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>::FullO3CPU(Params *params) 91 : BaseO3CPU(params), 92 tickEvent(this), 93 removeInstsThisCycle(false), 94 fetch(params), 95 decode(params), 96 rename(params), 97 iew(params), 98 commit(params), 99 100 regFile(params->numPhysIntRegs, params->numPhysFloatRegs), 101 102 freeList(params->numberOfThreads,//number of activeThreads 103 TheISA::NumIntRegs, params->numPhysIntRegs, 104 TheISA::NumFloatRegs, params->numPhysFloatRegs), 105 106 rob(params->numROBEntries, params->squashWidth, 107 params->smtROBPolicy, params->smtROBThreshold, 108 params->numberOfThreads), 109 110 scoreboard(params->numberOfThreads,//number of activeThreads 111 TheISA::NumIntRegs, params->numPhysIntRegs, 112 TheISA::NumFloatRegs, params->numPhysFloatRegs, 113 TheISA::NumMiscRegs * number_of_threads, 114 TheISA::ZeroReg), 115 116 // For now just have these time buffers be pretty big. 117 // @todo: Make these time buffer sizes parameters or derived 118 // from latencies 119 timeBuffer(5, 5), 120 fetchQueue(5, 5), 121 decodeQueue(5, 5), 122 renameQueue(5, 5), 123 iewQueue(5, 5), 124 activityRec(NumStages, 10, params->activity), 125 126 globalSeqNum(1), 127 128#if FULL_SYSTEM 129 system(params->system), 130 physmem(system->physmem), 131#endif // FULL_SYSTEM 132 mem(params->mem), 133 switchCount(0), 134 deferRegistration(params->deferRegistration), 135 numThreads(number_of_threads) 136{ 137 _status = Idle; 138 139 checker = NULL; 140 141 if (params->checker) { 142#if USE_CHECKER 143 BaseCPU *temp_checker = params->checker; 144 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 145 checker->setMemory(mem); 146#if FULL_SYSTEM 147 checker->setSystem(params->system); 148#endif 149#else 150 panic("Checker enabled but not compiled in!"); 151#endif // USE_CHECKER 152 } 153 154#if !FULL_SYSTEM 155 thread.resize(number_of_threads); 156 tids.resize(number_of_threads); 157#endif 158 159 // The stages also need their CPU pointer setup. However this 160 // must be done at the upper level CPU because they have pointers 161 // to the upper level CPU, and not this FullO3CPU. 162 163 // Set up Pointers to the activeThreads list for each stage 164 fetch.setActiveThreads(&activeThreads); 165 decode.setActiveThreads(&activeThreads); 166 rename.setActiveThreads(&activeThreads); 167 iew.setActiveThreads(&activeThreads); 168 commit.setActiveThreads(&activeThreads); 169 170 // Give each of the stages the time buffer they will use. 171 fetch.setTimeBuffer(&timeBuffer); 172 decode.setTimeBuffer(&timeBuffer); 173 rename.setTimeBuffer(&timeBuffer); 174 iew.setTimeBuffer(&timeBuffer); 175 commit.setTimeBuffer(&timeBuffer); 176 177 // Also setup each of the stages' queues. 178 fetch.setFetchQueue(&fetchQueue); 179 decode.setFetchQueue(&fetchQueue); 180 commit.setFetchQueue(&fetchQueue); 181 decode.setDecodeQueue(&decodeQueue); 182 rename.setDecodeQueue(&decodeQueue); 183 rename.setRenameQueue(&renameQueue); 184 iew.setRenameQueue(&renameQueue); 185 iew.setIEWQueue(&iewQueue); 186 commit.setIEWQueue(&iewQueue); 187 commit.setRenameQueue(&renameQueue); 188 189 commit.setFetchStage(&fetch); 190 commit.setIEWStage(&iew); 191 rename.setIEWStage(&iew); 192 rename.setCommitStage(&commit); 193 194#if !FULL_SYSTEM 195 int active_threads = params->workload.size(); 196#else 197 int active_threads = 1; 198#endif 199 200 //Make Sure That this a Valid Architeture 201 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 202 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 203 204 rename.setScoreboard(&scoreboard); 205 iew.setScoreboard(&scoreboard); 206 207 // Setup the rename map for whichever stages need it. 208 PhysRegIndex lreg_idx = 0; 209 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 210 211 for (int tid=0; tid < numThreads; tid++) { 212 bool bindRegs = (tid <= active_threads - 1); 213 214 commitRenameMap[tid].init(TheISA::NumIntRegs, 215 params->numPhysIntRegs, 216 lreg_idx, //Index for Logical. Regs 217 218 TheISA::NumFloatRegs, 219 params->numPhysFloatRegs, 220 freg_idx, //Index for Float Regs 221 222 TheISA::NumMiscRegs, 223 224 TheISA::ZeroReg, 225 TheISA::ZeroReg, 226 227 tid, 228 false); 229 230 renameMap[tid].init(TheISA::NumIntRegs, 231 params->numPhysIntRegs, 232 lreg_idx, //Index for Logical. Regs 233 234 TheISA::NumFloatRegs, 235 params->numPhysFloatRegs, 236 freg_idx, //Index for Float Regs 237 238 TheISA::NumMiscRegs, 239 240 TheISA::ZeroReg, 241 TheISA::ZeroReg, 242 243 tid, 244 bindRegs); 245 } 246 247 rename.setRenameMap(renameMap); 248 commit.setRenameMap(commitRenameMap); 249 250 // Give renameMap & rename stage access to the freeList; 251 for (int i=0; i < numThreads; i++) { 252 renameMap[i].setFreeList(&freeList); 253 } 254 rename.setFreeList(&freeList); 255 256 // Setup the ROB for whichever stages need it. 257 commit.setROB(&rob); 258 259 lastRunningCycle = curTick; 260 261 contextSwitch = false; 262} 263 264template <class Impl> 265FullO3CPU<Impl>::~FullO3CPU() 266{ 267} 268 269template <class Impl> 270void 271FullO3CPU<Impl>::fullCPURegStats() 272{ 273 BaseO3CPU::regStats(); 274 275 // Register any of the O3CPU's stats here. 276 timesIdled 277 .name(name() + ".timesIdled") 278 .desc("Number of times that the entire CPU went into an idle state and" 279 " unscheduled itself") 280 .prereq(timesIdled); 281 282 idleCycles 283 .name(name() + ".idleCycles") 284 .desc("Total number of cycles that the CPU has spent unscheduled due " 285 "to idling") 286 .prereq(idleCycles); 287 288 // Number of Instructions simulated 289 // -------------------------------- 290 // Should probably be in Base CPU but need templated 291 // MaxThreads so put in here instead 292 committedInsts 293 .init(numThreads) 294 .name(name() + ".committedInsts") 295 .desc("Number of Instructions Simulated"); 296 297 totalCommittedInsts 298 .name(name() + ".committedInsts_total") 299 .desc("Number of Instructions Simulated"); 300 301 cpi 302 .name(name() + ".cpi") 303 .desc("CPI: Cycles Per Instruction") 304 .precision(6); 305 cpi = simTicks / committedInsts; 306 307 totalCpi 308 .name(name() + ".cpi_total") 309 .desc("CPI: Total CPI of All Threads") 310 .precision(6); 311 totalCpi = simTicks / totalCommittedInsts; 312 313 ipc 314 .name(name() + ".ipc") 315 .desc("IPC: Instructions Per Cycle") 316 .precision(6); 317 ipc = committedInsts / simTicks; 318 319 totalIpc 320 .name(name() + ".ipc_total") 321 .desc("IPC: Total IPC of All Threads") 322 .precision(6); 323 totalIpc = totalCommittedInsts / simTicks; 324 325} 326 327template <class Impl> 328void 329FullO3CPU<Impl>::tick() 330{ 331 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 332 333 ++numCycles; 334 335// activity = false; 336 337 //Tick each of the stages 338 fetch.tick(); 339 340 decode.tick(); 341 342 rename.tick(); 343 344 iew.tick(); 345 346 commit.tick(); 347 348#if !FULL_SYSTEM 349 doContextSwitch(); 350#endif 351 352 // Now advance the time buffers 353 timeBuffer.advance(); 354 355 fetchQueue.advance(); 356 decodeQueue.advance(); 357 renameQueue.advance(); 358 iewQueue.advance(); 359 360 activityRec.advance(); 361 362 if (removeInstsThisCycle) { 363 cleanUpRemovedInsts(); 364 } 365 366 if (!tickEvent.scheduled()) { 367 if (_status == SwitchedOut) { 368 // increment stat 369 lastRunningCycle = curTick; 370 } else if (!activityRec.active()) { 371 lastRunningCycle = curTick; 372 timesIdled++; 373 } else { 374 tickEvent.schedule(curTick + cycles(1)); 375 } 376 } 377 378#if !FULL_SYSTEM 379 updateThreadPriority(); 380#endif 381 382} 383 384template <class Impl> 385void 386FullO3CPU<Impl>::init() 387{ 388 if (!deferRegistration) { 389 registerThreadContexts(); 390 } 391 392 // Set inSyscall so that the CPU doesn't squash when initially 393 // setting up registers. 394 for (int i = 0; i < number_of_threads; ++i) 395 thread[i]->inSyscall = true; 396 397 for (int tid=0; tid < number_of_threads; tid++) { 398#if FULL_SYSTEM 399 ThreadContext *src_tc = threadContexts[tid]; 400#else 401 ThreadContext *src_tc = thread[tid]->getTC(); 402#endif 403 // Threads start in the Suspended State 404 if (src_tc->status() != ThreadContext::Suspended) { 405 continue; 406 } 407 408#if FULL_SYSTEM 409 TheISA::initCPU(src_tc, src_tc->readCpuId()); 410#endif 411 } 412 413 // Clear inSyscall. 414 for (int i = 0; i < number_of_threads; ++i) 415 thread[i]->inSyscall = false; 416 417 // Initialize stages. 418 fetch.initStage(); 419 iew.initStage(); 420 rename.initStage(); 421 commit.initStage(); 422 423 commit.setThreads(thread); 424} 425 426template <class Impl> 427void 428FullO3CPU<Impl>::insertThread(unsigned tid) 429{ 430 DPRINTF(O3CPU,"[tid:%i] Initializing thread data"); 431 // Will change now that the PC and thread state is internal to the CPU 432 // and not in the ThreadContext. 433#if 0 434#if FULL_SYSTEM 435 ThreadContext *src_tc = system->threadContexts[tid]; 436#else 437 ThreadContext *src_tc = thread[tid]; 438#endif 439 440 //Bind Int Regs to Rename Map 441 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 442 PhysRegIndex phys_reg = freeList.getIntReg(); 443 444 renameMap[tid].setEntry(ireg,phys_reg); 445 scoreboard.setReg(phys_reg); 446 } 447 448 //Bind Float Regs to Rename Map 449 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 450 PhysRegIndex phys_reg = freeList.getFloatReg(); 451 452 renameMap[tid].setEntry(freg,phys_reg); 453 scoreboard.setReg(phys_reg); 454 } 455 456 //Copy Thread Data Into RegFile 457 this->copyFromTC(tid); 458 459 //Set PC/NPC 460 regFile.pc[tid] = src_tc->readPC(); 461 regFile.npc[tid] = src_tc->readNextPC(); 462 463 src_tc->setStatus(ThreadContext::Active); 464 465 activateContext(tid,1); 466 467 //Reset ROB/IQ/LSQ Entries 468 commit.rob->resetEntries(); 469 iew.resetEntries(); 470#endif 471} 472 473template <class Impl> 474void 475FullO3CPU<Impl>::removeThread(unsigned tid) 476{ 477 DPRINTF(O3CPU,"[tid:%i] Removing thread data"); 478#if 0 479 //Unbind Int Regs from Rename Map 480 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 481 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 482 483 scoreboard.unsetReg(phys_reg); 484 freeList.addReg(phys_reg); 485 } 486 487 //Unbind Float Regs from Rename Map 488 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 489 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 490 491 scoreboard.unsetReg(phys_reg); 492 freeList.addReg(phys_reg); 493 } 494 495 //Copy Thread Data From RegFile 496 /* Fix Me: 497 * Do we really need to do this if we are removing a thread 498 * in the sense that it's finished (exiting)? If the thread is just 499 * being suspended we might... 500 */ 501// this->copyToTC(tid); 502 503 //Squash Throughout Pipeline 504 fetch.squash(0,tid); 505 decode.squash(tid); 506 rename.squash(tid); 507 508 assert(iew.ldstQueue.getCount(tid) == 0); 509 510 //Reset ROB/IQ/LSQ Entries 511 if (activeThreads.size() >= 1) { 512 commit.rob->resetEntries(); 513 iew.resetEntries(); 514 } 515#endif 516} 517 518 519template <class Impl> 520void 521FullO3CPU<Impl>::activateWhenReady(int tid) 522{ 523 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 524 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 525 tid); 526 527 bool ready = true; 528 529 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 530 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 531 "Phys. Int. Regs.\n", 532 tid); 533 ready = false; 534 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 535 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 536 "Phys. Float. Regs.\n", 537 tid); 538 ready = false; 539 } else if (commit.rob->numFreeEntries() >= 540 commit.rob->entryAmount(activeThreads.size() + 1)) { 541 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 542 "ROB entries.\n", 543 tid); 544 ready = false; 545 } else if (iew.instQueue.numFreeEntries() >= 546 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 547 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 548 "IQ entries.\n", 549 tid); 550 ready = false; 551 } else if (iew.ldstQueue.numFreeEntries() >= 552 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 553 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 554 "LSQ entries.\n", 555 tid); 556 ready = false; 557 } 558 559 if (ready) { 560 insertThread(tid); 561 562 contextSwitch = false; 563 564 cpuWaitList.remove(tid); 565 } else { 566 suspendContext(tid); 567 568 //blocks fetch 569 contextSwitch = true; 570 571 //do waitlist 572 cpuWaitList.push_back(tid); 573 } 574} 575 576template <class Impl> 577void 578FullO3CPU<Impl>::activateContext(int tid, int delay) 579{ 580 // Needs to set each stage to running as well. 581 list<unsigned>::iterator isActive = find( 582 activeThreads.begin(), activeThreads.end(), tid); 583 584 if (isActive == activeThreads.end()) { 585 //May Need to Re-code this if the delay variable is the 586 //delay needed for thread to activate 587 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 588 tid); 589 590 activeThreads.push_back(tid); 591 } 592 593 assert(_status == Idle || _status == SwitchedOut); 594 595 scheduleTickEvent(delay); 596 597 // Be sure to signal that there's some activity so the CPU doesn't 598 // deschedule itself. 599 activityRec.activity(); 600 fetch.wakeFromQuiesce(); 601 602 _status = Running; 603} 604 605template <class Impl> 606void 607FullO3CPU<Impl>::suspendContext(int tid) 608{ 609 DPRINTF(O3CPU,"[tid: %i]: Suspended ...\n", tid); 610 unscheduleTickEvent(); 611 _status = Idle; 612/* 613 //Remove From Active List, if Active 614 list<unsigned>::iterator isActive = find( 615 activeThreads.begin(), activeThreads.end(), tid); 616 617 if (isActive != activeThreads.end()) { 618 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 619 tid); 620 activeThreads.erase(isActive); 621 } 622*/ 623} 624 625template <class Impl> 626void 627FullO3CPU<Impl>::deallocateContext(int tid) 628{ 629 DPRINTF(O3CPU,"[tid:%i]: Deallocating ...", tid); 630/* 631 //Remove From Active List, if Active 632 list<unsigned>::iterator isActive = find( 633 activeThreads.begin(), activeThreads.end(), tid); 634 635 if (isActive != activeThreads.end()) { 636 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 637 tid); 638 activeThreads.erase(isActive); 639 640 removeThread(tid); 641 } 642*/ 643} 644 645template <class Impl> 646void 647FullO3CPU<Impl>::haltContext(int tid) 648{ 649 DPRINTF(O3CPU,"[tid:%i]: Halted ...", tid); 650/* 651 //Remove From Active List, if Active 652 list<unsigned>::iterator isActive = find( 653 activeThreads.begin(), activeThreads.end(), tid); 654 655 if (isActive != activeThreads.end()) { 656 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 657 tid); 658 activeThreads.erase(isActive); 659 660 removeThread(tid); 661 } 662*/ 663} 664 665template <class Impl> 666void 667FullO3CPU<Impl>::switchOut(Sampler *_sampler) 668{ 669 sampler = _sampler; 670 switchCount = 0; 671 fetch.switchOut(); 672 decode.switchOut(); 673 rename.switchOut(); 674 iew.switchOut(); 675 commit.switchOut(); 676 677 // Wake the CPU and record activity so everything can drain out if 678 // the CPU is currently idle. 679 wakeCPU(); 680 activityRec.activity(); 681} 682 683template <class Impl> 684void 685FullO3CPU<Impl>::signalSwitched() 686{ 687 if (++switchCount == NumStages) { 688 fetch.doSwitchOut(); 689 rename.doSwitchOut(); 690 commit.doSwitchOut(); 691 instList.clear(); 692 while (!removeList.empty()) { 693 removeList.pop(); 694 } 695 696#if USE_CHECKER 697 if (checker) 698 checker->switchOut(sampler); 699#endif 700 701 if (tickEvent.scheduled()) 702 tickEvent.squash(); 703 sampler->signalSwitched(); 704 _status = SwitchedOut; 705 } 706 assert(switchCount <= 5); 707} 708 709template <class Impl> 710void 711FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 712{ 713 // Flush out any old data from the time buffers. 714 for (int i = 0; i < 10; ++i) { 715 timeBuffer.advance(); 716 fetchQueue.advance(); 717 decodeQueue.advance(); 718 renameQueue.advance(); 719 iewQueue.advance(); 720 } 721 722 activityRec.reset(); 723 724 BaseCPU::takeOverFrom(oldCPU); 725 726 fetch.takeOverFrom(); 727 decode.takeOverFrom(); 728 rename.takeOverFrom(); 729 iew.takeOverFrom(); 730 commit.takeOverFrom(); 731 732 assert(!tickEvent.scheduled()); 733 734 // @todo: Figure out how to properly select the tid to put onto 735 // the active threads list. 736 int tid = 0; 737 738 list<unsigned>::iterator isActive = find( 739 activeThreads.begin(), activeThreads.end(), tid); 740 741 if (isActive == activeThreads.end()) { 742 //May Need to Re-code this if the delay variable is the delay 743 //needed for thread to activate 744 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 745 tid); 746 747 activeThreads.push_back(tid); 748 } 749 750 // Set all statuses to active, schedule the CPU's tick event. 751 // @todo: Fix up statuses so this is handled properly 752 for (int i = 0; i < threadContexts.size(); ++i) { 753 ThreadContext *tc = threadContexts[i]; 754 if (tc->status() == ThreadContext::Active && _status != Running) { 755 _status = Running; 756 tickEvent.schedule(curTick); 757 } 758 } 759 if (!tickEvent.scheduled()) 760 tickEvent.schedule(curTick); 761} 762 763template <class Impl> 764uint64_t 765FullO3CPU<Impl>::readIntReg(int reg_idx) 766{ 767 return regFile.readIntReg(reg_idx); 768} 769 770template <class Impl> 771FloatReg 772FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 773{ 774 return regFile.readFloatReg(reg_idx, width); 775} 776 777template <class Impl> 778FloatReg 779FullO3CPU<Impl>::readFloatReg(int reg_idx) 780{ 781 return regFile.readFloatReg(reg_idx); 782} 783 784template <class Impl> 785FloatRegBits 786FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 787{ 788 return regFile.readFloatRegBits(reg_idx, width); 789} 790 791template <class Impl> 792FloatRegBits 793FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 794{ 795 return regFile.readFloatRegBits(reg_idx); 796} 797 798template <class Impl> 799void 800FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 801{ 802 regFile.setIntReg(reg_idx, val); 803} 804 805template <class Impl> 806void 807FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 808{ 809 regFile.setFloatReg(reg_idx, val, width); 810} 811 812template <class Impl> 813void 814FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 815{ 816 regFile.setFloatReg(reg_idx, val); 817} 818 819template <class Impl> 820void 821FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 822{ 823 regFile.setFloatRegBits(reg_idx, val, width); 824} 825 826template <class Impl> 827void 828FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 829{ 830 regFile.setFloatRegBits(reg_idx, val); 831} 832 833template <class Impl> 834uint64_t 835FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 836{ 837 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 838 839 return regFile.readIntReg(phys_reg); 840} 841 842template <class Impl> 843float 844FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 845{ 846 int idx = reg_idx + TheISA::FP_Base_DepTag; 847 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 848 849 return regFile.readFloatReg(phys_reg); 850} 851 852template <class Impl> 853double 854FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 855{ 856 int idx = reg_idx + TheISA::FP_Base_DepTag; 857 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 858 859 return regFile.readFloatReg(phys_reg, 64); 860} 861 862template <class Impl> 863uint64_t 864FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 865{ 866 int idx = reg_idx + TheISA::FP_Base_DepTag; 867 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 868 869 return regFile.readFloatRegBits(phys_reg); 870} 871 872template <class Impl> 873void 874FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 875{ 876 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 877 878 regFile.setIntReg(phys_reg, val); 879} 880 881template <class Impl> 882void 883FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 884{ 885 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 886 887 regFile.setFloatReg(phys_reg, val); 888} 889 890template <class Impl> 891void 892FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 893{ 894 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 895 896 regFile.setFloatReg(phys_reg, val, 64); 897} 898 899template <class Impl> 900void 901FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 902{ 903 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 904 905 regFile.setFloatRegBits(phys_reg, val); 906} 907 908template <class Impl> 909uint64_t 910FullO3CPU<Impl>::readPC(unsigned tid) 911{ 912 return commit.readPC(tid); 913} 914 915template <class Impl> 916void 917FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 918{ 919 commit.setPC(new_PC, tid); 920} 921 922template <class Impl> 923uint64_t 924FullO3CPU<Impl>::readNextPC(unsigned tid) 925{ 926 return commit.readNextPC(tid); 927} 928 929template <class Impl> 930void 931FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 932{ 933 commit.setNextPC(val, tid); 934} 935 936#if THE_ISA != ALPHA_ISA 937template <class Impl> 938uint64_t 939FullO3CPU<Impl>::readNextNPC(unsigned tid) 940{ 941 return commit.readNextNPC(tid); 942} 943 944template <class Impl> 945void 946FullO3CPU<Impl>::setNextNNPC(uint64_t val,unsigned tid) 947{ 948 commit.setNextNPC(val, tid); 949} 950#endif 951 952template <class Impl> 953typename FullO3CPU<Impl>::ListIt 954FullO3CPU<Impl>::addInst(DynInstPtr &inst) 955{ 956 instList.push_back(inst); 957 958 return --(instList.end()); 959} 960 961template <class Impl> 962void 963FullO3CPU<Impl>::instDone(unsigned tid) 964{ 965 // Keep an instruction count. 966 thread[tid]->numInst++; 967 thread[tid]->numInsts++; 968 committedInsts[tid]++; 969 totalCommittedInsts++; 970 971 // Check for instruction-count-based events. 972 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 973} 974 975template <class Impl> 976void 977FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 978{ 979 removeInstsThisCycle = true; 980 981 removeList.push(inst->getInstListIt()); 982} 983 984template <class Impl> 985void 986FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 987{ 988 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 989 "[sn:%lli]\n", 990 inst->threadNumber, inst->readPC(), inst->seqNum); 991 992 removeInstsThisCycle = true; 993 994 // Remove the front instruction. 995 removeList.push(inst->getInstListIt()); 996} 997 998template <class Impl> 999void 1000FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 1001{ 1002 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1003 " list.\n", tid); 1004 1005 ListIt end_it; 1006 1007 bool rob_empty = false; 1008 1009 if (instList.empty()) { 1010 return; 1011 } else if (rob.isEmpty(/*tid*/)) { 1012 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1013 end_it = instList.begin(); 1014 rob_empty = true; 1015 } else { 1016 end_it = (rob.readTailInst(tid))->getInstListIt(); 1017 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1018 } 1019 1020 removeInstsThisCycle = true; 1021 1022 ListIt inst_it = instList.end(); 1023 1024 inst_it--; 1025 1026 // Walk through the instruction list, removing any instructions 1027 // that were inserted after the given instruction iterator, end_it. 1028 while (inst_it != end_it) { 1029 assert(!instList.empty()); 1030 1031 squashInstIt(inst_it, tid); 1032 1033 inst_it--; 1034 } 1035 1036 // If the ROB was empty, then we actually need to remove the first 1037 // instruction as well. 1038 if (rob_empty) { 1039 squashInstIt(inst_it, tid); 1040 } 1041} 1042 1043template <class Impl> 1044void 1045FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1046 unsigned tid) 1047{ 1048 assert(!instList.empty()); 1049 1050 removeInstsThisCycle = true; 1051 1052 ListIt inst_iter = instList.end(); 1053 1054 inst_iter--; 1055 1056 DPRINTF(O3CPU, "Deleting instructions from instruction " 1057 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1058 tid, seq_num, (*inst_iter)->seqNum); 1059 1060 while ((*inst_iter)->seqNum > seq_num) { 1061 1062 bool break_loop = (inst_iter == instList.begin()); 1063 1064 squashInstIt(inst_iter, tid); 1065 1066 inst_iter--; 1067 1068 if (break_loop) 1069 break; 1070 } 1071} 1072 1073template <class Impl> 1074inline void 1075FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1076{ 1077 if ((*instIt)->threadNumber == tid) { 1078 DPRINTF(O3CPU, "Squashing instruction, " 1079 "[tid:%i] [sn:%lli] PC %#x\n", 1080 (*instIt)->threadNumber, 1081 (*instIt)->seqNum, 1082 (*instIt)->readPC()); 1083 1084 // Mark it as squashed. 1085 (*instIt)->setSquashed(); 1086 1087 // @todo: Formulate a consistent method for deleting 1088 // instructions from the instruction list 1089 // Remove the instruction from the list. 1090 removeList.push(instIt); 1091 } 1092} 1093 1094template <class Impl> 1095void 1096FullO3CPU<Impl>::cleanUpRemovedInsts() 1097{ 1098 while (!removeList.empty()) { 1099 DPRINTF(O3CPU, "Removing instruction, " 1100 "[tid:%i] [sn:%lli] PC %#x\n", 1101 (*removeList.front())->threadNumber, 1102 (*removeList.front())->seqNum, 1103 (*removeList.front())->readPC()); 1104 1105 instList.erase(removeList.front()); 1106 1107 removeList.pop(); 1108 } 1109 1110 removeInstsThisCycle = false; 1111} 1112/* 1113template <class Impl> 1114void 1115FullO3CPU<Impl>::removeAllInsts() 1116{ 1117 instList.clear(); 1118} 1119*/ 1120template <class Impl> 1121void 1122FullO3CPU<Impl>::dumpInsts() 1123{ 1124 int num = 0; 1125 1126 ListIt inst_list_it = instList.begin(); 1127 1128 cprintf("Dumping Instruction List\n"); 1129 1130 while (inst_list_it != instList.end()) { 1131 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1132 "Squashed:%i\n\n", 1133 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1134 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1135 (*inst_list_it)->isSquashed()); 1136 inst_list_it++; 1137 ++num; 1138 } 1139} 1140/* 1141template <class Impl> 1142void 1143FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1144{ 1145 iew.wakeDependents(inst); 1146} 1147*/ 1148template <class Impl> 1149void 1150FullO3CPU<Impl>::wakeCPU() 1151{ 1152 if (activityRec.active() || tickEvent.scheduled()) { 1153 DPRINTF(Activity, "CPU already running.\n"); 1154 return; 1155 } 1156 1157 DPRINTF(Activity, "Waking up CPU\n"); 1158 1159 idleCycles += (curTick - 1) - lastRunningCycle; 1160 1161 tickEvent.schedule(curTick); 1162} 1163 1164template <class Impl> 1165int 1166FullO3CPU<Impl>::getFreeTid() 1167{ 1168 for (int i=0; i < numThreads; i++) { 1169 if (!tids[i]) { 1170 tids[i] = true; 1171 return i; 1172 } 1173 } 1174 1175 return -1; 1176} 1177 1178template <class Impl> 1179void 1180FullO3CPU<Impl>::doContextSwitch() 1181{ 1182 if (contextSwitch) { 1183 1184 //ADD CODE TO DEACTIVE THREAD HERE (???) 1185 1186 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1187 activateWhenReady(tid); 1188 } 1189 1190 if (cpuWaitList.size() == 0) 1191 contextSwitch = true; 1192 } 1193} 1194 1195template <class Impl> 1196void 1197FullO3CPU<Impl>::updateThreadPriority() 1198{ 1199 if (activeThreads.size() > 1) 1200 { 1201 //DEFAULT TO ROUND ROBIN SCHEME 1202 //e.g. Move highest priority to end of thread list 1203 list<unsigned>::iterator list_begin = activeThreads.begin(); 1204 list<unsigned>::iterator list_end = activeThreads.end(); 1205 1206 unsigned high_thread = *list_begin; 1207 1208 activeThreads.erase(list_begin); 1209 1210 activeThreads.push_back(high_thread); 1211 } 1212} 1213 1214// Forward declaration of FullO3CPU. 1215template class FullO3CPU<AlphaSimpleImpl>;
| 45#include "cpu/o3/cpu.hh" 46 47#include "sim/root.hh" 48#include "sim/stat_control.hh" 49 50#if USE_CHECKER 51#include "cpu/checker/cpu.hh" 52#endif 53 54using namespace std; 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() 84{ 85 return "FullO3CPU tick event"; 86} 87 88template <class Impl> 89FullO3CPU<Impl>::FullO3CPU(Params *params) 90 : BaseO3CPU(params), 91 tickEvent(this), 92 removeInstsThisCycle(false), 93 fetch(params), 94 decode(params), 95 rename(params), 96 iew(params), 97 commit(params), 98 99 regFile(params->numPhysIntRegs, params->numPhysFloatRegs), 100 101 freeList(params->numberOfThreads,//number of activeThreads 102 TheISA::NumIntRegs, params->numPhysIntRegs, 103 TheISA::NumFloatRegs, params->numPhysFloatRegs), 104 105 rob(params->numROBEntries, params->squashWidth, 106 params->smtROBPolicy, params->smtROBThreshold, 107 params->numberOfThreads), 108 109 scoreboard(params->numberOfThreads,//number of activeThreads 110 TheISA::NumIntRegs, params->numPhysIntRegs, 111 TheISA::NumFloatRegs, params->numPhysFloatRegs, 112 TheISA::NumMiscRegs * number_of_threads, 113 TheISA::ZeroReg), 114 115 // For now just have these time buffers be pretty big. 116 // @todo: Make these time buffer sizes parameters or derived 117 // from latencies 118 timeBuffer(5, 5), 119 fetchQueue(5, 5), 120 decodeQueue(5, 5), 121 renameQueue(5, 5), 122 iewQueue(5, 5), 123 activityRec(NumStages, 10, params->activity), 124 125 globalSeqNum(1), 126 127#if FULL_SYSTEM 128 system(params->system), 129 physmem(system->physmem), 130#endif // FULL_SYSTEM 131 mem(params->mem), 132 switchCount(0), 133 deferRegistration(params->deferRegistration), 134 numThreads(number_of_threads) 135{ 136 _status = Idle; 137 138 checker = NULL; 139 140 if (params->checker) { 141#if USE_CHECKER 142 BaseCPU *temp_checker = params->checker; 143 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 144 checker->setMemory(mem); 145#if FULL_SYSTEM 146 checker->setSystem(params->system); 147#endif 148#else 149 panic("Checker enabled but not compiled in!"); 150#endif // USE_CHECKER 151 } 152 153#if !FULL_SYSTEM 154 thread.resize(number_of_threads); 155 tids.resize(number_of_threads); 156#endif 157 158 // The stages also need their CPU pointer setup. However this 159 // must be done at the upper level CPU because they have pointers 160 // to the upper level CPU, and not this FullO3CPU. 161 162 // Set up Pointers to the activeThreads list for each stage 163 fetch.setActiveThreads(&activeThreads); 164 decode.setActiveThreads(&activeThreads); 165 rename.setActiveThreads(&activeThreads); 166 iew.setActiveThreads(&activeThreads); 167 commit.setActiveThreads(&activeThreads); 168 169 // Give each of the stages the time buffer they will use. 170 fetch.setTimeBuffer(&timeBuffer); 171 decode.setTimeBuffer(&timeBuffer); 172 rename.setTimeBuffer(&timeBuffer); 173 iew.setTimeBuffer(&timeBuffer); 174 commit.setTimeBuffer(&timeBuffer); 175 176 // Also setup each of the stages' queues. 177 fetch.setFetchQueue(&fetchQueue); 178 decode.setFetchQueue(&fetchQueue); 179 commit.setFetchQueue(&fetchQueue); 180 decode.setDecodeQueue(&decodeQueue); 181 rename.setDecodeQueue(&decodeQueue); 182 rename.setRenameQueue(&renameQueue); 183 iew.setRenameQueue(&renameQueue); 184 iew.setIEWQueue(&iewQueue); 185 commit.setIEWQueue(&iewQueue); 186 commit.setRenameQueue(&renameQueue); 187 188 commit.setFetchStage(&fetch); 189 commit.setIEWStage(&iew); 190 rename.setIEWStage(&iew); 191 rename.setCommitStage(&commit); 192 193#if !FULL_SYSTEM 194 int active_threads = params->workload.size(); 195#else 196 int active_threads = 1; 197#endif 198 199 //Make Sure That this a Valid Architeture 200 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 201 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 202 203 rename.setScoreboard(&scoreboard); 204 iew.setScoreboard(&scoreboard); 205 206 // Setup the rename map for whichever stages need it. 207 PhysRegIndex lreg_idx = 0; 208 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 209 210 for (int tid=0; tid < numThreads; tid++) { 211 bool bindRegs = (tid <= active_threads - 1); 212 213 commitRenameMap[tid].init(TheISA::NumIntRegs, 214 params->numPhysIntRegs, 215 lreg_idx, //Index for Logical. Regs 216 217 TheISA::NumFloatRegs, 218 params->numPhysFloatRegs, 219 freg_idx, //Index for Float Regs 220 221 TheISA::NumMiscRegs, 222 223 TheISA::ZeroReg, 224 TheISA::ZeroReg, 225 226 tid, 227 false); 228 229 renameMap[tid].init(TheISA::NumIntRegs, 230 params->numPhysIntRegs, 231 lreg_idx, //Index for Logical. Regs 232 233 TheISA::NumFloatRegs, 234 params->numPhysFloatRegs, 235 freg_idx, //Index for Float Regs 236 237 TheISA::NumMiscRegs, 238 239 TheISA::ZeroReg, 240 TheISA::ZeroReg, 241 242 tid, 243 bindRegs); 244 } 245 246 rename.setRenameMap(renameMap); 247 commit.setRenameMap(commitRenameMap); 248 249 // Give renameMap & rename stage access to the freeList; 250 for (int i=0; i < numThreads; i++) { 251 renameMap[i].setFreeList(&freeList); 252 } 253 rename.setFreeList(&freeList); 254 255 // Setup the ROB for whichever stages need it. 256 commit.setROB(&rob); 257 258 lastRunningCycle = curTick; 259 260 contextSwitch = false; 261} 262 263template <class Impl> 264FullO3CPU<Impl>::~FullO3CPU() 265{ 266} 267 268template <class Impl> 269void 270FullO3CPU<Impl>::fullCPURegStats() 271{ 272 BaseO3CPU::regStats(); 273 274 // Register any of the O3CPU's stats here. 275 timesIdled 276 .name(name() + ".timesIdled") 277 .desc("Number of times that the entire CPU went into an idle state and" 278 " unscheduled itself") 279 .prereq(timesIdled); 280 281 idleCycles 282 .name(name() + ".idleCycles") 283 .desc("Total number of cycles that the CPU has spent unscheduled due " 284 "to idling") 285 .prereq(idleCycles); 286 287 // Number of Instructions simulated 288 // -------------------------------- 289 // Should probably be in Base CPU but need templated 290 // MaxThreads so put in here instead 291 committedInsts 292 .init(numThreads) 293 .name(name() + ".committedInsts") 294 .desc("Number of Instructions Simulated"); 295 296 totalCommittedInsts 297 .name(name() + ".committedInsts_total") 298 .desc("Number of Instructions Simulated"); 299 300 cpi 301 .name(name() + ".cpi") 302 .desc("CPI: Cycles Per Instruction") 303 .precision(6); 304 cpi = simTicks / committedInsts; 305 306 totalCpi 307 .name(name() + ".cpi_total") 308 .desc("CPI: Total CPI of All Threads") 309 .precision(6); 310 totalCpi = simTicks / totalCommittedInsts; 311 312 ipc 313 .name(name() + ".ipc") 314 .desc("IPC: Instructions Per Cycle") 315 .precision(6); 316 ipc = committedInsts / simTicks; 317 318 totalIpc 319 .name(name() + ".ipc_total") 320 .desc("IPC: Total IPC of All Threads") 321 .precision(6); 322 totalIpc = totalCommittedInsts / simTicks; 323 324} 325 326template <class Impl> 327void 328FullO3CPU<Impl>::tick() 329{ 330 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 331 332 ++numCycles; 333 334// activity = false; 335 336 //Tick each of the stages 337 fetch.tick(); 338 339 decode.tick(); 340 341 rename.tick(); 342 343 iew.tick(); 344 345 commit.tick(); 346 347#if !FULL_SYSTEM 348 doContextSwitch(); 349#endif 350 351 // Now advance the time buffers 352 timeBuffer.advance(); 353 354 fetchQueue.advance(); 355 decodeQueue.advance(); 356 renameQueue.advance(); 357 iewQueue.advance(); 358 359 activityRec.advance(); 360 361 if (removeInstsThisCycle) { 362 cleanUpRemovedInsts(); 363 } 364 365 if (!tickEvent.scheduled()) { 366 if (_status == SwitchedOut) { 367 // increment stat 368 lastRunningCycle = curTick; 369 } else if (!activityRec.active()) { 370 lastRunningCycle = curTick; 371 timesIdled++; 372 } else { 373 tickEvent.schedule(curTick + cycles(1)); 374 } 375 } 376 377#if !FULL_SYSTEM 378 updateThreadPriority(); 379#endif 380 381} 382 383template <class Impl> 384void 385FullO3CPU<Impl>::init() 386{ 387 if (!deferRegistration) { 388 registerThreadContexts(); 389 } 390 391 // Set inSyscall so that the CPU doesn't squash when initially 392 // setting up registers. 393 for (int i = 0; i < number_of_threads; ++i) 394 thread[i]->inSyscall = true; 395 396 for (int tid=0; tid < number_of_threads; tid++) { 397#if FULL_SYSTEM 398 ThreadContext *src_tc = threadContexts[tid]; 399#else 400 ThreadContext *src_tc = thread[tid]->getTC(); 401#endif 402 // Threads start in the Suspended State 403 if (src_tc->status() != ThreadContext::Suspended) { 404 continue; 405 } 406 407#if FULL_SYSTEM 408 TheISA::initCPU(src_tc, src_tc->readCpuId()); 409#endif 410 } 411 412 // Clear inSyscall. 413 for (int i = 0; i < number_of_threads; ++i) 414 thread[i]->inSyscall = false; 415 416 // Initialize stages. 417 fetch.initStage(); 418 iew.initStage(); 419 rename.initStage(); 420 commit.initStage(); 421 422 commit.setThreads(thread); 423} 424 425template <class Impl> 426void 427FullO3CPU<Impl>::insertThread(unsigned tid) 428{ 429 DPRINTF(O3CPU,"[tid:%i] Initializing thread data"); 430 // Will change now that the PC and thread state is internal to the CPU 431 // and not in the ThreadContext. 432#if 0 433#if FULL_SYSTEM 434 ThreadContext *src_tc = system->threadContexts[tid]; 435#else 436 ThreadContext *src_tc = thread[tid]; 437#endif 438 439 //Bind Int Regs to Rename Map 440 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 441 PhysRegIndex phys_reg = freeList.getIntReg(); 442 443 renameMap[tid].setEntry(ireg,phys_reg); 444 scoreboard.setReg(phys_reg); 445 } 446 447 //Bind Float Regs to Rename Map 448 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 449 PhysRegIndex phys_reg = freeList.getFloatReg(); 450 451 renameMap[tid].setEntry(freg,phys_reg); 452 scoreboard.setReg(phys_reg); 453 } 454 455 //Copy Thread Data Into RegFile 456 this->copyFromTC(tid); 457 458 //Set PC/NPC 459 regFile.pc[tid] = src_tc->readPC(); 460 regFile.npc[tid] = src_tc->readNextPC(); 461 462 src_tc->setStatus(ThreadContext::Active); 463 464 activateContext(tid,1); 465 466 //Reset ROB/IQ/LSQ Entries 467 commit.rob->resetEntries(); 468 iew.resetEntries(); 469#endif 470} 471 472template <class Impl> 473void 474FullO3CPU<Impl>::removeThread(unsigned tid) 475{ 476 DPRINTF(O3CPU,"[tid:%i] Removing thread data"); 477#if 0 478 //Unbind Int Regs from Rename Map 479 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 480 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 481 482 scoreboard.unsetReg(phys_reg); 483 freeList.addReg(phys_reg); 484 } 485 486 //Unbind Float Regs from Rename Map 487 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 488 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 489 490 scoreboard.unsetReg(phys_reg); 491 freeList.addReg(phys_reg); 492 } 493 494 //Copy Thread Data From RegFile 495 /* Fix Me: 496 * Do we really need to do this if we are removing a thread 497 * in the sense that it's finished (exiting)? If the thread is just 498 * being suspended we might... 499 */ 500// this->copyToTC(tid); 501 502 //Squash Throughout Pipeline 503 fetch.squash(0,tid); 504 decode.squash(tid); 505 rename.squash(tid); 506 507 assert(iew.ldstQueue.getCount(tid) == 0); 508 509 //Reset ROB/IQ/LSQ Entries 510 if (activeThreads.size() >= 1) { 511 commit.rob->resetEntries(); 512 iew.resetEntries(); 513 } 514#endif 515} 516 517 518template <class Impl> 519void 520FullO3CPU<Impl>::activateWhenReady(int tid) 521{ 522 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 523 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 524 tid); 525 526 bool ready = true; 527 528 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 529 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 530 "Phys. Int. Regs.\n", 531 tid); 532 ready = false; 533 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 534 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 535 "Phys. Float. Regs.\n", 536 tid); 537 ready = false; 538 } else if (commit.rob->numFreeEntries() >= 539 commit.rob->entryAmount(activeThreads.size() + 1)) { 540 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 541 "ROB entries.\n", 542 tid); 543 ready = false; 544 } else if (iew.instQueue.numFreeEntries() >= 545 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 546 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 547 "IQ entries.\n", 548 tid); 549 ready = false; 550 } else if (iew.ldstQueue.numFreeEntries() >= 551 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 552 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 553 "LSQ entries.\n", 554 tid); 555 ready = false; 556 } 557 558 if (ready) { 559 insertThread(tid); 560 561 contextSwitch = false; 562 563 cpuWaitList.remove(tid); 564 } else { 565 suspendContext(tid); 566 567 //blocks fetch 568 contextSwitch = true; 569 570 //do waitlist 571 cpuWaitList.push_back(tid); 572 } 573} 574 575template <class Impl> 576void 577FullO3CPU<Impl>::activateContext(int tid, int delay) 578{ 579 // Needs to set each stage to running as well. 580 list<unsigned>::iterator isActive = find( 581 activeThreads.begin(), activeThreads.end(), tid); 582 583 if (isActive == activeThreads.end()) { 584 //May Need to Re-code this if the delay variable is the 585 //delay needed for thread to activate 586 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 587 tid); 588 589 activeThreads.push_back(tid); 590 } 591 592 assert(_status == Idle || _status == SwitchedOut); 593 594 scheduleTickEvent(delay); 595 596 // Be sure to signal that there's some activity so the CPU doesn't 597 // deschedule itself. 598 activityRec.activity(); 599 fetch.wakeFromQuiesce(); 600 601 _status = Running; 602} 603 604template <class Impl> 605void 606FullO3CPU<Impl>::suspendContext(int tid) 607{ 608 DPRINTF(O3CPU,"[tid: %i]: Suspended ...\n", tid); 609 unscheduleTickEvent(); 610 _status = Idle; 611/* 612 //Remove From Active List, if Active 613 list<unsigned>::iterator isActive = find( 614 activeThreads.begin(), activeThreads.end(), tid); 615 616 if (isActive != activeThreads.end()) { 617 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 618 tid); 619 activeThreads.erase(isActive); 620 } 621*/ 622} 623 624template <class Impl> 625void 626FullO3CPU<Impl>::deallocateContext(int tid) 627{ 628 DPRINTF(O3CPU,"[tid:%i]: Deallocating ...", tid); 629/* 630 //Remove From Active List, if Active 631 list<unsigned>::iterator isActive = find( 632 activeThreads.begin(), activeThreads.end(), tid); 633 634 if (isActive != activeThreads.end()) { 635 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 636 tid); 637 activeThreads.erase(isActive); 638 639 removeThread(tid); 640 } 641*/ 642} 643 644template <class Impl> 645void 646FullO3CPU<Impl>::haltContext(int tid) 647{ 648 DPRINTF(O3CPU,"[tid:%i]: Halted ...", tid); 649/* 650 //Remove From Active List, if Active 651 list<unsigned>::iterator isActive = find( 652 activeThreads.begin(), activeThreads.end(), tid); 653 654 if (isActive != activeThreads.end()) { 655 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 656 tid); 657 activeThreads.erase(isActive); 658 659 removeThread(tid); 660 } 661*/ 662} 663 664template <class Impl> 665void 666FullO3CPU<Impl>::switchOut(Sampler *_sampler) 667{ 668 sampler = _sampler; 669 switchCount = 0; 670 fetch.switchOut(); 671 decode.switchOut(); 672 rename.switchOut(); 673 iew.switchOut(); 674 commit.switchOut(); 675 676 // Wake the CPU and record activity so everything can drain out if 677 // the CPU is currently idle. 678 wakeCPU(); 679 activityRec.activity(); 680} 681 682template <class Impl> 683void 684FullO3CPU<Impl>::signalSwitched() 685{ 686 if (++switchCount == NumStages) { 687 fetch.doSwitchOut(); 688 rename.doSwitchOut(); 689 commit.doSwitchOut(); 690 instList.clear(); 691 while (!removeList.empty()) { 692 removeList.pop(); 693 } 694 695#if USE_CHECKER 696 if (checker) 697 checker->switchOut(sampler); 698#endif 699 700 if (tickEvent.scheduled()) 701 tickEvent.squash(); 702 sampler->signalSwitched(); 703 _status = SwitchedOut; 704 } 705 assert(switchCount <= 5); 706} 707 708template <class Impl> 709void 710FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 711{ 712 // Flush out any old data from the time buffers. 713 for (int i = 0; i < 10; ++i) { 714 timeBuffer.advance(); 715 fetchQueue.advance(); 716 decodeQueue.advance(); 717 renameQueue.advance(); 718 iewQueue.advance(); 719 } 720 721 activityRec.reset(); 722 723 BaseCPU::takeOverFrom(oldCPU); 724 725 fetch.takeOverFrom(); 726 decode.takeOverFrom(); 727 rename.takeOverFrom(); 728 iew.takeOverFrom(); 729 commit.takeOverFrom(); 730 731 assert(!tickEvent.scheduled()); 732 733 // @todo: Figure out how to properly select the tid to put onto 734 // the active threads list. 735 int tid = 0; 736 737 list<unsigned>::iterator isActive = find( 738 activeThreads.begin(), activeThreads.end(), tid); 739 740 if (isActive == activeThreads.end()) { 741 //May Need to Re-code this if the delay variable is the delay 742 //needed for thread to activate 743 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 744 tid); 745 746 activeThreads.push_back(tid); 747 } 748 749 // Set all statuses to active, schedule the CPU's tick event. 750 // @todo: Fix up statuses so this is handled properly 751 for (int i = 0; i < threadContexts.size(); ++i) { 752 ThreadContext *tc = threadContexts[i]; 753 if (tc->status() == ThreadContext::Active && _status != Running) { 754 _status = Running; 755 tickEvent.schedule(curTick); 756 } 757 } 758 if (!tickEvent.scheduled()) 759 tickEvent.schedule(curTick); 760} 761 762template <class Impl> 763uint64_t 764FullO3CPU<Impl>::readIntReg(int reg_idx) 765{ 766 return regFile.readIntReg(reg_idx); 767} 768 769template <class Impl> 770FloatReg 771FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 772{ 773 return regFile.readFloatReg(reg_idx, width); 774} 775 776template <class Impl> 777FloatReg 778FullO3CPU<Impl>::readFloatReg(int reg_idx) 779{ 780 return regFile.readFloatReg(reg_idx); 781} 782 783template <class Impl> 784FloatRegBits 785FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 786{ 787 return regFile.readFloatRegBits(reg_idx, width); 788} 789 790template <class Impl> 791FloatRegBits 792FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 793{ 794 return regFile.readFloatRegBits(reg_idx); 795} 796 797template <class Impl> 798void 799FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 800{ 801 regFile.setIntReg(reg_idx, val); 802} 803 804template <class Impl> 805void 806FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 807{ 808 regFile.setFloatReg(reg_idx, val, width); 809} 810 811template <class Impl> 812void 813FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 814{ 815 regFile.setFloatReg(reg_idx, val); 816} 817 818template <class Impl> 819void 820FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 821{ 822 regFile.setFloatRegBits(reg_idx, val, width); 823} 824 825template <class Impl> 826void 827FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 828{ 829 regFile.setFloatRegBits(reg_idx, val); 830} 831 832template <class Impl> 833uint64_t 834FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 835{ 836 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 837 838 return regFile.readIntReg(phys_reg); 839} 840 841template <class Impl> 842float 843FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 844{ 845 int idx = reg_idx + TheISA::FP_Base_DepTag; 846 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 847 848 return regFile.readFloatReg(phys_reg); 849} 850 851template <class Impl> 852double 853FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 854{ 855 int idx = reg_idx + TheISA::FP_Base_DepTag; 856 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 857 858 return regFile.readFloatReg(phys_reg, 64); 859} 860 861template <class Impl> 862uint64_t 863FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 864{ 865 int idx = reg_idx + TheISA::FP_Base_DepTag; 866 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 867 868 return regFile.readFloatRegBits(phys_reg); 869} 870 871template <class Impl> 872void 873FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 874{ 875 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 876 877 regFile.setIntReg(phys_reg, val); 878} 879 880template <class Impl> 881void 882FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 883{ 884 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 885 886 regFile.setFloatReg(phys_reg, val); 887} 888 889template <class Impl> 890void 891FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 892{ 893 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 894 895 regFile.setFloatReg(phys_reg, val, 64); 896} 897 898template <class Impl> 899void 900FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 901{ 902 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 903 904 regFile.setFloatRegBits(phys_reg, val); 905} 906 907template <class Impl> 908uint64_t 909FullO3CPU<Impl>::readPC(unsigned tid) 910{ 911 return commit.readPC(tid); 912} 913 914template <class Impl> 915void 916FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 917{ 918 commit.setPC(new_PC, tid); 919} 920 921template <class Impl> 922uint64_t 923FullO3CPU<Impl>::readNextPC(unsigned tid) 924{ 925 return commit.readNextPC(tid); 926} 927 928template <class Impl> 929void 930FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 931{ 932 commit.setNextPC(val, tid); 933} 934 935#if THE_ISA != ALPHA_ISA 936template <class Impl> 937uint64_t 938FullO3CPU<Impl>::readNextNPC(unsigned tid) 939{ 940 return commit.readNextNPC(tid); 941} 942 943template <class Impl> 944void 945FullO3CPU<Impl>::setNextNNPC(uint64_t val,unsigned tid) 946{ 947 commit.setNextNPC(val, tid); 948} 949#endif 950 951template <class Impl> 952typename FullO3CPU<Impl>::ListIt 953FullO3CPU<Impl>::addInst(DynInstPtr &inst) 954{ 955 instList.push_back(inst); 956 957 return --(instList.end()); 958} 959 960template <class Impl> 961void 962FullO3CPU<Impl>::instDone(unsigned tid) 963{ 964 // Keep an instruction count. 965 thread[tid]->numInst++; 966 thread[tid]->numInsts++; 967 committedInsts[tid]++; 968 totalCommittedInsts++; 969 970 // Check for instruction-count-based events. 971 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 972} 973 974template <class Impl> 975void 976FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 977{ 978 removeInstsThisCycle = true; 979 980 removeList.push(inst->getInstListIt()); 981} 982 983template <class Impl> 984void 985FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 986{ 987 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 988 "[sn:%lli]\n", 989 inst->threadNumber, inst->readPC(), inst->seqNum); 990 991 removeInstsThisCycle = true; 992 993 // Remove the front instruction. 994 removeList.push(inst->getInstListIt()); 995} 996 997template <class Impl> 998void 999FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 1000{ 1001 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1002 " list.\n", tid); 1003 1004 ListIt end_it; 1005 1006 bool rob_empty = false; 1007 1008 if (instList.empty()) { 1009 return; 1010 } else if (rob.isEmpty(/*tid*/)) { 1011 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1012 end_it = instList.begin(); 1013 rob_empty = true; 1014 } else { 1015 end_it = (rob.readTailInst(tid))->getInstListIt(); 1016 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1017 } 1018 1019 removeInstsThisCycle = true; 1020 1021 ListIt inst_it = instList.end(); 1022 1023 inst_it--; 1024 1025 // Walk through the instruction list, removing any instructions 1026 // that were inserted after the given instruction iterator, end_it. 1027 while (inst_it != end_it) { 1028 assert(!instList.empty()); 1029 1030 squashInstIt(inst_it, tid); 1031 1032 inst_it--; 1033 } 1034 1035 // If the ROB was empty, then we actually need to remove the first 1036 // instruction as well. 1037 if (rob_empty) { 1038 squashInstIt(inst_it, tid); 1039 } 1040} 1041 1042template <class Impl> 1043void 1044FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1045 unsigned tid) 1046{ 1047 assert(!instList.empty()); 1048 1049 removeInstsThisCycle = true; 1050 1051 ListIt inst_iter = instList.end(); 1052 1053 inst_iter--; 1054 1055 DPRINTF(O3CPU, "Deleting instructions from instruction " 1056 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1057 tid, seq_num, (*inst_iter)->seqNum); 1058 1059 while ((*inst_iter)->seqNum > seq_num) { 1060 1061 bool break_loop = (inst_iter == instList.begin()); 1062 1063 squashInstIt(inst_iter, tid); 1064 1065 inst_iter--; 1066 1067 if (break_loop) 1068 break; 1069 } 1070} 1071 1072template <class Impl> 1073inline void 1074FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1075{ 1076 if ((*instIt)->threadNumber == tid) { 1077 DPRINTF(O3CPU, "Squashing instruction, " 1078 "[tid:%i] [sn:%lli] PC %#x\n", 1079 (*instIt)->threadNumber, 1080 (*instIt)->seqNum, 1081 (*instIt)->readPC()); 1082 1083 // Mark it as squashed. 1084 (*instIt)->setSquashed(); 1085 1086 // @todo: Formulate a consistent method for deleting 1087 // instructions from the instruction list 1088 // Remove the instruction from the list. 1089 removeList.push(instIt); 1090 } 1091} 1092 1093template <class Impl> 1094void 1095FullO3CPU<Impl>::cleanUpRemovedInsts() 1096{ 1097 while (!removeList.empty()) { 1098 DPRINTF(O3CPU, "Removing instruction, " 1099 "[tid:%i] [sn:%lli] PC %#x\n", 1100 (*removeList.front())->threadNumber, 1101 (*removeList.front())->seqNum, 1102 (*removeList.front())->readPC()); 1103 1104 instList.erase(removeList.front()); 1105 1106 removeList.pop(); 1107 } 1108 1109 removeInstsThisCycle = false; 1110} 1111/* 1112template <class Impl> 1113void 1114FullO3CPU<Impl>::removeAllInsts() 1115{ 1116 instList.clear(); 1117} 1118*/ 1119template <class Impl> 1120void 1121FullO3CPU<Impl>::dumpInsts() 1122{ 1123 int num = 0; 1124 1125 ListIt inst_list_it = instList.begin(); 1126 1127 cprintf("Dumping Instruction List\n"); 1128 1129 while (inst_list_it != instList.end()) { 1130 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1131 "Squashed:%i\n\n", 1132 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1133 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1134 (*inst_list_it)->isSquashed()); 1135 inst_list_it++; 1136 ++num; 1137 } 1138} 1139/* 1140template <class Impl> 1141void 1142FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1143{ 1144 iew.wakeDependents(inst); 1145} 1146*/ 1147template <class Impl> 1148void 1149FullO3CPU<Impl>::wakeCPU() 1150{ 1151 if (activityRec.active() || tickEvent.scheduled()) { 1152 DPRINTF(Activity, "CPU already running.\n"); 1153 return; 1154 } 1155 1156 DPRINTF(Activity, "Waking up CPU\n"); 1157 1158 idleCycles += (curTick - 1) - lastRunningCycle; 1159 1160 tickEvent.schedule(curTick); 1161} 1162 1163template <class Impl> 1164int 1165FullO3CPU<Impl>::getFreeTid() 1166{ 1167 for (int i=0; i < numThreads; i++) { 1168 if (!tids[i]) { 1169 tids[i] = true; 1170 return i; 1171 } 1172 } 1173 1174 return -1; 1175} 1176 1177template <class Impl> 1178void 1179FullO3CPU<Impl>::doContextSwitch() 1180{ 1181 if (contextSwitch) { 1182 1183 //ADD CODE TO DEACTIVE THREAD HERE (???) 1184 1185 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1186 activateWhenReady(tid); 1187 } 1188 1189 if (cpuWaitList.size() == 0) 1190 contextSwitch = true; 1191 } 1192} 1193 1194template <class Impl> 1195void 1196FullO3CPU<Impl>::updateThreadPriority() 1197{ 1198 if (activeThreads.size() > 1) 1199 { 1200 //DEFAULT TO ROUND ROBIN SCHEME 1201 //e.g. Move highest priority to end of thread list 1202 list<unsigned>::iterator list_begin = activeThreads.begin(); 1203 list<unsigned>::iterator list_end = activeThreads.end(); 1204 1205 unsigned high_thread = *list_begin; 1206 1207 activeThreads.erase(list_begin); 1208 1209 activeThreads.push_back(high_thread); 1210 } 1211} 1212 1213// Forward declaration of FullO3CPU. 1214template class FullO3CPU<AlphaSimpleImpl>;
|