cpu.cc revision 5336
1/* 2 * Copyright (c) 2004-2006 The Regents of The University of Michigan 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are 7 * met: redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer; 9 * redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution; 12 * neither the name of the copyright holders nor the names of its 13 * contributors may be used to endorse or promote products derived from 14 * this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 * 28 * Authors: Kevin Lim 29 * Korey Sewell 30 */ 31 32#include "config/full_system.hh" 33#include "config/use_checker.hh" 34 35#include "cpu/activity.hh" 36#include "cpu/simple_thread.hh" 37#include "cpu/thread_context.hh" 38#include "cpu/o3/isa_specific.hh" 39#include "cpu/o3/cpu.hh" 40#include "enums/MemoryMode.hh" 41#include "sim/core.hh" 42#include "sim/stat_control.hh" 43 44#if FULL_SYSTEM 45#include "cpu/quiesce_event.hh" 46#include "sim/system.hh" 47#else 48#include "sim/process.hh" 49#endif 50 51#if USE_CHECKER 52#include "cpu/checker/cpu.hh" 53#endif 54 55using namespace TheISA; 56 57BaseO3CPU::BaseO3CPU(Params *params) 58 : BaseCPU(params), cpu_id(0) 59{ 60} 61 62void 63BaseO3CPU::regStats() 64{ 65 BaseCPU::regStats(); 66} 67 68template <class Impl> 69FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 70 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) 71{ 72} 73 74template <class Impl> 75void 76FullO3CPU<Impl>::TickEvent::process() 77{ 78 cpu->tick(); 79} 80 81template <class Impl> 82const char * 83FullO3CPU<Impl>::TickEvent::description() const 84{ 85 return "FullO3CPU tick"; 86} 87 88template <class Impl> 89FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 90 : Event(&mainEventQueue, CPU_Switch_Pri) 91{ 92} 93 94template <class Impl> 95void 96FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 97 FullO3CPU<Impl> *thread_cpu) 98{ 99 tid = thread_num; 100 cpu = thread_cpu; 101} 102 103template <class Impl> 104void 105FullO3CPU<Impl>::ActivateThreadEvent::process() 106{ 107 cpu->activateThread(tid); 108} 109 110template <class Impl> 111const char * 112FullO3CPU<Impl>::ActivateThreadEvent::description() const 113{ 114 return "FullO3CPU \"Activate Thread\""; 115} 116 117template <class Impl> 118FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 119 : Event(&mainEventQueue, CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 120{ 121} 122 123template <class Impl> 124void 125FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 126 FullO3CPU<Impl> *thread_cpu) 127{ 128 tid = thread_num; 129 cpu = thread_cpu; 130 remove = false; 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() const 145{ 146 return "FullO3CPU \"Deallocate Context\""; 147} 148 149template <class Impl> 150FullO3CPU<Impl>::FullO3CPU(O3CPU *o3_cpu, Params *params) 151 : BaseO3CPU(params), 152 itb(params->itb), 153 dtb(params->dtb), 154 tickEvent(this), 155 removeInstsThisCycle(false), 156 fetch(o3_cpu, params), 157 decode(o3_cpu, params), 158 rename(o3_cpu, params), 159 iew(o3_cpu, params), 160 commit(o3_cpu, params), 161 162 regFile(o3_cpu, params->numPhysIntRegs, 163 params->numPhysFloatRegs), 164 165 freeList(params->numberOfThreads, 166 TheISA::NumIntRegs, params->numPhysIntRegs, 167 TheISA::NumFloatRegs, params->numPhysFloatRegs), 168 169 rob(o3_cpu, 170 params->numROBEntries, params->squashWidth, 171 params->smtROBPolicy, params->smtROBThreshold, 172 params->numberOfThreads), 173 174 scoreboard(params->numberOfThreads, 175 TheISA::NumIntRegs, params->numPhysIntRegs, 176 TheISA::NumFloatRegs, params->numPhysFloatRegs, 177 TheISA::NumMiscRegs * number_of_threads, 178 TheISA::ZeroReg), 179 180 timeBuffer(params->backComSize, params->forwardComSize), 181 fetchQueue(params->backComSize, params->forwardComSize), 182 decodeQueue(params->backComSize, params->forwardComSize), 183 renameQueue(params->backComSize, params->forwardComSize), 184 iewQueue(params->backComSize, params->forwardComSize), 185 activityRec(NumStages, 186 params->backComSize + params->forwardComSize, 187 params->activity), 188 189 globalSeqNum(1), 190#if FULL_SYSTEM 191 system(params->system), 192 physmem(system->physmem), 193#endif // FULL_SYSTEM 194 drainCount(0), 195 deferRegistration(params->deferRegistration), 196 numThreads(number_of_threads) 197{ 198 if (!deferRegistration) { 199 _status = Running; 200 } else { 201 _status = Idle; 202 } 203 204#if USE_CHECKER 205 if (params->checker) { 206 BaseCPU *temp_checker = params->checker; 207 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 208#if FULL_SYSTEM 209 checker->setSystem(params->system); 210#endif 211 } else { 212 checker = NULL; 213 } 214#endif // USE_CHECKER 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 = numCycles / committedInsts; 383 384 totalCpi 385 .name(name() + ".cpi_total") 386 .desc("CPI: Total CPI of All Threads") 387 .precision(6); 388 totalCpi = numCycles / totalCommittedInsts; 389 390 ipc 391 .name(name() + ".ipc") 392 .desc("IPC: Instructions Per Cycle") 393 .precision(6); 394 ipc = committedInsts / numCycles; 395 396 totalIpc 397 .name(name() + ".ipc_total") 398 .desc("IPC: Total IPC of All Threads") 399 .precision(6); 400 totalIpc = totalCommittedInsts / numCycles; 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(nextCycle(curTick + ticks(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 std::list<unsigned>::iterator isActive = 524 std::find(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 std::list<unsigned>::iterator thread_it = 542 std::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 + ticks(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 + ticks(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 || 606 activeThreads.size() == 0) 607 unscheduleTickEvent(); 608 _status = Idle; 609} 610 611template <class Impl> 612void 613FullO3CPU<Impl>::haltContext(int tid) 614{ 615 //For now, this is the same as deallocate 616 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 617 deallocateContext(tid, true, 1); 618} 619 620template <class Impl> 621void 622FullO3CPU<Impl>::insertThread(unsigned tid) 623{ 624 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 625 // Will change now that the PC and thread state is internal to the CPU 626 // and not in the ThreadContext. 627#if FULL_SYSTEM 628 ThreadContext *src_tc = system->threadContexts[tid]; 629#else 630 ThreadContext *src_tc = tcBase(tid); 631#endif 632 633 //Bind Int Regs to Rename Map 634 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 635 PhysRegIndex phys_reg = freeList.getIntReg(); 636 637 renameMap[tid].setEntry(ireg,phys_reg); 638 scoreboard.setReg(phys_reg); 639 } 640 641 //Bind Float Regs to Rename Map 642 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 643 PhysRegIndex phys_reg = freeList.getFloatReg(); 644 645 renameMap[tid].setEntry(freg,phys_reg); 646 scoreboard.setReg(phys_reg); 647 } 648 649 //Copy Thread Data Into RegFile 650 //this->copyFromTC(tid); 651 652 //Set PC/NPC/NNPC 653 setPC(src_tc->readPC(), tid); 654 setNextPC(src_tc->readNextPC(), tid); 655 setNextNPC(src_tc->readNextNPC(), tid); 656 657 src_tc->setStatus(ThreadContext::Active); 658 659 activateContext(tid,1); 660 661 //Reset ROB/IQ/LSQ Entries 662 commit.rob->resetEntries(); 663 iew.resetEntries(); 664} 665 666template <class Impl> 667void 668FullO3CPU<Impl>::removeThread(unsigned tid) 669{ 670 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 671 672 // Copy Thread Data From RegFile 673 // If thread is suspended, it might be re-allocated 674 //this->copyToTC(tid); 675 676 // Unbind Int Regs from Rename Map 677 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 678 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 679 680 scoreboard.unsetReg(phys_reg); 681 freeList.addReg(phys_reg); 682 } 683 684 // Unbind Float Regs from Rename Map 685 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 686 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 687 688 scoreboard.unsetReg(phys_reg); 689 freeList.addReg(phys_reg); 690 } 691 692 // Squash Throughout Pipeline 693 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; 694 fetch.squash(0, sizeof(TheISA::MachInst), 0, squash_seq_num, tid); 695 decode.squash(tid); 696 rename.squash(squash_seq_num, tid); 697 iew.squash(tid); 698 commit.rob->squash(squash_seq_num, tid); 699 700 assert(iew.ldstQueue.getCount(tid) == 0); 701 702 // Reset ROB/IQ/LSQ Entries 703 704 // Commented out for now. This should be possible to do by 705 // telling all the pipeline stages to drain first, and then 706 // checking until the drain completes. Once the pipeline is 707 // drained, call resetEntries(). - 10-09-06 ktlim 708/* 709 if (activeThreads.size() >= 1) { 710 commit.rob->resetEntries(); 711 iew.resetEntries(); 712 } 713*/ 714} 715 716 717template <class Impl> 718void 719FullO3CPU<Impl>::activateWhenReady(int tid) 720{ 721 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 722 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 723 tid); 724 725 bool ready = true; 726 727 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 728 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 729 "Phys. Int. Regs.\n", 730 tid); 731 ready = false; 732 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 733 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 734 "Phys. Float. Regs.\n", 735 tid); 736 ready = false; 737 } else if (commit.rob->numFreeEntries() >= 738 commit.rob->entryAmount(activeThreads.size() + 1)) { 739 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 740 "ROB entries.\n", 741 tid); 742 ready = false; 743 } else if (iew.instQueue.numFreeEntries() >= 744 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 745 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 746 "IQ entries.\n", 747 tid); 748 ready = false; 749 } else if (iew.ldstQueue.numFreeEntries() >= 750 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 751 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 752 "LSQ entries.\n", 753 tid); 754 ready = false; 755 } 756 757 if (ready) { 758 insertThread(tid); 759 760 contextSwitch = false; 761 762 cpuWaitList.remove(tid); 763 } else { 764 suspendContext(tid); 765 766 //blocks fetch 767 contextSwitch = true; 768 769 //@todo: dont always add to waitlist 770 //do waitlist 771 cpuWaitList.push_back(tid); 772 } 773} 774 775#if FULL_SYSTEM 776template <class Impl> 777void 778FullO3CPU<Impl>::updateMemPorts() 779{ 780 // Update all ThreadContext's memory ports (Functional/Virtual 781 // Ports) 782 for (int i = 0; i < thread.size(); ++i) 783 thread[i]->connectMemPorts(); 784} 785#endif 786 787template <class Impl> 788void 789FullO3CPU<Impl>::serialize(std::ostream &os) 790{ 791 SimObject::State so_state = SimObject::getState(); 792 SERIALIZE_ENUM(so_state); 793 BaseCPU::serialize(os); 794 nameOut(os, csprintf("%s.tickEvent", name())); 795 tickEvent.serialize(os); 796 797 // Use SimpleThread's ability to checkpoint to make it easier to 798 // write out the registers. Also make this static so it doesn't 799 // get instantiated multiple times (causes a panic in statistics). 800 static SimpleThread temp; 801 802 for (int i = 0; i < thread.size(); i++) { 803 nameOut(os, csprintf("%s.xc.%i", name(), i)); 804 temp.copyTC(thread[i]->getTC()); 805 temp.serialize(os); 806 } 807} 808 809template <class Impl> 810void 811FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 812{ 813 SimObject::State so_state; 814 UNSERIALIZE_ENUM(so_state); 815 BaseCPU::unserialize(cp, section); 816 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 817 818 // Use SimpleThread's ability to checkpoint to make it easier to 819 // read in the registers. Also make this static so it doesn't 820 // get instantiated multiple times (causes a panic in statistics). 821 static SimpleThread temp; 822 823 for (int i = 0; i < thread.size(); i++) { 824 temp.copyTC(thread[i]->getTC()); 825 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 826 thread[i]->getTC()->copyArchRegs(temp.getTC()); 827 } 828} 829 830template <class Impl> 831unsigned int 832FullO3CPU<Impl>::drain(Event *drain_event) 833{ 834 DPRINTF(O3CPU, "Switching out\n"); 835 836 // If the CPU isn't doing anything, then return immediately. 837 if (_status == Idle || _status == SwitchedOut) { 838 return 0; 839 } 840 841 drainCount = 0; 842 fetch.drain(); 843 decode.drain(); 844 rename.drain(); 845 iew.drain(); 846 commit.drain(); 847 848 // Wake the CPU and record activity so everything can drain out if 849 // the CPU was not able to immediately drain. 850 if (getState() != SimObject::Drained) { 851 // A bit of a hack...set the drainEvent after all the drain() 852 // calls have been made, that way if all of the stages drain 853 // immediately, the signalDrained() function knows not to call 854 // process on the drain event. 855 drainEvent = drain_event; 856 857 wakeCPU(); 858 activityRec.activity(); 859 860 return 1; 861 } else { 862 return 0; 863 } 864} 865 866template <class Impl> 867void 868FullO3CPU<Impl>::resume() 869{ 870 fetch.resume(); 871 decode.resume(); 872 rename.resume(); 873 iew.resume(); 874 commit.resume(); 875 876 changeState(SimObject::Running); 877 878 if (_status == SwitchedOut || _status == Idle) 879 return; 880 881#if FULL_SYSTEM 882 assert(system->getMemoryMode() == Enums::timing); 883#endif 884 885 if (!tickEvent.scheduled()) 886 tickEvent.schedule(nextCycle()); 887 _status = Running; 888} 889 890template <class Impl> 891void 892FullO3CPU<Impl>::signalDrained() 893{ 894 if (++drainCount == NumStages) { 895 if (tickEvent.scheduled()) 896 tickEvent.squash(); 897 898 changeState(SimObject::Drained); 899 900 BaseCPU::switchOut(); 901 902 if (drainEvent) { 903 drainEvent->process(); 904 drainEvent = NULL; 905 } 906 } 907 assert(drainCount <= 5); 908} 909 910template <class Impl> 911void 912FullO3CPU<Impl>::switchOut() 913{ 914 fetch.switchOut(); 915 rename.switchOut(); 916 iew.switchOut(); 917 commit.switchOut(); 918 instList.clear(); 919 while (!removeList.empty()) { 920 removeList.pop(); 921 } 922 923 _status = SwitchedOut; 924#if USE_CHECKER 925 if (checker) 926 checker->switchOut(); 927#endif 928 if (tickEvent.scheduled()) 929 tickEvent.squash(); 930} 931 932template <class Impl> 933void 934FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 935{ 936 // Flush out any old data from the time buffers. 937 for (int i = 0; i < timeBuffer.getSize(); ++i) { 938 timeBuffer.advance(); 939 fetchQueue.advance(); 940 decodeQueue.advance(); 941 renameQueue.advance(); 942 iewQueue.advance(); 943 } 944 945 activityRec.reset(); 946 947 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort()); 948 949 fetch.takeOverFrom(); 950 decode.takeOverFrom(); 951 rename.takeOverFrom(); 952 iew.takeOverFrom(); 953 commit.takeOverFrom(); 954 955 assert(!tickEvent.scheduled()); 956 957 // @todo: Figure out how to properly select the tid to put onto 958 // the active threads list. 959 int tid = 0; 960 961 std::list<unsigned>::iterator isActive = 962 std::find(activeThreads.begin(), activeThreads.end(), tid); 963 964 if (isActive == activeThreads.end()) { 965 //May Need to Re-code this if the delay variable is the delay 966 //needed for thread to activate 967 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 968 tid); 969 970 activeThreads.push_back(tid); 971 } 972 973 // Set all statuses to active, schedule the CPU's tick event. 974 // @todo: Fix up statuses so this is handled properly 975 for (int i = 0; i < threadContexts.size(); ++i) { 976 ThreadContext *tc = threadContexts[i]; 977 if (tc->status() == ThreadContext::Active && _status != Running) { 978 _status = Running; 979 tickEvent.schedule(nextCycle()); 980 } 981 } 982 if (!tickEvent.scheduled()) 983 tickEvent.schedule(nextCycle()); 984} 985 986template <class Impl> 987uint64_t 988FullO3CPU<Impl>::readIntReg(int reg_idx) 989{ 990 return regFile.readIntReg(reg_idx); 991} 992 993template <class Impl> 994FloatReg 995FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 996{ 997 return regFile.readFloatReg(reg_idx, width); 998} 999 1000template <class Impl> 1001FloatReg 1002FullO3CPU<Impl>::readFloatReg(int reg_idx) 1003{ 1004 return regFile.readFloatReg(reg_idx); 1005} 1006 1007template <class Impl> 1008FloatRegBits 1009FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1010{ 1011 return regFile.readFloatRegBits(reg_idx, width); 1012} 1013 1014template <class Impl> 1015FloatRegBits 1016FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1017{ 1018 return regFile.readFloatRegBits(reg_idx); 1019} 1020 1021template <class Impl> 1022void 1023FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1024{ 1025 regFile.setIntReg(reg_idx, val); 1026} 1027 1028template <class Impl> 1029void 1030FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1031{ 1032 regFile.setFloatReg(reg_idx, val, width); 1033} 1034 1035template <class Impl> 1036void 1037FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1038{ 1039 regFile.setFloatReg(reg_idx, val); 1040} 1041 1042template <class Impl> 1043void 1044FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1045{ 1046 regFile.setFloatRegBits(reg_idx, val, width); 1047} 1048 1049template <class Impl> 1050void 1051FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1052{ 1053 regFile.setFloatRegBits(reg_idx, val); 1054} 1055 1056template <class Impl> 1057uint64_t 1058FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1059{ 1060 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1061 1062 return regFile.readIntReg(phys_reg); 1063} 1064 1065template <class Impl> 1066float 1067FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 1068{ 1069 int idx = reg_idx + TheISA::FP_Base_DepTag; 1070 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1071 1072 return regFile.readFloatReg(phys_reg); 1073} 1074 1075template <class Impl> 1076double 1077FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 1078{ 1079 int idx = reg_idx + TheISA::FP_Base_DepTag; 1080 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1081 1082 return regFile.readFloatReg(phys_reg, 64); 1083} 1084 1085template <class Impl> 1086uint64_t 1087FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1088{ 1089 int idx = reg_idx + TheISA::FP_Base_DepTag; 1090 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1091 1092 return regFile.readFloatRegBits(phys_reg); 1093} 1094 1095template <class Impl> 1096void 1097FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1098{ 1099 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1100 1101 regFile.setIntReg(phys_reg, val); 1102} 1103 1104template <class Impl> 1105void 1106FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 1107{ 1108 int idx = reg_idx + TheISA::FP_Base_DepTag; 1109 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1110 1111 regFile.setFloatReg(phys_reg, val); 1112} 1113 1114template <class Impl> 1115void 1116FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1117{ 1118 int idx = reg_idx + TheISA::FP_Base_DepTag; 1119 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1120 1121 regFile.setFloatReg(phys_reg, val, 64); 1122} 1123 1124template <class Impl> 1125void 1126FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1127{ 1128 int idx = reg_idx + TheISA::FP_Base_DepTag; 1129 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1130 1131 regFile.setFloatRegBits(phys_reg, val); 1132} 1133 1134template <class Impl> 1135uint64_t 1136FullO3CPU<Impl>::readPC(unsigned tid) 1137{ 1138 return commit.readPC(tid); 1139} 1140 1141template <class Impl> 1142void 1143FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1144{ 1145 commit.setPC(new_PC, tid); 1146} 1147 1148template <class Impl> 1149uint64_t 1150FullO3CPU<Impl>::readMicroPC(unsigned tid) 1151{ 1152 return commit.readMicroPC(tid); 1153} 1154 1155template <class Impl> 1156void 1157FullO3CPU<Impl>::setMicroPC(Addr new_PC,unsigned tid) 1158{ 1159 commit.setMicroPC(new_PC, tid); 1160} 1161 1162template <class Impl> 1163uint64_t 1164FullO3CPU<Impl>::readNextPC(unsigned tid) 1165{ 1166 return commit.readNextPC(tid); 1167} 1168 1169template <class Impl> 1170void 1171FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1172{ 1173 commit.setNextPC(val, tid); 1174} 1175 1176template <class Impl> 1177uint64_t 1178FullO3CPU<Impl>::readNextNPC(unsigned tid) 1179{ 1180 return commit.readNextNPC(tid); 1181} 1182 1183template <class Impl> 1184void 1185FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1186{ 1187 commit.setNextNPC(val, tid); 1188} 1189 1190template <class Impl> 1191uint64_t 1192FullO3CPU<Impl>::readNextMicroPC(unsigned tid) 1193{ 1194 return commit.readNextMicroPC(tid); 1195} 1196 1197template <class Impl> 1198void 1199FullO3CPU<Impl>::setNextMicroPC(Addr new_PC,unsigned tid) 1200{ 1201 commit.setNextMicroPC(new_PC, tid); 1202} 1203 1204template <class Impl> 1205typename FullO3CPU<Impl>::ListIt 1206FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1207{ 1208 instList.push_back(inst); 1209 1210 return --(instList.end()); 1211} 1212 1213template <class Impl> 1214void 1215FullO3CPU<Impl>::instDone(unsigned tid) 1216{ 1217 // Keep an instruction count. 1218 thread[tid]->numInst++; 1219 thread[tid]->numInsts++; 1220 committedInsts[tid]++; 1221 totalCommittedInsts++; 1222 1223 // Check for instruction-count-based events. 1224 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1225} 1226 1227template <class Impl> 1228void 1229FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1230{ 1231 removeInstsThisCycle = true; 1232 1233 removeList.push(inst->getInstListIt()); 1234} 1235 1236template <class Impl> 1237void 1238FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1239{ 1240 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1241 "[sn:%lli]\n", 1242 inst->threadNumber, inst->readPC(), inst->seqNum); 1243 1244 removeInstsThisCycle = true; 1245 1246 // Remove the front instruction. 1247 removeList.push(inst->getInstListIt()); 1248} 1249 1250template <class Impl> 1251void 1252FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 1253{ 1254 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1255 " list.\n", tid); 1256 1257 ListIt end_it; 1258 1259 bool rob_empty = false; 1260 1261 if (instList.empty()) { 1262 return; 1263 } else if (rob.isEmpty(/*tid*/)) { 1264 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1265 end_it = instList.begin(); 1266 rob_empty = true; 1267 } else { 1268 end_it = (rob.readTailInst(tid))->getInstListIt(); 1269 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1270 } 1271 1272 removeInstsThisCycle = true; 1273 1274 ListIt inst_it = instList.end(); 1275 1276 inst_it--; 1277 1278 // Walk through the instruction list, removing any instructions 1279 // that were inserted after the given instruction iterator, end_it. 1280 while (inst_it != end_it) { 1281 assert(!instList.empty()); 1282 1283 squashInstIt(inst_it, tid); 1284 1285 inst_it--; 1286 } 1287 1288 // If the ROB was empty, then we actually need to remove the first 1289 // instruction as well. 1290 if (rob_empty) { 1291 squashInstIt(inst_it, tid); 1292 } 1293} 1294 1295template <class Impl> 1296void 1297FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1298 unsigned tid) 1299{ 1300 assert(!instList.empty()); 1301 1302 removeInstsThisCycle = true; 1303 1304 ListIt inst_iter = instList.end(); 1305 1306 inst_iter--; 1307 1308 DPRINTF(O3CPU, "Deleting instructions from instruction " 1309 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1310 tid, seq_num, (*inst_iter)->seqNum); 1311 1312 while ((*inst_iter)->seqNum > seq_num) { 1313 1314 bool break_loop = (inst_iter == instList.begin()); 1315 1316 squashInstIt(inst_iter, tid); 1317 1318 inst_iter--; 1319 1320 if (break_loop) 1321 break; 1322 } 1323} 1324 1325template <class Impl> 1326inline void 1327FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1328{ 1329 if ((*instIt)->threadNumber == tid) { 1330 DPRINTF(O3CPU, "Squashing instruction, " 1331 "[tid:%i] [sn:%lli] PC %#x\n", 1332 (*instIt)->threadNumber, 1333 (*instIt)->seqNum, 1334 (*instIt)->readPC()); 1335 1336 // Mark it as squashed. 1337 (*instIt)->setSquashed(); 1338 1339 // @todo: Formulate a consistent method for deleting 1340 // instructions from the instruction list 1341 // Remove the instruction from the list. 1342 removeList.push(instIt); 1343 } 1344} 1345 1346template <class Impl> 1347void 1348FullO3CPU<Impl>::cleanUpRemovedInsts() 1349{ 1350 while (!removeList.empty()) { 1351 DPRINTF(O3CPU, "Removing instruction, " 1352 "[tid:%i] [sn:%lli] PC %#x\n", 1353 (*removeList.front())->threadNumber, 1354 (*removeList.front())->seqNum, 1355 (*removeList.front())->readPC()); 1356 1357 instList.erase(removeList.front()); 1358 1359 removeList.pop(); 1360 } 1361 1362 removeInstsThisCycle = false; 1363} 1364/* 1365template <class Impl> 1366void 1367FullO3CPU<Impl>::removeAllInsts() 1368{ 1369 instList.clear(); 1370} 1371*/ 1372template <class Impl> 1373void 1374FullO3CPU<Impl>::dumpInsts() 1375{ 1376 int num = 0; 1377 1378 ListIt inst_list_it = instList.begin(); 1379 1380 cprintf("Dumping Instruction List\n"); 1381 1382 while (inst_list_it != instList.end()) { 1383 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1384 "Squashed:%i\n\n", 1385 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1386 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1387 (*inst_list_it)->isSquashed()); 1388 inst_list_it++; 1389 ++num; 1390 } 1391} 1392/* 1393template <class Impl> 1394void 1395FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1396{ 1397 iew.wakeDependents(inst); 1398} 1399*/ 1400template <class Impl> 1401void 1402FullO3CPU<Impl>::wakeCPU() 1403{ 1404 if (activityRec.active() || tickEvent.scheduled()) { 1405 DPRINTF(Activity, "CPU already running.\n"); 1406 return; 1407 } 1408 1409 DPRINTF(Activity, "Waking up CPU\n"); 1410 1411 idleCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1412 numCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1413 1414 tickEvent.schedule(nextCycle()); 1415} 1416 1417template <class Impl> 1418int 1419FullO3CPU<Impl>::getFreeTid() 1420{ 1421 for (int i=0; i < numThreads; i++) { 1422 if (!tids[i]) { 1423 tids[i] = true; 1424 return i; 1425 } 1426 } 1427 1428 return -1; 1429} 1430 1431template <class Impl> 1432void 1433FullO3CPU<Impl>::doContextSwitch() 1434{ 1435 if (contextSwitch) { 1436 1437 //ADD CODE TO DEACTIVE THREAD HERE (???) 1438 1439 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1440 activateWhenReady(tid); 1441 } 1442 1443 if (cpuWaitList.size() == 0) 1444 contextSwitch = true; 1445 } 1446} 1447 1448template <class Impl> 1449void 1450FullO3CPU<Impl>::updateThreadPriority() 1451{ 1452 if (activeThreads.size() > 1) 1453 { 1454 //DEFAULT TO ROUND ROBIN SCHEME 1455 //e.g. Move highest priority to end of thread list 1456 std::list<unsigned>::iterator list_begin = activeThreads.begin(); 1457 std::list<unsigned>::iterator list_end = activeThreads.end(); 1458 1459 unsigned high_thread = *list_begin; 1460 1461 activeThreads.erase(list_begin); 1462 1463 activeThreads.push_back(high_thread); 1464 } 1465} 1466 1467// Forward declaration of FullO3CPU. 1468template class FullO3CPU<O3CPUImpl>; 1469