cpu.cc revision 3229
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#if FULL_SYSTEM 854 assert(system->getMemoryMode() == System::Timing); 855#endif 856 fetch.resume(); 857 decode.resume(); 858 rename.resume(); 859 iew.resume(); 860 commit.resume(); 861 862 changeState(SimObject::Running); 863 864 if (_status == SwitchedOut || _status == Idle) 865 return; 866 867 if (!tickEvent.scheduled()) 868 tickEvent.schedule(curTick); 869 _status = Running; 870} 871 872template <class Impl> 873void 874FullO3CPU<Impl>::signalDrained() 875{ 876 if (++drainCount == NumStages) { 877 if (tickEvent.scheduled()) 878 tickEvent.squash(); 879 880 changeState(SimObject::Drained); 881 882 BaseCPU::switchOut(); 883 884 if (drainEvent) { 885 drainEvent->process(); 886 drainEvent = NULL; 887 } 888 } 889 assert(drainCount <= 5); 890} 891 892template <class Impl> 893void 894FullO3CPU<Impl>::switchOut() 895{ 896 fetch.switchOut(); 897 rename.switchOut(); 898 iew.switchOut(); 899 commit.switchOut(); 900 instList.clear(); 901 while (!removeList.empty()) { 902 removeList.pop(); 903 } 904 905 _status = SwitchedOut; 906#if USE_CHECKER 907 if (checker) 908 checker->switchOut(); 909#endif 910 if (tickEvent.scheduled()) 911 tickEvent.squash(); 912} 913 914template <class Impl> 915void 916FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 917{ 918 // Flush out any old data from the time buffers. 919 for (int i = 0; i < timeBuffer.getSize(); ++i) { 920 timeBuffer.advance(); 921 fetchQueue.advance(); 922 decodeQueue.advance(); 923 renameQueue.advance(); 924 iewQueue.advance(); 925 } 926 927 activityRec.reset(); 928 929 BaseCPU::takeOverFrom(oldCPU); 930 931 fetch.takeOverFrom(); 932 decode.takeOverFrom(); 933 rename.takeOverFrom(); 934 iew.takeOverFrom(); 935 commit.takeOverFrom(); 936 937 assert(!tickEvent.scheduled()); 938 939 // @todo: Figure out how to properly select the tid to put onto 940 // the active threads list. 941 int tid = 0; 942 943 list<unsigned>::iterator isActive = find( 944 activeThreads.begin(), activeThreads.end(), tid); 945 946 if (isActive == activeThreads.end()) { 947 //May Need to Re-code this if the delay variable is the delay 948 //needed for thread to activate 949 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 950 tid); 951 952 activeThreads.push_back(tid); 953 } 954 955 // Set all statuses to active, schedule the CPU's tick event. 956 // @todo: Fix up statuses so this is handled properly 957 for (int i = 0; i < threadContexts.size(); ++i) { 958 ThreadContext *tc = threadContexts[i]; 959 if (tc->status() == ThreadContext::Active && _status != Running) { 960 _status = Running; 961 tickEvent.schedule(curTick); 962 } 963 } 964 if (!tickEvent.scheduled()) 965 tickEvent.schedule(curTick); 966 967 Port *peer; 968 Port *icachePort = fetch.getIcachePort(); 969 if (icachePort->getPeer() == NULL) { 970 peer = oldCPU->getPort("icache_port")->getPeer(); 971 icachePort->setPeer(peer); 972 } else { 973 peer = icachePort->getPeer(); 974 } 975 peer->setPeer(icachePort); 976 977 Port *dcachePort = iew.getDcachePort(); 978 if (dcachePort->getPeer() == NULL) { 979 peer = oldCPU->getPort("dcache_port")->getPeer(); 980 dcachePort->setPeer(peer); 981 } else { 982 peer = dcachePort->getPeer(); 983 } 984 peer->setPeer(dcachePort); 985} 986 987template <class Impl> 988uint64_t 989FullO3CPU<Impl>::readIntReg(int reg_idx) 990{ 991 return regFile.readIntReg(reg_idx); 992} 993 994template <class Impl> 995FloatReg 996FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 997{ 998 return regFile.readFloatReg(reg_idx, width); 999} 1000 1001template <class Impl> 1002FloatReg 1003FullO3CPU<Impl>::readFloatReg(int reg_idx) 1004{ 1005 return regFile.readFloatReg(reg_idx); 1006} 1007 1008template <class Impl> 1009FloatRegBits 1010FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1011{ 1012 return regFile.readFloatRegBits(reg_idx, width); 1013} 1014 1015template <class Impl> 1016FloatRegBits 1017FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1018{ 1019 return regFile.readFloatRegBits(reg_idx); 1020} 1021 1022template <class Impl> 1023void 1024FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1025{ 1026 regFile.setIntReg(reg_idx, val); 1027} 1028 1029template <class Impl> 1030void 1031FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1032{ 1033 regFile.setFloatReg(reg_idx, val, width); 1034} 1035 1036template <class Impl> 1037void 1038FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1039{ 1040 regFile.setFloatReg(reg_idx, val); 1041} 1042 1043template <class Impl> 1044void 1045FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1046{ 1047 regFile.setFloatRegBits(reg_idx, val, width); 1048} 1049 1050template <class Impl> 1051void 1052FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1053{ 1054 regFile.setFloatRegBits(reg_idx, val); 1055} 1056 1057template <class Impl> 1058uint64_t 1059FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1060{ 1061 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1062 1063 return regFile.readIntReg(phys_reg); 1064} 1065 1066template <class Impl> 1067float 1068FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 1069{ 1070 int idx = reg_idx + TheISA::FP_Base_DepTag; 1071 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1072 1073 return regFile.readFloatReg(phys_reg); 1074} 1075 1076template <class Impl> 1077double 1078FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 1079{ 1080 int idx = reg_idx + TheISA::FP_Base_DepTag; 1081 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1082 1083 return regFile.readFloatReg(phys_reg, 64); 1084} 1085 1086template <class Impl> 1087uint64_t 1088FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1089{ 1090 int idx = reg_idx + TheISA::FP_Base_DepTag; 1091 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1092 1093 return regFile.readFloatRegBits(phys_reg); 1094} 1095 1096template <class Impl> 1097void 1098FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1099{ 1100 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1101 1102 regFile.setIntReg(phys_reg, val); 1103} 1104 1105template <class Impl> 1106void 1107FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 1108{ 1109 int idx = reg_idx + TheISA::FP_Base_DepTag; 1110 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1111 1112 regFile.setFloatReg(phys_reg, val); 1113} 1114 1115template <class Impl> 1116void 1117FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1118{ 1119 int idx = reg_idx + TheISA::FP_Base_DepTag; 1120 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1121 1122 regFile.setFloatReg(phys_reg, val, 64); 1123} 1124 1125template <class Impl> 1126void 1127FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1128{ 1129 int idx = reg_idx + TheISA::FP_Base_DepTag; 1130 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1131 1132 regFile.setFloatRegBits(phys_reg, val); 1133} 1134 1135template <class Impl> 1136uint64_t 1137FullO3CPU<Impl>::readPC(unsigned tid) 1138{ 1139 return commit.readPC(tid); 1140} 1141 1142template <class Impl> 1143void 1144FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1145{ 1146 commit.setPC(new_PC, tid); 1147} 1148 1149template <class Impl> 1150uint64_t 1151FullO3CPU<Impl>::readNextPC(unsigned tid) 1152{ 1153 return commit.readNextPC(tid); 1154} 1155 1156template <class Impl> 1157void 1158FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1159{ 1160 commit.setNextPC(val, tid); 1161} 1162 1163template <class Impl> 1164uint64_t 1165FullO3CPU<Impl>::readNextNPC(unsigned tid) 1166{ 1167 return commit.readNextNPC(tid); 1168} 1169 1170template <class Impl> 1171void 1172FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1173{ 1174 commit.setNextNPC(val, tid); 1175} 1176 1177template <class Impl> 1178typename FullO3CPU<Impl>::ListIt 1179FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1180{ 1181 instList.push_back(inst); 1182 1183 return --(instList.end()); 1184} 1185 1186template <class Impl> 1187void 1188FullO3CPU<Impl>::instDone(unsigned tid) 1189{ 1190 // Keep an instruction count. 1191 thread[tid]->numInst++; 1192 thread[tid]->numInsts++; 1193 committedInsts[tid]++; 1194 totalCommittedInsts++; 1195 1196 // Check for instruction-count-based events. 1197 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1198} 1199 1200template <class Impl> 1201void 1202FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1203{ 1204 removeInstsThisCycle = true; 1205 1206 removeList.push(inst->getInstListIt()); 1207} 1208 1209template <class Impl> 1210void 1211FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1212{ 1213 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1214 "[sn:%lli]\n", 1215 inst->threadNumber, inst->readPC(), inst->seqNum); 1216 1217 removeInstsThisCycle = true; 1218 1219 // Remove the front instruction. 1220 removeList.push(inst->getInstListIt()); 1221} 1222 1223template <class Impl> 1224void 1225FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid, 1226 bool squash_delay_slot, 1227 const InstSeqNum &delay_slot_seq_num) 1228{ 1229 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1230 " list.\n", tid); 1231 1232 ListIt end_it; 1233 1234 bool rob_empty = false; 1235 1236 if (instList.empty()) { 1237 return; 1238 } else if (rob.isEmpty(/*tid*/)) { 1239 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1240 end_it = instList.begin(); 1241 rob_empty = true; 1242 } else { 1243 end_it = (rob.readTailInst(tid))->getInstListIt(); 1244 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1245 } 1246 1247 removeInstsThisCycle = true; 1248 1249 ListIt inst_it = instList.end(); 1250 1251 inst_it--; 1252 1253 // Walk through the instruction list, removing any instructions 1254 // that were inserted after the given instruction iterator, end_it. 1255 while (inst_it != end_it) { 1256 assert(!instList.empty()); 1257 1258#if ISA_HAS_DELAY_SLOT 1259 if(!squash_delay_slot && 1260 delay_slot_seq_num >= (*inst_it)->seqNum) { 1261 break; 1262 } 1263#endif 1264 squashInstIt(inst_it, tid); 1265 1266 inst_it--; 1267 } 1268 1269 // If the ROB was empty, then we actually need to remove the first 1270 // instruction as well. 1271 if (rob_empty) { 1272 squashInstIt(inst_it, tid); 1273 } 1274} 1275 1276template <class Impl> 1277void 1278FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1279 unsigned tid) 1280{ 1281 assert(!instList.empty()); 1282 1283 removeInstsThisCycle = true; 1284 1285 ListIt inst_iter = instList.end(); 1286 1287 inst_iter--; 1288 1289 DPRINTF(O3CPU, "Deleting instructions from instruction " 1290 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1291 tid, seq_num, (*inst_iter)->seqNum); 1292 1293 while ((*inst_iter)->seqNum > seq_num) { 1294 1295 bool break_loop = (inst_iter == instList.begin()); 1296 1297 squashInstIt(inst_iter, tid); 1298 1299 inst_iter--; 1300 1301 if (break_loop) 1302 break; 1303 } 1304} 1305 1306template <class Impl> 1307inline void 1308FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1309{ 1310 if ((*instIt)->threadNumber == tid) { 1311 DPRINTF(O3CPU, "Squashing instruction, " 1312 "[tid:%i] [sn:%lli] PC %#x\n", 1313 (*instIt)->threadNumber, 1314 (*instIt)->seqNum, 1315 (*instIt)->readPC()); 1316 1317 // Mark it as squashed. 1318 (*instIt)->setSquashed(); 1319 1320 // @todo: Formulate a consistent method for deleting 1321 // instructions from the instruction list 1322 // Remove the instruction from the list. 1323 removeList.push(instIt); 1324 } 1325} 1326 1327template <class Impl> 1328void 1329FullO3CPU<Impl>::cleanUpRemovedInsts() 1330{ 1331 while (!removeList.empty()) { 1332 DPRINTF(O3CPU, "Removing instruction, " 1333 "[tid:%i] [sn:%lli] PC %#x\n", 1334 (*removeList.front())->threadNumber, 1335 (*removeList.front())->seqNum, 1336 (*removeList.front())->readPC()); 1337 1338 instList.erase(removeList.front()); 1339 1340 removeList.pop(); 1341 } 1342 1343 removeInstsThisCycle = false; 1344} 1345/* 1346template <class Impl> 1347void 1348FullO3CPU<Impl>::removeAllInsts() 1349{ 1350 instList.clear(); 1351} 1352*/ 1353template <class Impl> 1354void 1355FullO3CPU<Impl>::dumpInsts() 1356{ 1357 int num = 0; 1358 1359 ListIt inst_list_it = instList.begin(); 1360 1361 cprintf("Dumping Instruction List\n"); 1362 1363 while (inst_list_it != instList.end()) { 1364 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1365 "Squashed:%i\n\n", 1366 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1367 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1368 (*inst_list_it)->isSquashed()); 1369 inst_list_it++; 1370 ++num; 1371 } 1372} 1373/* 1374template <class Impl> 1375void 1376FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1377{ 1378 iew.wakeDependents(inst); 1379} 1380*/ 1381template <class Impl> 1382void 1383FullO3CPU<Impl>::wakeCPU() 1384{ 1385 if (activityRec.active() || tickEvent.scheduled()) { 1386 DPRINTF(Activity, "CPU already running.\n"); 1387 return; 1388 } 1389 1390 DPRINTF(Activity, "Waking up CPU\n"); 1391 1392 idleCycles += (curTick - 1) - lastRunningCycle; 1393 1394 tickEvent.schedule(curTick); 1395} 1396 1397template <class Impl> 1398int 1399FullO3CPU<Impl>::getFreeTid() 1400{ 1401 for (int i=0; i < numThreads; i++) { 1402 if (!tids[i]) { 1403 tids[i] = true; 1404 return i; 1405 } 1406 } 1407 1408 return -1; 1409} 1410 1411template <class Impl> 1412void 1413FullO3CPU<Impl>::doContextSwitch() 1414{ 1415 if (contextSwitch) { 1416 1417 //ADD CODE TO DEACTIVE THREAD HERE (???) 1418 1419 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1420 activateWhenReady(tid); 1421 } 1422 1423 if (cpuWaitList.size() == 0) 1424 contextSwitch = true; 1425 } 1426} 1427 1428template <class Impl> 1429void 1430FullO3CPU<Impl>::updateThreadPriority() 1431{ 1432 if (activeThreads.size() > 1) 1433 { 1434 //DEFAULT TO ROUND ROBIN SCHEME 1435 //e.g. Move highest priority to end of thread list 1436 list<unsigned>::iterator list_begin = activeThreads.begin(); 1437 list<unsigned>::iterator list_end = activeThreads.end(); 1438 1439 unsigned high_thread = *list_begin; 1440 1441 activeThreads.erase(list_begin); 1442 1443 activeThreads.push_back(high_thread); 1444 } 1445} 1446 1447// Forward declaration of FullO3CPU. 1448template class FullO3CPU<O3CPUImpl>; 1449