cpu.cc revision 3686:fa8d8b90cd8a
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 drainCount(0), 191 deferRegistration(params->deferRegistration), 192 numThreads(number_of_threads) 193{ 194 if (!deferRegistration) { 195 _status = Running; 196 } else { 197 _status = Idle; 198 } 199 200 checker = NULL; 201 202 if (params->checker) { 203#if USE_CHECKER 204 BaseCPU *temp_checker = params->checker; 205 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 206#if FULL_SYSTEM 207 checker->setSystem(params->system); 208#endif 209#else 210 panic("Checker enabled but not compiled in!"); 211#endif // USE_CHECKER 212 } 213 214#if !FULL_SYSTEM 215 thread.resize(number_of_threads); 216 tids.resize(number_of_threads); 217#endif 218 219 // The stages also need their CPU pointer setup. However this 220 // must be done at the upper level CPU because they have pointers 221 // to the upper level CPU, and not this FullO3CPU. 222 223 // Set up Pointers to the activeThreads list for each stage 224 fetch.setActiveThreads(&activeThreads); 225 decode.setActiveThreads(&activeThreads); 226 rename.setActiveThreads(&activeThreads); 227 iew.setActiveThreads(&activeThreads); 228 commit.setActiveThreads(&activeThreads); 229 230 // Give each of the stages the time buffer they will use. 231 fetch.setTimeBuffer(&timeBuffer); 232 decode.setTimeBuffer(&timeBuffer); 233 rename.setTimeBuffer(&timeBuffer); 234 iew.setTimeBuffer(&timeBuffer); 235 commit.setTimeBuffer(&timeBuffer); 236 237 // Also setup each of the stages' queues. 238 fetch.setFetchQueue(&fetchQueue); 239 decode.setFetchQueue(&fetchQueue); 240 commit.setFetchQueue(&fetchQueue); 241 decode.setDecodeQueue(&decodeQueue); 242 rename.setDecodeQueue(&decodeQueue); 243 rename.setRenameQueue(&renameQueue); 244 iew.setRenameQueue(&renameQueue); 245 iew.setIEWQueue(&iewQueue); 246 commit.setIEWQueue(&iewQueue); 247 commit.setRenameQueue(&renameQueue); 248 249 commit.setIEWStage(&iew); 250 rename.setIEWStage(&iew); 251 rename.setCommitStage(&commit); 252 253#if !FULL_SYSTEM 254 int active_threads = params->workload.size(); 255 256 if (active_threads > Impl::MaxThreads) { 257 panic("Workload Size too large. Increase the 'MaxThreads'" 258 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or " 259 "edit your workload size."); 260 } 261#else 262 int active_threads = 1; 263#endif 264 265 //Make Sure That this a Valid Architeture 266 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 267 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 268 269 rename.setScoreboard(&scoreboard); 270 iew.setScoreboard(&scoreboard); 271 272 // Setup the rename map for whichever stages need it. 273 PhysRegIndex lreg_idx = 0; 274 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 275 276 for (int tid=0; tid < numThreads; tid++) { 277 bool bindRegs = (tid <= active_threads - 1); 278 279 commitRenameMap[tid].init(TheISA::NumIntRegs, 280 params->numPhysIntRegs, 281 lreg_idx, //Index for Logical. Regs 282 283 TheISA::NumFloatRegs, 284 params->numPhysFloatRegs, 285 freg_idx, //Index for Float Regs 286 287 TheISA::NumMiscRegs, 288 289 TheISA::ZeroReg, 290 TheISA::ZeroReg, 291 292 tid, 293 false); 294 295 renameMap[tid].init(TheISA::NumIntRegs, 296 params->numPhysIntRegs, 297 lreg_idx, //Index for Logical. Regs 298 299 TheISA::NumFloatRegs, 300 params->numPhysFloatRegs, 301 freg_idx, //Index for Float Regs 302 303 TheISA::NumMiscRegs, 304 305 TheISA::ZeroReg, 306 TheISA::ZeroReg, 307 308 tid, 309 bindRegs); 310 311 activateThreadEvent[tid].init(tid, this); 312 deallocateContextEvent[tid].init(tid, this); 313 } 314 315 rename.setRenameMap(renameMap); 316 commit.setRenameMap(commitRenameMap); 317 318 // Give renameMap & rename stage access to the freeList; 319 for (int i=0; i < numThreads; i++) { 320 renameMap[i].setFreeList(&freeList); 321 } 322 rename.setFreeList(&freeList); 323 324 // Setup the ROB for whichever stages need it. 325 commit.setROB(&rob); 326 327 lastRunningCycle = curTick; 328 329 lastActivatedCycle = -1; 330 331 // Give renameMap & rename stage access to the freeList; 332 //for (int i=0; i < numThreads; i++) { 333 //globalSeqNum[i] = 1; 334 //} 335 336 contextSwitch = false; 337} 338 339template <class Impl> 340FullO3CPU<Impl>::~FullO3CPU() 341{ 342} 343 344template <class Impl> 345void 346FullO3CPU<Impl>::fullCPURegStats() 347{ 348 BaseO3CPU::regStats(); 349 350 // Register any of the O3CPU's stats here. 351 timesIdled 352 .name(name() + ".timesIdled") 353 .desc("Number of times that the entire CPU went into an idle state and" 354 " unscheduled itself") 355 .prereq(timesIdled); 356 357 idleCycles 358 .name(name() + ".idleCycles") 359 .desc("Total number of cycles that the CPU has spent unscheduled due " 360 "to idling") 361 .prereq(idleCycles); 362 363 // Number of Instructions simulated 364 // -------------------------------- 365 // Should probably be in Base CPU but need templated 366 // MaxThreads so put in here instead 367 committedInsts 368 .init(numThreads) 369 .name(name() + ".committedInsts") 370 .desc("Number of Instructions Simulated"); 371 372 totalCommittedInsts 373 .name(name() + ".committedInsts_total") 374 .desc("Number of Instructions Simulated"); 375 376 cpi 377 .name(name() + ".cpi") 378 .desc("CPI: Cycles Per Instruction") 379 .precision(6); 380 cpi = simTicks / committedInsts; 381 382 totalCpi 383 .name(name() + ".cpi_total") 384 .desc("CPI: Total CPI of All Threads") 385 .precision(6); 386 totalCpi = simTicks / totalCommittedInsts; 387 388 ipc 389 .name(name() + ".ipc") 390 .desc("IPC: Instructions Per Cycle") 391 .precision(6); 392 ipc = committedInsts / simTicks; 393 394 totalIpc 395 .name(name() + ".ipc_total") 396 .desc("IPC: Total IPC of All Threads") 397 .precision(6); 398 totalIpc = totalCommittedInsts / simTicks; 399 400} 401 402template <class Impl> 403Port * 404FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 405{ 406 if (if_name == "dcache_port") 407 return iew.getDcachePort(); 408 else if (if_name == "icache_port") 409 return fetch.getIcachePort(); 410 else 411 panic("No Such Port\n"); 412} 413 414template <class Impl> 415void 416FullO3CPU<Impl>::tick() 417{ 418 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 419 420 ++numCycles; 421 422// activity = false; 423 424 //Tick each of the stages 425 fetch.tick(); 426 427 decode.tick(); 428 429 rename.tick(); 430 431 iew.tick(); 432 433 commit.tick(); 434 435#if !FULL_SYSTEM 436 doContextSwitch(); 437#endif 438 439 // Now advance the time buffers 440 timeBuffer.advance(); 441 442 fetchQueue.advance(); 443 decodeQueue.advance(); 444 renameQueue.advance(); 445 iewQueue.advance(); 446 447 activityRec.advance(); 448 449 if (removeInstsThisCycle) { 450 cleanUpRemovedInsts(); 451 } 452 453 if (!tickEvent.scheduled()) { 454 if (_status == SwitchedOut || 455 getState() == SimObject::Drained) { 456 DPRINTF(O3CPU, "Switched out!\n"); 457 // increment stat 458 lastRunningCycle = curTick; 459 } else if (!activityRec.active() || _status == Idle) { 460 DPRINTF(O3CPU, "Idle!\n"); 461 lastRunningCycle = curTick; 462 timesIdled++; 463 } else { 464 tickEvent.schedule(curTick + cycles(1)); 465 DPRINTF(O3CPU, "Scheduling next tick!\n"); 466 } 467 } 468 469#if !FULL_SYSTEM 470 updateThreadPriority(); 471#endif 472 473} 474 475template <class Impl> 476void 477FullO3CPU<Impl>::init() 478{ 479 if (!deferRegistration) { 480 registerThreadContexts(); 481 } 482 483 // Set inSyscall so that the CPU doesn't squash when initially 484 // setting up registers. 485 for (int i = 0; i < number_of_threads; ++i) 486 thread[i]->inSyscall = true; 487 488 for (int tid=0; tid < number_of_threads; tid++) { 489#if FULL_SYSTEM 490 ThreadContext *src_tc = threadContexts[tid]; 491#else 492 ThreadContext *src_tc = thread[tid]->getTC(); 493#endif 494 // Threads start in the Suspended State 495 if (src_tc->status() != ThreadContext::Suspended) { 496 continue; 497 } 498 499#if FULL_SYSTEM 500 TheISA::initCPU(src_tc, src_tc->readCpuId()); 501#endif 502 } 503 504 // Clear inSyscall. 505 for (int i = 0; i < number_of_threads; ++i) 506 thread[i]->inSyscall = false; 507 508 // Initialize stages. 509 fetch.initStage(); 510 iew.initStage(); 511 rename.initStage(); 512 commit.initStage(); 513 514 commit.setThreads(thread); 515} 516 517template <class Impl> 518void 519FullO3CPU<Impl>::activateThread(unsigned tid) 520{ 521 list<unsigned>::iterator isActive = find( 522 activeThreads.begin(), activeThreads.end(), tid); 523 524 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 525 526 if (isActive == activeThreads.end()) { 527 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 528 tid); 529 530 activeThreads.push_back(tid); 531 } 532} 533 534template <class Impl> 535void 536FullO3CPU<Impl>::deactivateThread(unsigned tid) 537{ 538 //Remove From Active List, if Active 539 list<unsigned>::iterator thread_it = 540 find(activeThreads.begin(), activeThreads.end(), tid); 541 542 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 543 544 if (thread_it != activeThreads.end()) { 545 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 546 tid); 547 activeThreads.erase(thread_it); 548 } 549} 550 551template <class Impl> 552void 553FullO3CPU<Impl>::activateContext(int tid, int delay) 554{ 555#if FULL_SYSTEM 556 // Connect the ThreadContext's memory ports (Functional/Virtual 557 // Ports) 558 threadContexts[tid]->connectMemPorts(); 559#endif 560 561 // Needs to set each stage to running as well. 562 if (delay){ 563 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 564 "on cycle %d\n", tid, curTick + cycles(delay)); 565 scheduleActivateThreadEvent(tid, delay); 566 } else { 567 activateThread(tid); 568 } 569 570 if (lastActivatedCycle < curTick) { 571 scheduleTickEvent(delay); 572 573 // Be sure to signal that there's some activity so the CPU doesn't 574 // deschedule itself. 575 activityRec.activity(); 576 fetch.wakeFromQuiesce(); 577 578 lastActivatedCycle = curTick; 579 580 _status = Running; 581 } 582} 583 584template <class Impl> 585bool 586FullO3CPU<Impl>::deallocateContext(int tid, bool remove, int delay) 587{ 588 // Schedule removal of thread data from CPU 589 if (delay){ 590 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 591 "on cycle %d\n", tid, curTick + cycles(delay)); 592 scheduleDeallocateContextEvent(tid, remove, delay); 593 return false; 594 } else { 595 deactivateThread(tid); 596 if (remove) 597 removeThread(tid); 598 return true; 599 } 600} 601 602template <class Impl> 603void 604FullO3CPU<Impl>::suspendContext(int tid) 605{ 606 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 607 bool deallocated = deallocateContext(tid, false, 1); 608 // If this was the last thread then unschedule the tick event. 609 if ((activeThreads.size() == 1 && !deallocated) || activeThreads.size() == 0) 610 unscheduleTickEvent(); 611 _status = Idle; 612} 613 614template <class Impl> 615void 616FullO3CPU<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 623template <class Impl> 624void 625FullO3CPU<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#if ISA_HAS_DELAY_SLOT 659 setNextNPC(src_tc->readNextNPC(), tid); 660#endif 661 662 src_tc->setStatus(ThreadContext::Active); 663 664 activateContext(tid,1); 665 666 //Reset ROB/IQ/LSQ Entries 667 commit.rob->resetEntries(); 668 iew.resetEntries(); 669} 670 671template <class Impl> 672void 673FullO3CPU<Impl>::removeThread(unsigned tid) 674{ 675 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 676 677 // Copy Thread Data From RegFile 678 // If thread is suspended, it might be re-allocated 679 //this->copyToTC(tid); 680 681 // Unbind Int Regs from Rename Map 682 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 683 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 684 685 scoreboard.unsetReg(phys_reg); 686 freeList.addReg(phys_reg); 687 } 688 689 // Unbind Float Regs from Rename Map 690 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 691 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 692 693 scoreboard.unsetReg(phys_reg); 694 freeList.addReg(phys_reg); 695 } 696 697 // Squash Throughout Pipeline 698 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; 699 fetch.squash(0, squash_seq_num, true, tid); 700 decode.squash(tid); 701 rename.squash(squash_seq_num, tid); 702 iew.squash(tid); 703 commit.rob->squash(squash_seq_num, tid); 704 705 assert(iew.ldstQueue.getCount(tid) == 0); 706 707 // Reset ROB/IQ/LSQ Entries 708 709 // Commented out for now. This should be possible to do by 710 // telling all the pipeline stages to drain first, and then 711 // checking until the drain completes. Once the pipeline is 712 // drained, call resetEntries(). - 10-09-06 ktlim 713/* 714 if (activeThreads.size() >= 1) { 715 commit.rob->resetEntries(); 716 iew.resetEntries(); 717 } 718*/ 719} 720 721 722template <class Impl> 723void 724FullO3CPU<Impl>::activateWhenReady(int tid) 725{ 726 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 727 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 728 tid); 729 730 bool ready = true; 731 732 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 733 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 734 "Phys. Int. Regs.\n", 735 tid); 736 ready = false; 737 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 738 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 739 "Phys. Float. Regs.\n", 740 tid); 741 ready = false; 742 } else if (commit.rob->numFreeEntries() >= 743 commit.rob->entryAmount(activeThreads.size() + 1)) { 744 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 745 "ROB entries.\n", 746 tid); 747 ready = false; 748 } else if (iew.instQueue.numFreeEntries() >= 749 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 750 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 751 "IQ entries.\n", 752 tid); 753 ready = false; 754 } else if (iew.ldstQueue.numFreeEntries() >= 755 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 756 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 757 "LSQ entries.\n", 758 tid); 759 ready = false; 760 } 761 762 if (ready) { 763 insertThread(tid); 764 765 contextSwitch = false; 766 767 cpuWaitList.remove(tid); 768 } else { 769 suspendContext(tid); 770 771 //blocks fetch 772 contextSwitch = true; 773 774 //@todo: dont always add to waitlist 775 //do waitlist 776 cpuWaitList.push_back(tid); 777 } 778} 779 780template <class Impl> 781void 782FullO3CPU<Impl>::serialize(std::ostream &os) 783{ 784 SimObject::State so_state = SimObject::getState(); 785 SERIALIZE_ENUM(so_state); 786 BaseCPU::serialize(os); 787 nameOut(os, csprintf("%s.tickEvent", name())); 788 tickEvent.serialize(os); 789 790 // Use SimpleThread's ability to checkpoint to make it easier to 791 // write out the registers. Also make this static so it doesn't 792 // get instantiated multiple times (causes a panic in statistics). 793 static SimpleThread temp; 794 795 for (int i = 0; i < thread.size(); i++) { 796 nameOut(os, csprintf("%s.xc.%i", name(), i)); 797 temp.copyTC(thread[i]->getTC()); 798 temp.serialize(os); 799 } 800} 801 802template <class Impl> 803void 804FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 805{ 806 SimObject::State so_state; 807 UNSERIALIZE_ENUM(so_state); 808 BaseCPU::unserialize(cp, section); 809 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 810 811 // Use SimpleThread's ability to checkpoint to make it easier to 812 // read in the registers. Also make this static so it doesn't 813 // get instantiated multiple times (causes a panic in statistics). 814 static SimpleThread temp; 815 816 for (int i = 0; i < thread.size(); i++) { 817 temp.copyTC(thread[i]->getTC()); 818 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 819 thread[i]->getTC()->copyArchRegs(temp.getTC()); 820 } 821} 822 823template <class Impl> 824unsigned int 825FullO3CPU<Impl>::drain(Event *drain_event) 826{ 827 DPRINTF(O3CPU, "Switching out\n"); 828 829 // If the CPU isn't doing anything, then return immediately. 830 if (_status == Idle || _status == SwitchedOut) { 831 return 0; 832 } 833 834 drainCount = 0; 835 fetch.drain(); 836 decode.drain(); 837 rename.drain(); 838 iew.drain(); 839 commit.drain(); 840 841 // Wake the CPU and record activity so everything can drain out if 842 // the CPU was not able to immediately drain. 843 if (getState() != SimObject::Drained) { 844 // A bit of a hack...set the drainEvent after all the drain() 845 // calls have been made, that way if all of the stages drain 846 // immediately, the signalDrained() function knows not to call 847 // process on the drain event. 848 drainEvent = drain_event; 849 850 wakeCPU(); 851 activityRec.activity(); 852 853 return 1; 854 } else { 855 return 0; 856 } 857} 858 859template <class Impl> 860void 861FullO3CPU<Impl>::resume() 862{ 863 fetch.resume(); 864 decode.resume(); 865 rename.resume(); 866 iew.resume(); 867 commit.resume(); 868 869 changeState(SimObject::Running); 870 871 if (_status == SwitchedOut || _status == Idle) 872 return; 873 874#if FULL_SYSTEM 875 assert(system->getMemoryMode() == System::Timing); 876#endif 877 878 if (!tickEvent.scheduled()) 879 tickEvent.schedule(curTick); 880 _status = Running; 881} 882 883template <class Impl> 884void 885FullO3CPU<Impl>::signalDrained() 886{ 887 if (++drainCount == NumStages) { 888 if (tickEvent.scheduled()) 889 tickEvent.squash(); 890 891 changeState(SimObject::Drained); 892 893 BaseCPU::switchOut(); 894 895 if (drainEvent) { 896 drainEvent->process(); 897 drainEvent = NULL; 898 } 899 } 900 assert(drainCount <= 5); 901} 902 903template <class Impl> 904void 905FullO3CPU<Impl>::switchOut() 906{ 907 fetch.switchOut(); 908 rename.switchOut(); 909 iew.switchOut(); 910 commit.switchOut(); 911 instList.clear(); 912 while (!removeList.empty()) { 913 removeList.pop(); 914 } 915 916 _status = SwitchedOut; 917#if USE_CHECKER 918 if (checker) 919 checker->switchOut(); 920#endif 921 if (tickEvent.scheduled()) 922 tickEvent.squash(); 923} 924 925template <class Impl> 926void 927FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 928{ 929 // Flush out any old data from the time buffers. 930 for (int i = 0; i < timeBuffer.getSize(); ++i) { 931 timeBuffer.advance(); 932 fetchQueue.advance(); 933 decodeQueue.advance(); 934 renameQueue.advance(); 935 iewQueue.advance(); 936 } 937 938 activityRec.reset(); 939 940 BaseCPU::takeOverFrom(oldCPU); 941 942 fetch.takeOverFrom(); 943 decode.takeOverFrom(); 944 rename.takeOverFrom(); 945 iew.takeOverFrom(); 946 commit.takeOverFrom(); 947 948 assert(!tickEvent.scheduled()); 949 950 // @todo: Figure out how to properly select the tid to put onto 951 // the active threads list. 952 int tid = 0; 953 954 list<unsigned>::iterator isActive = find( 955 activeThreads.begin(), activeThreads.end(), tid); 956 957 if (isActive == activeThreads.end()) { 958 //May Need to Re-code this if the delay variable is the delay 959 //needed for thread to activate 960 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 961 tid); 962 963 activeThreads.push_back(tid); 964 } 965 966 // Set all statuses to active, schedule the CPU's tick event. 967 // @todo: Fix up statuses so this is handled properly 968 for (int i = 0; i < threadContexts.size(); ++i) { 969 ThreadContext *tc = threadContexts[i]; 970 if (tc->status() == ThreadContext::Active && _status != Running) { 971 _status = Running; 972 tickEvent.schedule(curTick); 973 } 974 } 975 if (!tickEvent.scheduled()) 976 tickEvent.schedule(curTick); 977 978 Port *peer; 979 Port *icachePort = fetch.getIcachePort(); 980 if (icachePort->getPeer() == NULL) { 981 peer = oldCPU->getPort("icache_port")->getPeer(); 982 icachePort->setPeer(peer); 983 } else { 984 peer = icachePort->getPeer(); 985 } 986 peer->setPeer(icachePort); 987 988 Port *dcachePort = iew.getDcachePort(); 989 if (dcachePort->getPeer() == NULL) { 990 peer = oldCPU->getPort("dcache_port")->getPeer(); 991 dcachePort->setPeer(peer); 992 } else { 993 peer = dcachePort->getPeer(); 994 } 995 peer->setPeer(dcachePort); 996} 997 998template <class Impl> 999uint64_t 1000FullO3CPU<Impl>::readIntReg(int reg_idx) 1001{ 1002 return regFile.readIntReg(reg_idx); 1003} 1004 1005template <class Impl> 1006FloatReg 1007FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 1008{ 1009 return regFile.readFloatReg(reg_idx, width); 1010} 1011 1012template <class Impl> 1013FloatReg 1014FullO3CPU<Impl>::readFloatReg(int reg_idx) 1015{ 1016 return regFile.readFloatReg(reg_idx); 1017} 1018 1019template <class Impl> 1020FloatRegBits 1021FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1022{ 1023 return regFile.readFloatRegBits(reg_idx, width); 1024} 1025 1026template <class Impl> 1027FloatRegBits 1028FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1029{ 1030 return regFile.readFloatRegBits(reg_idx); 1031} 1032 1033template <class Impl> 1034void 1035FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1036{ 1037 regFile.setIntReg(reg_idx, val); 1038} 1039 1040template <class Impl> 1041void 1042FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1043{ 1044 regFile.setFloatReg(reg_idx, val, width); 1045} 1046 1047template <class Impl> 1048void 1049FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1050{ 1051 regFile.setFloatReg(reg_idx, val); 1052} 1053 1054template <class Impl> 1055void 1056FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1057{ 1058 regFile.setFloatRegBits(reg_idx, val, width); 1059} 1060 1061template <class Impl> 1062void 1063FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1064{ 1065 regFile.setFloatRegBits(reg_idx, val); 1066} 1067 1068template <class Impl> 1069uint64_t 1070FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1071{ 1072 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1073 1074 return regFile.readIntReg(phys_reg); 1075} 1076 1077template <class Impl> 1078float 1079FullO3CPU<Impl>::readArchFloatRegSingle(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); 1085} 1086 1087template <class Impl> 1088double 1089FullO3CPU<Impl>::readArchFloatRegDouble(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.readFloatReg(phys_reg, 64); 1095} 1096 1097template <class Impl> 1098uint64_t 1099FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1100{ 1101 int idx = reg_idx + TheISA::FP_Base_DepTag; 1102 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1103 1104 return regFile.readFloatRegBits(phys_reg); 1105} 1106 1107template <class Impl> 1108void 1109FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1110{ 1111 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1112 1113 regFile.setIntReg(phys_reg, val); 1114} 1115 1116template <class Impl> 1117void 1118FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float 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); 1124} 1125 1126template <class Impl> 1127void 1128FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1129{ 1130 int idx = reg_idx + TheISA::FP_Base_DepTag; 1131 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1132 1133 regFile.setFloatReg(phys_reg, val, 64); 1134} 1135 1136template <class Impl> 1137void 1138FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1139{ 1140 int idx = reg_idx + TheISA::FP_Base_DepTag; 1141 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1142 1143 regFile.setFloatRegBits(phys_reg, val); 1144} 1145 1146template <class Impl> 1147uint64_t 1148FullO3CPU<Impl>::readPC(unsigned tid) 1149{ 1150 return commit.readPC(tid); 1151} 1152 1153template <class Impl> 1154void 1155FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1156{ 1157 commit.setPC(new_PC, tid); 1158} 1159 1160template <class Impl> 1161uint64_t 1162FullO3CPU<Impl>::readNextPC(unsigned tid) 1163{ 1164 return commit.readNextPC(tid); 1165} 1166 1167template <class Impl> 1168void 1169FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1170{ 1171 commit.setNextPC(val, tid); 1172} 1173 1174template <class Impl> 1175uint64_t 1176FullO3CPU<Impl>::readNextNPC(unsigned tid) 1177{ 1178 return commit.readNextNPC(tid); 1179} 1180 1181template <class Impl> 1182void 1183FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1184{ 1185 commit.setNextNPC(val, tid); 1186} 1187 1188template <class Impl> 1189typename FullO3CPU<Impl>::ListIt 1190FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1191{ 1192 instList.push_back(inst); 1193 1194 return --(instList.end()); 1195} 1196 1197template <class Impl> 1198void 1199FullO3CPU<Impl>::instDone(unsigned tid) 1200{ 1201 // Keep an instruction count. 1202 thread[tid]->numInst++; 1203 thread[tid]->numInsts++; 1204 committedInsts[tid]++; 1205 totalCommittedInsts++; 1206 1207 // Check for instruction-count-based events. 1208 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1209} 1210 1211template <class Impl> 1212void 1213FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1214{ 1215 removeInstsThisCycle = true; 1216 1217 removeList.push(inst->getInstListIt()); 1218} 1219 1220template <class Impl> 1221void 1222FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1223{ 1224 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1225 "[sn:%lli]\n", 1226 inst->threadNumber, inst->readPC(), inst->seqNum); 1227 1228 removeInstsThisCycle = true; 1229 1230 // Remove the front instruction. 1231 removeList.push(inst->getInstListIt()); 1232} 1233 1234template <class Impl> 1235void 1236FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid, 1237 bool squash_delay_slot, 1238 const InstSeqNum &delay_slot_seq_num) 1239{ 1240 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1241 " list.\n", tid); 1242 1243 ListIt end_it; 1244 1245 bool rob_empty = false; 1246 1247 if (instList.empty()) { 1248 return; 1249 } else if (rob.isEmpty(/*tid*/)) { 1250 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1251 end_it = instList.begin(); 1252 rob_empty = true; 1253 } else { 1254 end_it = (rob.readTailInst(tid))->getInstListIt(); 1255 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1256 } 1257 1258 removeInstsThisCycle = true; 1259 1260 ListIt inst_it = instList.end(); 1261 1262 inst_it--; 1263 1264 // Walk through the instruction list, removing any instructions 1265 // that were inserted after the given instruction iterator, end_it. 1266 while (inst_it != end_it) { 1267 assert(!instList.empty()); 1268 1269#if ISA_HAS_DELAY_SLOT 1270 if(!squash_delay_slot && 1271 delay_slot_seq_num >= (*inst_it)->seqNum) { 1272 break; 1273 } 1274#endif 1275 squashInstIt(inst_it, tid); 1276 1277 inst_it--; 1278 } 1279 1280 // If the ROB was empty, then we actually need to remove the first 1281 // instruction as well. 1282 if (rob_empty) { 1283 squashInstIt(inst_it, tid); 1284 } 1285} 1286 1287template <class Impl> 1288void 1289FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1290 unsigned tid) 1291{ 1292 assert(!instList.empty()); 1293 1294 removeInstsThisCycle = true; 1295 1296 ListIt inst_iter = instList.end(); 1297 1298 inst_iter--; 1299 1300 DPRINTF(O3CPU, "Deleting instructions from instruction " 1301 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1302 tid, seq_num, (*inst_iter)->seqNum); 1303 1304 while ((*inst_iter)->seqNum > seq_num) { 1305 1306 bool break_loop = (inst_iter == instList.begin()); 1307 1308 squashInstIt(inst_iter, tid); 1309 1310 inst_iter--; 1311 1312 if (break_loop) 1313 break; 1314 } 1315} 1316 1317template <class Impl> 1318inline void 1319FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1320{ 1321 if ((*instIt)->threadNumber == tid) { 1322 DPRINTF(O3CPU, "Squashing instruction, " 1323 "[tid:%i] [sn:%lli] PC %#x\n", 1324 (*instIt)->threadNumber, 1325 (*instIt)->seqNum, 1326 (*instIt)->readPC()); 1327 1328 // Mark it as squashed. 1329 (*instIt)->setSquashed(); 1330 1331 // @todo: Formulate a consistent method for deleting 1332 // instructions from the instruction list 1333 // Remove the instruction from the list. 1334 removeList.push(instIt); 1335 } 1336} 1337 1338template <class Impl> 1339void 1340FullO3CPU<Impl>::cleanUpRemovedInsts() 1341{ 1342 while (!removeList.empty()) { 1343 DPRINTF(O3CPU, "Removing instruction, " 1344 "[tid:%i] [sn:%lli] PC %#x\n", 1345 (*removeList.front())->threadNumber, 1346 (*removeList.front())->seqNum, 1347 (*removeList.front())->readPC()); 1348 1349 instList.erase(removeList.front()); 1350 1351 removeList.pop(); 1352 } 1353 1354 removeInstsThisCycle = false; 1355} 1356/* 1357template <class Impl> 1358void 1359FullO3CPU<Impl>::removeAllInsts() 1360{ 1361 instList.clear(); 1362} 1363*/ 1364template <class Impl> 1365void 1366FullO3CPU<Impl>::dumpInsts() 1367{ 1368 int num = 0; 1369 1370 ListIt inst_list_it = instList.begin(); 1371 1372 cprintf("Dumping Instruction List\n"); 1373 1374 while (inst_list_it != instList.end()) { 1375 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1376 "Squashed:%i\n\n", 1377 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1378 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1379 (*inst_list_it)->isSquashed()); 1380 inst_list_it++; 1381 ++num; 1382 } 1383} 1384/* 1385template <class Impl> 1386void 1387FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1388{ 1389 iew.wakeDependents(inst); 1390} 1391*/ 1392template <class Impl> 1393void 1394FullO3CPU<Impl>::wakeCPU() 1395{ 1396 if (activityRec.active() || tickEvent.scheduled()) { 1397 DPRINTF(Activity, "CPU already running.\n"); 1398 return; 1399 } 1400 1401 DPRINTF(Activity, "Waking up CPU\n"); 1402 1403 idleCycles += (curTick - 1) - lastRunningCycle; 1404 1405 tickEvent.schedule(curTick); 1406} 1407 1408template <class Impl> 1409int 1410FullO3CPU<Impl>::getFreeTid() 1411{ 1412 for (int i=0; i < numThreads; i++) { 1413 if (!tids[i]) { 1414 tids[i] = true; 1415 return i; 1416 } 1417 } 1418 1419 return -1; 1420} 1421 1422template <class Impl> 1423void 1424FullO3CPU<Impl>::doContextSwitch() 1425{ 1426 if (contextSwitch) { 1427 1428 //ADD CODE TO DEACTIVE THREAD HERE (???) 1429 1430 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1431 activateWhenReady(tid); 1432 } 1433 1434 if (cpuWaitList.size() == 0) 1435 contextSwitch = true; 1436 } 1437} 1438 1439template <class Impl> 1440void 1441FullO3CPU<Impl>::updateThreadPriority() 1442{ 1443 if (activeThreads.size() > 1) 1444 { 1445 //DEFAULT TO ROUND ROBIN SCHEME 1446 //e.g. Move highest priority to end of thread list 1447 list<unsigned>::iterator list_begin = activeThreads.begin(); 1448 list<unsigned>::iterator list_end = activeThreads.end(); 1449 1450 unsigned high_thread = *list_begin; 1451 1452 activeThreads.erase(list_begin); 1453 1454 activeThreads.push_back(high_thread); 1455 } 1456} 1457 1458// Forward declaration of FullO3CPU. 1459template class FullO3CPU<O3CPUImpl>; 1460