cpu.cc revision 3319:1ec49a9bfaa3
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#if FULL_SYSTEM 36#include "cpu/quiesce_event.hh" 37#include "sim/system.hh" 38#else 39#include "sim/process.hh" 40#endif 41 42#include "cpu/activity.hh" 43#include "cpu/simple_thread.hh" 44#include "cpu/thread_context.hh" 45#include "cpu/o3/isa_specific.hh" 46#include "cpu/o3/cpu.hh" 47 48#include "sim/root.hh" 49#include "sim/stat_control.hh" 50 51#if USE_CHECKER 52#include "cpu/checker/cpu.hh" 53#endif 54 55using namespace std; 56using namespace TheISA; 57 58BaseO3CPU::BaseO3CPU(Params *params) 59 : BaseCPU(params), cpu_id(0) 60{ 61} 62 63void 64BaseO3CPU::regStats() 65{ 66 BaseCPU::regStats(); 67} 68 69template <class Impl> 70FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 71 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) 72{ 73} 74 75template <class Impl> 76void 77FullO3CPU<Impl>::TickEvent::process() 78{ 79 cpu->tick(); 80} 81 82template <class Impl> 83const char * 84FullO3CPU<Impl>::TickEvent::description() 85{ 86 return "FullO3CPU tick event"; 87} 88 89template <class Impl> 90FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 91 : Event(&mainEventQueue, CPU_Switch_Pri) 92{ 93} 94 95template <class Impl> 96void 97FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 98 FullO3CPU<Impl> *thread_cpu) 99{ 100 tid = thread_num; 101 cpu = thread_cpu; 102} 103 104template <class Impl> 105void 106FullO3CPU<Impl>::ActivateThreadEvent::process() 107{ 108 cpu->activateThread(tid); 109} 110 111template <class Impl> 112const char * 113FullO3CPU<Impl>::ActivateThreadEvent::description() 114{ 115 return "FullO3CPU \"Activate Thread\" event"; 116} 117 118template <class Impl> 119FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 120 : Event(&mainEventQueue, CPU_Tick_Pri) 121{ 122} 123 124template <class Impl> 125void 126FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 127 FullO3CPU<Impl> *thread_cpu) 128{ 129 tid = thread_num; 130 cpu = thread_cpu; 131} 132 133template <class Impl> 134void 135FullO3CPU<Impl>::DeallocateContextEvent::process() 136{ 137 cpu->deactivateThread(tid); 138 if (remove) 139 cpu->removeThread(tid); 140} 141 142template <class Impl> 143const char * 144FullO3CPU<Impl>::DeallocateContextEvent::description() 145{ 146 return "FullO3CPU \"Deallocate Context\" event"; 147} 148 149template <class Impl> 150FullO3CPU<Impl>::FullO3CPU(Params *params) 151 : BaseO3CPU(params), 152 tickEvent(this), 153 removeInstsThisCycle(false), 154 fetch(params), 155 decode(params), 156 rename(params), 157 iew(params), 158 commit(params), 159 160 regFile(params->numPhysIntRegs, params->numPhysFloatRegs), 161 162 freeList(params->numberOfThreads, 163 TheISA::NumIntRegs, params->numPhysIntRegs, 164 TheISA::NumFloatRegs, params->numPhysFloatRegs), 165 166 rob(params->numROBEntries, params->squashWidth, 167 params->smtROBPolicy, params->smtROBThreshold, 168 params->numberOfThreads), 169 170 scoreboard(params->numberOfThreads, 171 TheISA::NumIntRegs, params->numPhysIntRegs, 172 TheISA::NumFloatRegs, params->numPhysFloatRegs, 173 TheISA::NumMiscRegs * number_of_threads, 174 TheISA::ZeroReg), 175 176 timeBuffer(params->backComSize, params->forwardComSize), 177 fetchQueue(params->backComSize, params->forwardComSize), 178 decodeQueue(params->backComSize, params->forwardComSize), 179 renameQueue(params->backComSize, params->forwardComSize), 180 iewQueue(params->backComSize, params->forwardComSize), 181 activityRec(NumStages, 182 params->backComSize + params->forwardComSize, 183 params->activity), 184 185 globalSeqNum(1), 186#if FULL_SYSTEM 187 system(params->system), 188 physmem(system->physmem), 189#endif // FULL_SYSTEM 190 mem(params->mem), 191 drainCount(0), 192 deferRegistration(params->deferRegistration), 193 numThreads(number_of_threads) 194{ 195 if (!deferRegistration) { 196 _status = Running; 197 } else { 198 _status = Idle; 199 } 200 201 checker = NULL; 202 203 if (params->checker) { 204#if USE_CHECKER 205 BaseCPU *temp_checker = params->checker; 206 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 207 checker->setMemory(mem); 208#if FULL_SYSTEM 209 checker->setSystem(params->system); 210#endif 211#else 212 panic("Checker enabled but not compiled in!"); 213#endif // USE_CHECKER 214 } 215 216#if !FULL_SYSTEM 217 thread.resize(number_of_threads); 218 tids.resize(number_of_threads); 219#endif 220 221 // The stages also need their CPU pointer setup. However this 222 // must be done at the upper level CPU because they have pointers 223 // to the upper level CPU, and not this FullO3CPU. 224 225 // Set up Pointers to the activeThreads list for each stage 226 fetch.setActiveThreads(&activeThreads); 227 decode.setActiveThreads(&activeThreads); 228 rename.setActiveThreads(&activeThreads); 229 iew.setActiveThreads(&activeThreads); 230 commit.setActiveThreads(&activeThreads); 231 232 // Give each of the stages the time buffer they will use. 233 fetch.setTimeBuffer(&timeBuffer); 234 decode.setTimeBuffer(&timeBuffer); 235 rename.setTimeBuffer(&timeBuffer); 236 iew.setTimeBuffer(&timeBuffer); 237 commit.setTimeBuffer(&timeBuffer); 238 239 // Also setup each of the stages' queues. 240 fetch.setFetchQueue(&fetchQueue); 241 decode.setFetchQueue(&fetchQueue); 242 commit.setFetchQueue(&fetchQueue); 243 decode.setDecodeQueue(&decodeQueue); 244 rename.setDecodeQueue(&decodeQueue); 245 rename.setRenameQueue(&renameQueue); 246 iew.setRenameQueue(&renameQueue); 247 iew.setIEWQueue(&iewQueue); 248 commit.setIEWQueue(&iewQueue); 249 commit.setRenameQueue(&renameQueue); 250 251 commit.setIEWStage(&iew); 252 rename.setIEWStage(&iew); 253 rename.setCommitStage(&commit); 254 255#if !FULL_SYSTEM 256 int active_threads = params->workload.size(); 257 258 if (active_threads > Impl::MaxThreads) { 259 panic("Workload Size too large. Increase the 'MaxThreads'" 260 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or " 261 "edit your workload size."); 262 } 263#else 264 int active_threads = 1; 265#endif 266 267 //Make Sure That this a Valid Architeture 268 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 269 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 270 271 rename.setScoreboard(&scoreboard); 272 iew.setScoreboard(&scoreboard); 273 274 // Setup the rename map for whichever stages need it. 275 PhysRegIndex lreg_idx = 0; 276 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 277 278 for (int tid=0; tid < numThreads; tid++) { 279 bool bindRegs = (tid <= active_threads - 1); 280 281 commitRenameMap[tid].init(TheISA::NumIntRegs, 282 params->numPhysIntRegs, 283 lreg_idx, //Index for Logical. Regs 284 285 TheISA::NumFloatRegs, 286 params->numPhysFloatRegs, 287 freg_idx, //Index for Float Regs 288 289 TheISA::NumMiscRegs, 290 291 TheISA::ZeroReg, 292 TheISA::ZeroReg, 293 294 tid, 295 false); 296 297 renameMap[tid].init(TheISA::NumIntRegs, 298 params->numPhysIntRegs, 299 lreg_idx, //Index for Logical. Regs 300 301 TheISA::NumFloatRegs, 302 params->numPhysFloatRegs, 303 freg_idx, //Index for Float Regs 304 305 TheISA::NumMiscRegs, 306 307 TheISA::ZeroReg, 308 TheISA::ZeroReg, 309 310 tid, 311 bindRegs); 312 313 activateThreadEvent[tid].init(tid, this); 314 deallocateContextEvent[tid].init(tid, this); 315 } 316 317 rename.setRenameMap(renameMap); 318 commit.setRenameMap(commitRenameMap); 319 320 // Give renameMap & rename stage access to the freeList; 321 for (int i=0; i < numThreads; i++) { 322 renameMap[i].setFreeList(&freeList); 323 } 324 rename.setFreeList(&freeList); 325 326 // Setup the ROB for whichever stages need it. 327 commit.setROB(&rob); 328 329 lastRunningCycle = curTick; 330 331 lastActivatedCycle = -1; 332 333 // Give renameMap & rename stage access to the freeList; 334 //for (int i=0; i < numThreads; i++) { 335 //globalSeqNum[i] = 1; 336 //} 337 338 contextSwitch = false; 339} 340 341template <class Impl> 342FullO3CPU<Impl>::~FullO3CPU() 343{ 344} 345 346template <class Impl> 347void 348FullO3CPU<Impl>::fullCPURegStats() 349{ 350 BaseO3CPU::regStats(); 351 352 // Register any of the O3CPU's stats here. 353 timesIdled 354 .name(name() + ".timesIdled") 355 .desc("Number of times that the entire CPU went into an idle state and" 356 " unscheduled itself") 357 .prereq(timesIdled); 358 359 idleCycles 360 .name(name() + ".idleCycles") 361 .desc("Total number of cycles that the CPU has spent unscheduled due " 362 "to idling") 363 .prereq(idleCycles); 364 365 // Number of Instructions simulated 366 // -------------------------------- 367 // Should probably be in Base CPU but need templated 368 // MaxThreads so put in here instead 369 committedInsts 370 .init(numThreads) 371 .name(name() + ".committedInsts") 372 .desc("Number of Instructions Simulated"); 373 374 totalCommittedInsts 375 .name(name() + ".committedInsts_total") 376 .desc("Number of Instructions Simulated"); 377 378 cpi 379 .name(name() + ".cpi") 380 .desc("CPI: Cycles Per Instruction") 381 .precision(6); 382 cpi = simTicks / committedInsts; 383 384 totalCpi 385 .name(name() + ".cpi_total") 386 .desc("CPI: Total CPI of All Threads") 387 .precision(6); 388 totalCpi = simTicks / totalCommittedInsts; 389 390 ipc 391 .name(name() + ".ipc") 392 .desc("IPC: Instructions Per Cycle") 393 .precision(6); 394 ipc = committedInsts / simTicks; 395 396 totalIpc 397 .name(name() + ".ipc_total") 398 .desc("IPC: Total IPC of All Threads") 399 .precision(6); 400 totalIpc = totalCommittedInsts / simTicks; 401 402} 403 404template <class Impl> 405Port * 406FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 407{ 408 if (if_name == "dcache_port") 409 return iew.getDcachePort(); 410 else if (if_name == "icache_port") 411 return fetch.getIcachePort(); 412 else 413 panic("No Such Port\n"); 414} 415 416template <class Impl> 417void 418FullO3CPU<Impl>::tick() 419{ 420 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 421 422 ++numCycles; 423 424// activity = false; 425 426 //Tick each of the stages 427 fetch.tick(); 428 429 decode.tick(); 430 431 rename.tick(); 432 433 iew.tick(); 434 435 commit.tick(); 436 437#if !FULL_SYSTEM 438 doContextSwitch(); 439#endif 440 441 // Now advance the time buffers 442 timeBuffer.advance(); 443 444 fetchQueue.advance(); 445 decodeQueue.advance(); 446 renameQueue.advance(); 447 iewQueue.advance(); 448 449 activityRec.advance(); 450 451 if (removeInstsThisCycle) { 452 cleanUpRemovedInsts(); 453 } 454 455 if (!tickEvent.scheduled()) { 456 if (_status == SwitchedOut || 457 getState() == SimObject::Drained) { 458 DPRINTF(O3CPU, "Switched out!\n"); 459 // increment stat 460 lastRunningCycle = curTick; 461 } else if (!activityRec.active() || _status == Idle) { 462 DPRINTF(O3CPU, "Idle!\n"); 463 lastRunningCycle = curTick; 464 timesIdled++; 465 } else { 466 tickEvent.schedule(curTick + cycles(1)); 467 DPRINTF(O3CPU, "Scheduling next tick!\n"); 468 } 469 } 470 471#if !FULL_SYSTEM 472 updateThreadPriority(); 473#endif 474 475} 476 477template <class Impl> 478void 479FullO3CPU<Impl>::init() 480{ 481 if (!deferRegistration) { 482 registerThreadContexts(); 483 } 484 485 // Set inSyscall so that the CPU doesn't squash when initially 486 // setting up registers. 487 for (int i = 0; i < number_of_threads; ++i) 488 thread[i]->inSyscall = true; 489 490 for (int tid=0; tid < number_of_threads; tid++) { 491#if FULL_SYSTEM 492 ThreadContext *src_tc = threadContexts[tid]; 493#else 494 ThreadContext *src_tc = thread[tid]->getTC(); 495#endif 496 // Threads start in the Suspended State 497 if (src_tc->status() != ThreadContext::Suspended) { 498 continue; 499 } 500 501#if FULL_SYSTEM 502 TheISA::initCPU(src_tc, src_tc->readCpuId()); 503#endif 504 } 505 506 // Clear inSyscall. 507 for (int i = 0; i < number_of_threads; ++i) 508 thread[i]->inSyscall = false; 509 510 // Initialize stages. 511 fetch.initStage(); 512 iew.initStage(); 513 rename.initStage(); 514 commit.initStage(); 515 516 commit.setThreads(thread); 517} 518 519template <class Impl> 520void 521FullO3CPU<Impl>::activateThread(unsigned tid) 522{ 523 list<unsigned>::iterator isActive = find( 524 activeThreads.begin(), activeThreads.end(), tid); 525 526 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 527 528 if (isActive == activeThreads.end()) { 529 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 530 tid); 531 532 activeThreads.push_back(tid); 533 } 534} 535 536template <class Impl> 537void 538FullO3CPU<Impl>::deactivateThread(unsigned tid) 539{ 540 //Remove From Active List, if Active 541 list<unsigned>::iterator thread_it = 542 find(activeThreads.begin(), activeThreads.end(), tid); 543 544 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 545 546 if (thread_it != activeThreads.end()) { 547 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 548 tid); 549 activeThreads.erase(thread_it); 550 } 551} 552 553template <class Impl> 554void 555FullO3CPU<Impl>::activateContext(int tid, int delay) 556{ 557 // Needs to set each stage to running as well. 558 if (delay){ 559 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 560 "on cycle %d\n", tid, curTick + cycles(delay)); 561 scheduleActivateThreadEvent(tid, delay); 562 } else { 563 activateThread(tid); 564 } 565 566 if (lastActivatedCycle < curTick) { 567 scheduleTickEvent(delay); 568 569 // Be sure to signal that there's some activity so the CPU doesn't 570 // deschedule itself. 571 activityRec.activity(); 572 fetch.wakeFromQuiesce(); 573 574 lastActivatedCycle = curTick; 575 576 _status = Running; 577 } 578} 579 580template <class Impl> 581bool 582FullO3CPU<Impl>::deallocateContext(int tid, bool remove, int delay) 583{ 584 // Schedule removal of thread data from CPU 585 if (delay){ 586 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 587 "on cycle %d\n", tid, curTick + cycles(delay)); 588 scheduleDeallocateContextEvent(tid, remove, delay); 589 return false; 590 } else { 591 deactivateThread(tid); 592 if (remove) 593 removeThread(tid); 594 return true; 595 } 596} 597 598template <class Impl> 599void 600FullO3CPU<Impl>::suspendContext(int tid) 601{ 602 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 603 bool deallocated = deallocateContext(tid, false, 1); 604 // If this was the last thread then unschedule the tick event. 605 if ((activeThreads.size() == 1 && !deallocated) || activeThreads.size() == 0) 606 unscheduleTickEvent(); 607 _status = Idle; 608} 609 610template <class Impl> 611void 612FullO3CPU<Impl>::haltContext(int tid) 613{ 614 //For now, this is the same as deallocate 615 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 616 deallocateContext(tid, true, 1); 617} 618 619template <class Impl> 620void 621FullO3CPU<Impl>::insertThread(unsigned tid) 622{ 623 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 624 // Will change now that the PC and thread state is internal to the CPU 625 // and not in the ThreadContext. 626#if FULL_SYSTEM 627 ThreadContext *src_tc = system->threadContexts[tid]; 628#else 629 ThreadContext *src_tc = tcBase(tid); 630#endif 631 632 //Bind Int Regs to Rename Map 633 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 634 PhysRegIndex phys_reg = freeList.getIntReg(); 635 636 renameMap[tid].setEntry(ireg,phys_reg); 637 scoreboard.setReg(phys_reg); 638 } 639 640 //Bind Float Regs to Rename Map 641 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 642 PhysRegIndex phys_reg = freeList.getFloatReg(); 643 644 renameMap[tid].setEntry(freg,phys_reg); 645 scoreboard.setReg(phys_reg); 646 } 647 648 //Copy Thread Data Into RegFile 649 //this->copyFromTC(tid); 650 651 //Set PC/NPC/NNPC 652 setPC(src_tc->readPC(), tid); 653 setNextPC(src_tc->readNextPC(), tid); 654#if ISA_HAS_DELAY_SLOT 655 setNextNPC(src_tc->readNextNPC(), tid); 656#endif 657 658 src_tc->setStatus(ThreadContext::Active); 659 660 activateContext(tid,1); 661 662 //Reset ROB/IQ/LSQ Entries 663 commit.rob->resetEntries(); 664 iew.resetEntries(); 665} 666 667template <class Impl> 668void 669FullO3CPU<Impl>::removeThread(unsigned tid) 670{ 671 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 672 673 // Copy Thread Data From RegFile 674 // If thread is suspended, it might be re-allocated 675 //this->copyToTC(tid); 676 677 // Unbind Int Regs from Rename Map 678 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 679 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 680 681 scoreboard.unsetReg(phys_reg); 682 freeList.addReg(phys_reg); 683 } 684 685 // Unbind Float Regs from Rename Map 686 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 687 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 688 689 scoreboard.unsetReg(phys_reg); 690 freeList.addReg(phys_reg); 691 } 692 693 // Squash Throughout Pipeline 694 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; 695 fetch.squash(0, squash_seq_num, true, tid); 696 decode.squash(tid); 697 rename.squash(squash_seq_num, tid); 698 iew.squash(tid); 699 commit.rob->squash(squash_seq_num, tid); 700 701 assert(iew.ldstQueue.getCount(tid) == 0); 702 703 // Reset ROB/IQ/LSQ Entries 704 705 // Commented out for now. This should be possible to do by 706 // telling all the pipeline stages to drain first, and then 707 // checking until the drain completes. Once the pipeline is 708 // drained, call resetEntries(). - 10-09-06 ktlim 709/* 710 if (activeThreads.size() >= 1) { 711 commit.rob->resetEntries(); 712 iew.resetEntries(); 713 } 714*/ 715} 716 717 718template <class Impl> 719void 720FullO3CPU<Impl>::activateWhenReady(int tid) 721{ 722 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 723 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 724 tid); 725 726 bool ready = true; 727 728 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 729 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 730 "Phys. Int. Regs.\n", 731 tid); 732 ready = false; 733 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 734 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 735 "Phys. Float. Regs.\n", 736 tid); 737 ready = false; 738 } else if (commit.rob->numFreeEntries() >= 739 commit.rob->entryAmount(activeThreads.size() + 1)) { 740 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 741 "ROB entries.\n", 742 tid); 743 ready = false; 744 } else if (iew.instQueue.numFreeEntries() >= 745 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 746 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 747 "IQ entries.\n", 748 tid); 749 ready = false; 750 } else if (iew.ldstQueue.numFreeEntries() >= 751 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 752 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 753 "LSQ entries.\n", 754 tid); 755 ready = false; 756 } 757 758 if (ready) { 759 insertThread(tid); 760 761 contextSwitch = false; 762 763 cpuWaitList.remove(tid); 764 } else { 765 suspendContext(tid); 766 767 //blocks fetch 768 contextSwitch = true; 769 770 //@todo: dont always add to waitlist 771 //do waitlist 772 cpuWaitList.push_back(tid); 773 } 774} 775 776template <class Impl> 777void 778FullO3CPU<Impl>::serialize(std::ostream &os) 779{ 780 SimObject::State so_state = SimObject::getState(); 781 SERIALIZE_ENUM(so_state); 782 BaseCPU::serialize(os); 783 nameOut(os, csprintf("%s.tickEvent", name())); 784 tickEvent.serialize(os); 785 786 // Use SimpleThread's ability to checkpoint to make it easier to 787 // write out the registers. Also make this static so it doesn't 788 // get instantiated multiple times (causes a panic in statistics). 789 static SimpleThread temp; 790 791 for (int i = 0; i < thread.size(); i++) { 792 nameOut(os, csprintf("%s.xc.%i", name(), i)); 793 temp.copyTC(thread[i]->getTC()); 794 temp.serialize(os); 795 } 796} 797 798template <class Impl> 799void 800FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 801{ 802 SimObject::State so_state; 803 UNSERIALIZE_ENUM(so_state); 804 BaseCPU::unserialize(cp, section); 805 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 806 807 // Use SimpleThread's ability to checkpoint to make it easier to 808 // read in the registers. Also make this static so it doesn't 809 // get instantiated multiple times (causes a panic in statistics). 810 static SimpleThread temp; 811 812 for (int i = 0; i < thread.size(); i++) { 813 temp.copyTC(thread[i]->getTC()); 814 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 815 thread[i]->getTC()->copyArchRegs(temp.getTC()); 816 } 817} 818 819template <class Impl> 820unsigned int 821FullO3CPU<Impl>::drain(Event *drain_event) 822{ 823 DPRINTF(O3CPU, "Switching out\n"); 824 drainCount = 0; 825 fetch.drain(); 826 decode.drain(); 827 rename.drain(); 828 iew.drain(); 829 commit.drain(); 830 831 // Wake the CPU and record activity so everything can drain out if 832 // the CPU was not able to immediately drain. 833 if (getState() != SimObject::Drained) { 834 // A bit of a hack...set the drainEvent after all the drain() 835 // calls have been made, that way if all of the stages drain 836 // immediately, the signalDrained() function knows not to call 837 // process on the drain event. 838 drainEvent = drain_event; 839 840 wakeCPU(); 841 activityRec.activity(); 842 843 return 1; 844 } else { 845 return 0; 846 } 847} 848 849template <class Impl> 850void 851FullO3CPU<Impl>::resume() 852{ 853 fetch.resume(); 854 decode.resume(); 855 rename.resume(); 856 iew.resume(); 857 commit.resume(); 858 859 changeState(SimObject::Running); 860 861 if (_status == SwitchedOut || _status == Idle) 862 return; 863 864#if FULL_SYSTEM 865 assert(system->getMemoryMode() == System::Timing); 866#endif 867 868 if (!tickEvent.scheduled()) 869 tickEvent.schedule(curTick); 870 _status = Running; 871} 872 873template <class Impl> 874void 875FullO3CPU<Impl>::signalDrained() 876{ 877 if (++drainCount == NumStages) { 878 if (tickEvent.scheduled()) 879 tickEvent.squash(); 880 881 changeState(SimObject::Drained); 882 883 BaseCPU::switchOut(); 884 885 if (drainEvent) { 886 drainEvent->process(); 887 drainEvent = NULL; 888 } 889 } 890 assert(drainCount <= 5); 891} 892 893template <class Impl> 894void 895FullO3CPU<Impl>::switchOut() 896{ 897 fetch.switchOut(); 898 rename.switchOut(); 899 iew.switchOut(); 900 commit.switchOut(); 901 instList.clear(); 902 while (!removeList.empty()) { 903 removeList.pop(); 904 } 905 906 _status = SwitchedOut; 907#if USE_CHECKER 908 if (checker) 909 checker->switchOut(); 910#endif 911 if (tickEvent.scheduled()) 912 tickEvent.squash(); 913} 914 915template <class Impl> 916void 917FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 918{ 919 // Flush out any old data from the time buffers. 920 for (int i = 0; i < timeBuffer.getSize(); ++i) { 921 timeBuffer.advance(); 922 fetchQueue.advance(); 923 decodeQueue.advance(); 924 renameQueue.advance(); 925 iewQueue.advance(); 926 } 927 928 activityRec.reset(); 929 930 BaseCPU::takeOverFrom(oldCPU); 931 932 fetch.takeOverFrom(); 933 decode.takeOverFrom(); 934 rename.takeOverFrom(); 935 iew.takeOverFrom(); 936 commit.takeOverFrom(); 937 938 assert(!tickEvent.scheduled()); 939 940 // @todo: Figure out how to properly select the tid to put onto 941 // the active threads list. 942 int tid = 0; 943 944 list<unsigned>::iterator isActive = find( 945 activeThreads.begin(), activeThreads.end(), tid); 946 947 if (isActive == activeThreads.end()) { 948 //May Need to Re-code this if the delay variable is the delay 949 //needed for thread to activate 950 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 951 tid); 952 953 activeThreads.push_back(tid); 954 } 955 956 // Set all statuses to active, schedule the CPU's tick event. 957 // @todo: Fix up statuses so this is handled properly 958 for (int i = 0; i < threadContexts.size(); ++i) { 959 ThreadContext *tc = threadContexts[i]; 960 if (tc->status() == ThreadContext::Active && _status != Running) { 961 _status = Running; 962 tickEvent.schedule(curTick); 963 } 964 } 965 if (!tickEvent.scheduled()) 966 tickEvent.schedule(curTick); 967 968 Port *peer; 969 Port *icachePort = fetch.getIcachePort(); 970 if (icachePort->getPeer() == NULL) { 971 peer = oldCPU->getPort("icache_port")->getPeer(); 972 icachePort->setPeer(peer); 973 } else { 974 peer = icachePort->getPeer(); 975 } 976 peer->setPeer(icachePort); 977 978 Port *dcachePort = iew.getDcachePort(); 979 if (dcachePort->getPeer() == NULL) { 980 peer = oldCPU->getPort("dcache_port")->getPeer(); 981 dcachePort->setPeer(peer); 982 } else { 983 peer = dcachePort->getPeer(); 984 } 985 peer->setPeer(dcachePort); 986} 987 988template <class Impl> 989uint64_t 990FullO3CPU<Impl>::readIntReg(int reg_idx) 991{ 992 return regFile.readIntReg(reg_idx); 993} 994 995template <class Impl> 996FloatReg 997FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 998{ 999 return regFile.readFloatReg(reg_idx, width); 1000} 1001 1002template <class Impl> 1003FloatReg 1004FullO3CPU<Impl>::readFloatReg(int reg_idx) 1005{ 1006 return regFile.readFloatReg(reg_idx); 1007} 1008 1009template <class Impl> 1010FloatRegBits 1011FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1012{ 1013 return regFile.readFloatRegBits(reg_idx, width); 1014} 1015 1016template <class Impl> 1017FloatRegBits 1018FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1019{ 1020 return regFile.readFloatRegBits(reg_idx); 1021} 1022 1023template <class Impl> 1024void 1025FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1026{ 1027 regFile.setIntReg(reg_idx, val); 1028} 1029 1030template <class Impl> 1031void 1032FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1033{ 1034 regFile.setFloatReg(reg_idx, val, width); 1035} 1036 1037template <class Impl> 1038void 1039FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1040{ 1041 regFile.setFloatReg(reg_idx, val); 1042} 1043 1044template <class Impl> 1045void 1046FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1047{ 1048 regFile.setFloatRegBits(reg_idx, val, width); 1049} 1050 1051template <class Impl> 1052void 1053FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1054{ 1055 regFile.setFloatRegBits(reg_idx, val); 1056} 1057 1058template <class Impl> 1059uint64_t 1060FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1061{ 1062 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1063 1064 return regFile.readIntReg(phys_reg); 1065} 1066 1067template <class Impl> 1068float 1069FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 1070{ 1071 int idx = reg_idx + TheISA::FP_Base_DepTag; 1072 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1073 1074 return regFile.readFloatReg(phys_reg); 1075} 1076 1077template <class Impl> 1078double 1079FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 1080{ 1081 int idx = reg_idx + TheISA::FP_Base_DepTag; 1082 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1083 1084 return regFile.readFloatReg(phys_reg, 64); 1085} 1086 1087template <class Impl> 1088uint64_t 1089FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1090{ 1091 int idx = reg_idx + TheISA::FP_Base_DepTag; 1092 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1093 1094 return regFile.readFloatRegBits(phys_reg); 1095} 1096 1097template <class Impl> 1098void 1099FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1100{ 1101 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1102 1103 regFile.setIntReg(phys_reg, val); 1104} 1105 1106template <class Impl> 1107void 1108FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 1109{ 1110 int idx = reg_idx + TheISA::FP_Base_DepTag; 1111 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1112 1113 regFile.setFloatReg(phys_reg, val); 1114} 1115 1116template <class Impl> 1117void 1118FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1119{ 1120 int idx = reg_idx + TheISA::FP_Base_DepTag; 1121 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1122 1123 regFile.setFloatReg(phys_reg, val, 64); 1124} 1125 1126template <class Impl> 1127void 1128FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1129{ 1130 int idx = reg_idx + TheISA::FP_Base_DepTag; 1131 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1132 1133 regFile.setFloatRegBits(phys_reg, val); 1134} 1135 1136template <class Impl> 1137uint64_t 1138FullO3CPU<Impl>::readPC(unsigned tid) 1139{ 1140 return commit.readPC(tid); 1141} 1142 1143template <class Impl> 1144void 1145FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1146{ 1147 commit.setPC(new_PC, tid); 1148} 1149 1150template <class Impl> 1151uint64_t 1152FullO3CPU<Impl>::readNextPC(unsigned tid) 1153{ 1154 return commit.readNextPC(tid); 1155} 1156 1157template <class Impl> 1158void 1159FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1160{ 1161 commit.setNextPC(val, tid); 1162} 1163 1164template <class Impl> 1165uint64_t 1166FullO3CPU<Impl>::readNextNPC(unsigned tid) 1167{ 1168 return commit.readNextNPC(tid); 1169} 1170 1171template <class Impl> 1172void 1173FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1174{ 1175 commit.setNextNPC(val, tid); 1176} 1177 1178template <class Impl> 1179typename FullO3CPU<Impl>::ListIt 1180FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1181{ 1182 instList.push_back(inst); 1183 1184 return --(instList.end()); 1185} 1186 1187template <class Impl> 1188void 1189FullO3CPU<Impl>::instDone(unsigned tid) 1190{ 1191 // Keep an instruction count. 1192 thread[tid]->numInst++; 1193 thread[tid]->numInsts++; 1194 committedInsts[tid]++; 1195 totalCommittedInsts++; 1196 1197 // Check for instruction-count-based events. 1198 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1199} 1200 1201template <class Impl> 1202void 1203FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1204{ 1205 removeInstsThisCycle = true; 1206 1207 removeList.push(inst->getInstListIt()); 1208} 1209 1210template <class Impl> 1211void 1212FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1213{ 1214 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1215 "[sn:%lli]\n", 1216 inst->threadNumber, inst->readPC(), inst->seqNum); 1217 1218 removeInstsThisCycle = true; 1219 1220 // Remove the front instruction. 1221 removeList.push(inst->getInstListIt()); 1222} 1223 1224template <class Impl> 1225void 1226FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid, 1227 bool squash_delay_slot, 1228 const InstSeqNum &delay_slot_seq_num) 1229{ 1230 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1231 " list.\n", tid); 1232 1233 ListIt end_it; 1234 1235 bool rob_empty = false; 1236 1237 if (instList.empty()) { 1238 return; 1239 } else if (rob.isEmpty(/*tid*/)) { 1240 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1241 end_it = instList.begin(); 1242 rob_empty = true; 1243 } else { 1244 end_it = (rob.readTailInst(tid))->getInstListIt(); 1245 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1246 } 1247 1248 removeInstsThisCycle = true; 1249 1250 ListIt inst_it = instList.end(); 1251 1252 inst_it--; 1253 1254 // Walk through the instruction list, removing any instructions 1255 // that were inserted after the given instruction iterator, end_it. 1256 while (inst_it != end_it) { 1257 assert(!instList.empty()); 1258 1259#if ISA_HAS_DELAY_SLOT 1260 if(!squash_delay_slot && 1261 delay_slot_seq_num >= (*inst_it)->seqNum) { 1262 break; 1263 } 1264#endif 1265 squashInstIt(inst_it, tid); 1266 1267 inst_it--; 1268 } 1269 1270 // If the ROB was empty, then we actually need to remove the first 1271 // instruction as well. 1272 if (rob_empty) { 1273 squashInstIt(inst_it, tid); 1274 } 1275} 1276 1277template <class Impl> 1278void 1279FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1280 unsigned tid) 1281{ 1282 assert(!instList.empty()); 1283 1284 removeInstsThisCycle = true; 1285 1286 ListIt inst_iter = instList.end(); 1287 1288 inst_iter--; 1289 1290 DPRINTF(O3CPU, "Deleting instructions from instruction " 1291 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1292 tid, seq_num, (*inst_iter)->seqNum); 1293 1294 while ((*inst_iter)->seqNum > seq_num) { 1295 1296 bool break_loop = (inst_iter == instList.begin()); 1297 1298 squashInstIt(inst_iter, tid); 1299 1300 inst_iter--; 1301 1302 if (break_loop) 1303 break; 1304 } 1305} 1306 1307template <class Impl> 1308inline void 1309FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1310{ 1311 if ((*instIt)->threadNumber == tid) { 1312 DPRINTF(O3CPU, "Squashing instruction, " 1313 "[tid:%i] [sn:%lli] PC %#x\n", 1314 (*instIt)->threadNumber, 1315 (*instIt)->seqNum, 1316 (*instIt)->readPC()); 1317 1318 // Mark it as squashed. 1319 (*instIt)->setSquashed(); 1320 1321 // @todo: Formulate a consistent method for deleting 1322 // instructions from the instruction list 1323 // Remove the instruction from the list. 1324 removeList.push(instIt); 1325 } 1326} 1327 1328template <class Impl> 1329void 1330FullO3CPU<Impl>::cleanUpRemovedInsts() 1331{ 1332 while (!removeList.empty()) { 1333 DPRINTF(O3CPU, "Removing instruction, " 1334 "[tid:%i] [sn:%lli] PC %#x\n", 1335 (*removeList.front())->threadNumber, 1336 (*removeList.front())->seqNum, 1337 (*removeList.front())->readPC()); 1338 1339 instList.erase(removeList.front()); 1340 1341 removeList.pop(); 1342 } 1343 1344 removeInstsThisCycle = false; 1345} 1346/* 1347template <class Impl> 1348void 1349FullO3CPU<Impl>::removeAllInsts() 1350{ 1351 instList.clear(); 1352} 1353*/ 1354template <class Impl> 1355void 1356FullO3CPU<Impl>::dumpInsts() 1357{ 1358 int num = 0; 1359 1360 ListIt inst_list_it = instList.begin(); 1361 1362 cprintf("Dumping Instruction List\n"); 1363 1364 while (inst_list_it != instList.end()) { 1365 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1366 "Squashed:%i\n\n", 1367 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1368 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1369 (*inst_list_it)->isSquashed()); 1370 inst_list_it++; 1371 ++num; 1372 } 1373} 1374/* 1375template <class Impl> 1376void 1377FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1378{ 1379 iew.wakeDependents(inst); 1380} 1381*/ 1382template <class Impl> 1383void 1384FullO3CPU<Impl>::wakeCPU() 1385{ 1386 if (activityRec.active() || tickEvent.scheduled()) { 1387 DPRINTF(Activity, "CPU already running.\n"); 1388 return; 1389 } 1390 1391 DPRINTF(Activity, "Waking up CPU\n"); 1392 1393 idleCycles += (curTick - 1) - lastRunningCycle; 1394 1395 tickEvent.schedule(curTick); 1396} 1397 1398template <class Impl> 1399int 1400FullO3CPU<Impl>::getFreeTid() 1401{ 1402 for (int i=0; i < numThreads; i++) { 1403 if (!tids[i]) { 1404 tids[i] = true; 1405 return i; 1406 } 1407 } 1408 1409 return -1; 1410} 1411 1412template <class Impl> 1413void 1414FullO3CPU<Impl>::doContextSwitch() 1415{ 1416 if (contextSwitch) { 1417 1418 //ADD CODE TO DEACTIVE THREAD HERE (???) 1419 1420 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1421 activateWhenReady(tid); 1422 } 1423 1424 if (cpuWaitList.size() == 0) 1425 contextSwitch = true; 1426 } 1427} 1428 1429template <class Impl> 1430void 1431FullO3CPU<Impl>::updateThreadPriority() 1432{ 1433 if (activeThreads.size() > 1) 1434 { 1435 //DEFAULT TO ROUND ROBIN SCHEME 1436 //e.g. Move highest priority to end of thread list 1437 list<unsigned>::iterator list_begin = activeThreads.begin(); 1438 list<unsigned>::iterator list_end = activeThreads.end(); 1439 1440 unsigned high_thread = *list_begin; 1441 1442 activeThreads.erase(list_begin); 1443 1444 activeThreads.push_back(high_thread); 1445 } 1446} 1447 1448// Forward declaration of FullO3CPU. 1449template class FullO3CPU<O3CPUImpl>; 1450