180 uint64_t
1181 FullO3CPU<Impl>::readNextNPC(unsigned tid)
1182 {
1183 return commit.readNextNPC(tid);
1184 }
1185
1186 template <class Impl>
1187 void
1188 FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid)
1189 {
1190 commit.setNextNPC(val, tid);
1191 }
1192
1193 template <class Impl>
1194 uint64_t
1195 FullO3CPU<Impl>::readNextMicroPC(unsigned tid)
1196 {
1197 return commit.readNextMicroPC(tid);
1198 }
1199
1200 template <class Impl>
1201 void
1202 FullO3CPU<Impl>::setNextMicroPC(Addr new_PC,unsigned tid)
1203 {
1204 commit.setNextMicroPC(new_PC, tid);
1205 }
1206
1207 template <class Impl>
1208 typename FullO3CPU<Impl>::ListIt
1209 FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1210 {
1211 instList.push_back(inst);
1212
1213 return --(instList.end());
1214 }
1215
1216 template <class Impl>
1217 void
1218 FullO3CPU<Impl>::instDone(unsigned tid)
1219 {
1220 // Keep an instruction count.
1221 thread[tid]->numInst++;
1222 thread[tid]->numInsts++;
1223 committedInsts[tid]++;
1224 totalCommittedInsts++;
1225
1226 // Check for instruction-count-based events.
1227 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1228 }
1229
1230 template <class Impl>
1231 void
1232 FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst)
1233 {
1234 removeInstsThisCycle = true;
1235
1236 removeList.push(inst->getInstListIt());
1237 }
1238
1239 template <class Impl>
1240 void
1241 FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1242 {
1243 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x "
1244 "[sn:%lli]\n",
1245 inst->threadNumber, inst->readPC(), inst->seqNum);
1246
1247 removeInstsThisCycle = true;
1248
1249 // Remove the front instruction.
1250 removeList.push(inst->getInstListIt());
1251 }
1252
1253 template <class Impl>
1254 void
1255 FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid)
1256 {
1257 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1258 " list.\n", tid);
1259
1260 ListIt end_it;
1261
1262 bool rob_empty = false;
1263
1264 if (instList.empty()) {
1265 return;
1266 } else if (rob.isEmpty(/*tid*/)) {
1267 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1268 end_it = instList.begin();
1269 rob_empty = true;
1270 } else {
1271 end_it = (rob.readTailInst(tid))->getInstListIt();
1272 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1273 }
1274
1275 removeInstsThisCycle = true;
1276
1277 ListIt inst_it = instList.end();
1278
1279 inst_it--;
1280
1281 // Walk through the instruction list, removing any instructions
1282 // that were inserted after the given instruction iterator, end_it.
1283 while (inst_it != end_it) {
1284 assert(!instList.empty());
1285
1286 squashInstIt(inst_it, tid);
1287
1288 inst_it--;
1289 }
1290
1291 // If the ROB was empty, then we actually need to remove the first
1292 // instruction as well.
1293 if (rob_empty) {
1294 squashInstIt(inst_it, tid);
1295 }
1296 }
1297
1298 template <class Impl>
1299 void
1300 FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num,
1301 unsigned tid)
1302 {
1303 assert(!instList.empty());
1304
1305 removeInstsThisCycle = true;
1306
1307 ListIt inst_iter = instList.end();
1308
1309 inst_iter--;
1310
1311 DPRINTF(O3CPU, "Deleting instructions from instruction "
1312 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1313 tid, seq_num, (*inst_iter)->seqNum);
1314
1315 while ((*inst_iter)->seqNum > seq_num) {
1316
1317 bool break_loop = (inst_iter == instList.begin());
1318
1319 squashInstIt(inst_iter, tid);
1320
1321 inst_iter--;
1322
1323 if (break_loop)
1324 break;
1325 }
1326 }
1327
1328 template <class Impl>
1329 inline void
1330 FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid)
1331 {
1332 if ((*instIt)->threadNumber == tid) {
1333 DPRINTF(O3CPU, "Squashing instruction, "
1334 "[tid:%i] [sn:%lli] PC %#x\n",
1335 (*instIt)->threadNumber,
1336 (*instIt)->seqNum,
1337 (*instIt)->readPC());
1338
1339 // Mark it as squashed.
1340 (*instIt)->setSquashed();
1341
1342 // @todo: Formulate a consistent method for deleting
1343 // instructions from the instruction list
1344 // Remove the instruction from the list.
1345 removeList.push(instIt);
1346 }
1347 }
1348
1349 template <class Impl>
1350 void
1351 FullO3CPU<Impl>::cleanUpRemovedInsts()
1352 {
1353 while (!removeList.empty()) {
1354 DPRINTF(O3CPU, "Removing instruction, "
1355 "[tid:%i] [sn:%lli] PC %#x\n",
1356 (*removeList.front())->threadNumber,
1357 (*removeList.front())->seqNum,
1358 (*removeList.front())->readPC());
1359
1360 instList.erase(removeList.front());
1361
1362 removeList.pop();
1363 }
1364
1365 removeInstsThisCycle = false;
1366 }
1367 /*
1368 template <class Impl>
1369 void
1370 FullO3CPU<Impl>::removeAllInsts()
1371 {
1372 instList.clear();
1373 }
1374 */
1375 template <class Impl>
1376 void
1377 FullO3CPU<Impl>::dumpInsts()
1378 {
1379 int num = 0;
1380
1381 ListIt inst_list_it = instList.begin();
1382
1383 cprintf("Dumping Instruction List\n");
1384
1385 while (inst_list_it != instList.end()) {
1386 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1387 "Squashed:%i\n\n",
1388 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber,
1389 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1390 (*inst_list_it)->isSquashed());
1391 inst_list_it++;
1392 ++num;
1393 }
1394 }
1395 /*
1396 template <class Impl>
1397 void
1398 FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1399 {
1400 iew.wakeDependents(inst);
1401 }
1402 */
1403 template <class Impl>
1404 void
1405 FullO3CPU<Impl>::wakeCPU()
1406 {
1407 if (activityRec.active() || tickEvent.scheduled()) {
1408 DPRINTF(Activity, "CPU already running.\n");
1409 return;
1410 }
1411
1412 DPRINTF(Activity, "Waking up CPU\n");
1413
1414 idleCycles += (curTick - 1) - lastRunningCycle;
1415
1416 tickEvent.schedule(nextCycle());
1417 }
1418
1419 template <class Impl>
1420 int
1421 FullO3CPU<Impl>::getFreeTid()
1422 {
1423 for (int i=0; i < numThreads; i++) {
1424 if (!tids[i]) {
1425 tids[i] = true;
1426 return i;
1427 }
1428 }
1429
1430 return -1;
1431 }
1432
1433 template <class Impl>
1434 void
1435 FullO3CPU<Impl>::doContextSwitch()
1436 {
1437 if (contextSwitch) {
1438
1439 //ADD CODE TO DEACTIVE THREAD HERE (???)
1440
1441 for (int tid=0; tid < cpuWaitList.size(); tid++) {
1442 activateWhenReady(tid);
1443 }
1444
1445 if (cpuWaitList.size() == 0)
1446 contextSwitch = true;
1447 }
1448 }
1449
1450 template <class Impl>
1451 void
1452 FullO3CPU<Impl>::updateThreadPriority()
1453 {
1454 if (activeThreads.size() > 1)
1455 {
1456 //DEFAULT TO ROUND ROBIN SCHEME
1457 //e.g. Move highest priority to end of thread list
1458 list<unsigned>::iterator list_begin = activeThreads.begin();
1459 list<unsigned>::iterator list_end = activeThreads.end();
1460
1461 unsigned high_thread = *list_begin;
1462
1463 activeThreads.erase(list_begin);
1464
1465 activeThreads.push_back(high_thread);
1466 }
1467 }
1468
1469 // Forward declaration of FullO3CPU.
1470 template class FullO3CPU<O3CPUImpl>;