cpu.cc revision 9992:6e39e3641dd8
1/* 2 * Copyright (c) 2011-2012 ARM Limited 3 * Copyright (c) 2013 Advanced Micro Devices, Inc. 4 * All rights reserved 5 * 6 * The license below extends only to copyright in the software and shall 7 * not be construed as granting a license to any other intellectual 8 * property including but not limited to intellectual property relating 9 * to a hardware implementation of the functionality of the software 10 * licensed hereunder. You may use the software subject to the license 11 * terms below provided that you ensure that this notice is replicated 12 * unmodified and in its entirety in all distributions of the software, 13 * modified or unmodified, in source code or in binary form. 14 * 15 * Copyright (c) 2004-2006 The Regents of The University of Michigan 16 * Copyright (c) 2011 Regents of the University of California 17 * All rights reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions are 21 * met: redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer; 23 * redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution; 26 * neither the name of the copyright holders nor the names of its 27 * contributors may be used to endorse or promote products derived from 28 * this software without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Authors: Kevin Lim 43 * Korey Sewell 44 * Rick Strong 45 */ 46 47#include "arch/kernel_stats.hh" 48#include "config/the_isa.hh" 49#include "cpu/checker/cpu.hh" 50#include "cpu/checker/thread_context.hh" 51#include "cpu/o3/cpu.hh" 52#include "cpu/o3/isa_specific.hh" 53#include "cpu/o3/thread_context.hh" 54#include "cpu/activity.hh" 55#include "cpu/quiesce_event.hh" 56#include "cpu/simple_thread.hh" 57#include "cpu/thread_context.hh" 58#include "debug/Activity.hh" 59#include "debug/Drain.hh" 60#include "debug/O3CPU.hh" 61#include "debug/Quiesce.hh" 62#include "enums/MemoryMode.hh" 63#include "sim/core.hh" 64#include "sim/full_system.hh" 65#include "sim/process.hh" 66#include "sim/stat_control.hh" 67#include "sim/system.hh" 68 69#if THE_ISA == ALPHA_ISA 70#include "arch/alpha/osfpal.hh" 71#include "debug/Activity.hh" 72#endif 73 74struct BaseCPUParams; 75 76using namespace TheISA; 77using namespace std; 78 79BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 80 : BaseCPU(params) 81{ 82} 83 84void 85BaseO3CPU::regStats() 86{ 87 BaseCPU::regStats(); 88} 89 90template<class Impl> 91bool 92FullO3CPU<Impl>::IcachePort::recvTimingResp(PacketPtr pkt) 93{ 94 DPRINTF(O3CPU, "Fetch unit received timing\n"); 95 // We shouldn't ever get a block in ownership state 96 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 97 fetch->processCacheCompletion(pkt); 98 99 return true; 100} 101 102template<class Impl> 103void 104FullO3CPU<Impl>::IcachePort::recvRetry() 105{ 106 fetch->recvRetry(); 107} 108 109template <class Impl> 110bool 111FullO3CPU<Impl>::DcachePort::recvTimingResp(PacketPtr pkt) 112{ 113 return lsq->recvTimingResp(pkt); 114} 115 116template <class Impl> 117void 118FullO3CPU<Impl>::DcachePort::recvTimingSnoopReq(PacketPtr pkt) 119{ 120 lsq->recvTimingSnoopReq(pkt); 121} 122 123template <class Impl> 124void 125FullO3CPU<Impl>::DcachePort::recvRetry() 126{ 127 lsq->recvRetry(); 128} 129 130template <class Impl> 131FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 132 : Event(CPU_Tick_Pri), cpu(c) 133{ 134} 135 136template <class Impl> 137void 138FullO3CPU<Impl>::TickEvent::process() 139{ 140 cpu->tick(); 141} 142 143template <class Impl> 144const char * 145FullO3CPU<Impl>::TickEvent::description() const 146{ 147 return "FullO3CPU tick"; 148} 149 150template <class Impl> 151FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 152 : Event(CPU_Switch_Pri) 153{ 154} 155 156template <class Impl> 157void 158FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 159 FullO3CPU<Impl> *thread_cpu) 160{ 161 tid = thread_num; 162 cpu = thread_cpu; 163} 164 165template <class Impl> 166void 167FullO3CPU<Impl>::ActivateThreadEvent::process() 168{ 169 cpu->activateThread(tid); 170} 171 172template <class Impl> 173const char * 174FullO3CPU<Impl>::ActivateThreadEvent::description() const 175{ 176 return "FullO3CPU \"Activate Thread\""; 177} 178 179template <class Impl> 180FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 181 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 182{ 183} 184 185template <class Impl> 186void 187FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 188 FullO3CPU<Impl> *thread_cpu) 189{ 190 tid = thread_num; 191 cpu = thread_cpu; 192 remove = false; 193} 194 195template <class Impl> 196void 197FullO3CPU<Impl>::DeallocateContextEvent::process() 198{ 199 cpu->deactivateThread(tid); 200 if (remove) 201 cpu->removeThread(tid); 202} 203 204template <class Impl> 205const char * 206FullO3CPU<Impl>::DeallocateContextEvent::description() const 207{ 208 return "FullO3CPU \"Deallocate Context\""; 209} 210 211template <class Impl> 212FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 213 : BaseO3CPU(params), 214 itb(params->itb), 215 dtb(params->dtb), 216 tickEvent(this), 217#ifndef NDEBUG 218 instcount(0), 219#endif 220 removeInstsThisCycle(false), 221 fetch(this, params), 222 decode(this, params), 223 rename(this, params), 224 iew(this, params), 225 commit(this, params), 226 227 regFile(params->numPhysIntRegs, 228 params->numPhysFloatRegs, 229 params->numPhysCCRegs), 230 231 freeList(name() + ".freelist", ®File), 232 233 rob(this, params), 234 235 scoreboard(name() + ".scoreboard", 236 regFile.totalNumPhysRegs(), TheISA::NumMiscRegs, 237 TheISA::ZeroReg, TheISA::ZeroReg), 238 239 isa(numThreads, NULL), 240 241 icachePort(&fetch, this), 242 dcachePort(&iew.ldstQueue, this), 243 244 timeBuffer(params->backComSize, params->forwardComSize), 245 fetchQueue(params->backComSize, params->forwardComSize), 246 decodeQueue(params->backComSize, params->forwardComSize), 247 renameQueue(params->backComSize, params->forwardComSize), 248 iewQueue(params->backComSize, params->forwardComSize), 249 activityRec(name(), NumStages, 250 params->backComSize + params->forwardComSize, 251 params->activity), 252 253 globalSeqNum(1), 254 system(params->system), 255 drainManager(NULL), 256 lastRunningCycle(curCycle()) 257{ 258 if (!params->switched_out) { 259 _status = Running; 260 } else { 261 _status = SwitchedOut; 262 } 263 264 if (params->checker) { 265 BaseCPU *temp_checker = params->checker; 266 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 267 checker->setIcachePort(&icachePort); 268 checker->setSystem(params->system); 269 } else { 270 checker = NULL; 271 } 272 273 if (!FullSystem) { 274 thread.resize(numThreads); 275 tids.resize(numThreads); 276 } 277 278 // The stages also need their CPU pointer setup. However this 279 // must be done at the upper level CPU because they have pointers 280 // to the upper level CPU, and not this FullO3CPU. 281 282 // Set up Pointers to the activeThreads list for each stage 283 fetch.setActiveThreads(&activeThreads); 284 decode.setActiveThreads(&activeThreads); 285 rename.setActiveThreads(&activeThreads); 286 iew.setActiveThreads(&activeThreads); 287 commit.setActiveThreads(&activeThreads); 288 289 // Give each of the stages the time buffer they will use. 290 fetch.setTimeBuffer(&timeBuffer); 291 decode.setTimeBuffer(&timeBuffer); 292 rename.setTimeBuffer(&timeBuffer); 293 iew.setTimeBuffer(&timeBuffer); 294 commit.setTimeBuffer(&timeBuffer); 295 296 // Also setup each of the stages' queues. 297 fetch.setFetchQueue(&fetchQueue); 298 decode.setFetchQueue(&fetchQueue); 299 commit.setFetchQueue(&fetchQueue); 300 decode.setDecodeQueue(&decodeQueue); 301 rename.setDecodeQueue(&decodeQueue); 302 rename.setRenameQueue(&renameQueue); 303 iew.setRenameQueue(&renameQueue); 304 iew.setIEWQueue(&iewQueue); 305 commit.setIEWQueue(&iewQueue); 306 commit.setRenameQueue(&renameQueue); 307 308 commit.setIEWStage(&iew); 309 rename.setIEWStage(&iew); 310 rename.setCommitStage(&commit); 311 312 ThreadID active_threads; 313 if (FullSystem) { 314 active_threads = 1; 315 } else { 316 active_threads = params->workload.size(); 317 318 if (active_threads > Impl::MaxThreads) { 319 panic("Workload Size too large. Increase the 'MaxThreads' " 320 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 321 "or edit your workload size."); 322 } 323 } 324 325 //Make Sure That this a Valid Architeture 326 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 327 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 328 assert(params->numPhysCCRegs >= numThreads * TheISA::NumCCRegs); 329 330 rename.setScoreboard(&scoreboard); 331 iew.setScoreboard(&scoreboard); 332 333 // Setup the rename map for whichever stages need it. 334 for (ThreadID tid = 0; tid < numThreads; tid++) { 335 isa[tid] = params->isa[tid]; 336 337 // Only Alpha has an FP zero register, so for other ISAs we 338 // use an invalid FP register index to avoid special treatment 339 // of any valid FP reg. 340 RegIndex invalidFPReg = TheISA::NumFloatRegs + 1; 341 RegIndex fpZeroReg = 342 (THE_ISA == ALPHA_ISA) ? TheISA::ZeroReg : invalidFPReg; 343 344 commitRenameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 345 &freeList); 346 347 renameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 348 &freeList); 349 350 activateThreadEvent[tid].init(tid, this); 351 deallocateContextEvent[tid].init(tid, this); 352 } 353 354 // Initialize rename map to assign physical registers to the 355 // architectural registers for active threads only. 356 for (ThreadID tid = 0; tid < active_threads; tid++) { 357 for (RegIndex ridx = 0; ridx < TheISA::NumIntRegs; ++ridx) { 358 // Note that we can't use the rename() method because we don't 359 // want special treatment for the zero register at this point 360 PhysRegIndex phys_reg = freeList.getIntReg(); 361 renameMap[tid].setIntEntry(ridx, phys_reg); 362 commitRenameMap[tid].setIntEntry(ridx, phys_reg); 363 } 364 365 for (RegIndex ridx = 0; ridx < TheISA::NumFloatRegs; ++ridx) { 366 PhysRegIndex phys_reg = freeList.getFloatReg(); 367 renameMap[tid].setFloatEntry(ridx, phys_reg); 368 commitRenameMap[tid].setFloatEntry(ridx, phys_reg); 369 } 370 371 for (RegIndex ridx = 0; ridx < TheISA::NumCCRegs; ++ridx) { 372 PhysRegIndex phys_reg = freeList.getCCReg(); 373 renameMap[tid].setCCEntry(ridx, phys_reg); 374 commitRenameMap[tid].setCCEntry(ridx, phys_reg); 375 } 376 } 377 378 rename.setRenameMap(renameMap); 379 commit.setRenameMap(commitRenameMap); 380 rename.setFreeList(&freeList); 381 382 // Setup the ROB for whichever stages need it. 383 commit.setROB(&rob); 384 385 lastActivatedCycle = 0; 386#if 0 387 // Give renameMap & rename stage access to the freeList; 388 for (ThreadID tid = 0; tid < numThreads; tid++) 389 globalSeqNum[tid] = 1; 390#endif 391 392 contextSwitch = false; 393 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 394 395 // Setup any thread state. 396 this->thread.resize(this->numThreads); 397 398 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 399 if (FullSystem) { 400 // SMT is not supported in FS mode yet. 401 assert(this->numThreads == 1); 402 this->thread[tid] = new Thread(this, 0, NULL); 403 } else { 404 if (tid < params->workload.size()) { 405 DPRINTF(O3CPU, "Workload[%i] process is %#x", 406 tid, this->thread[tid]); 407 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 408 (typename Impl::O3CPU *)(this), 409 tid, params->workload[tid]); 410 411 //usedTids[tid] = true; 412 //threadMap[tid] = tid; 413 } else { 414 //Allocate Empty thread so M5 can use later 415 //when scheduling threads to CPU 416 Process* dummy_proc = NULL; 417 418 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 419 (typename Impl::O3CPU *)(this), 420 tid, dummy_proc); 421 //usedTids[tid] = false; 422 } 423 } 424 425 ThreadContext *tc; 426 427 // Setup the TC that will serve as the interface to the threads/CPU. 428 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 429 430 tc = o3_tc; 431 432 // If we're using a checker, then the TC should be the 433 // CheckerThreadContext. 434 if (params->checker) { 435 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 436 o3_tc, this->checker); 437 } 438 439 o3_tc->cpu = (typename Impl::O3CPU *)(this); 440 assert(o3_tc->cpu); 441 o3_tc->thread = this->thread[tid]; 442 443 if (FullSystem) { 444 // Setup quiesce event. 445 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 446 } 447 // Give the thread the TC. 448 this->thread[tid]->tc = tc; 449 450 // Add the TC to the CPU's list of TC's. 451 this->threadContexts.push_back(tc); 452 } 453 454 // FullO3CPU always requires an interrupt controller. 455 if (!params->switched_out && !interrupts) { 456 fatal("FullO3CPU %s has no interrupt controller.\n" 457 "Ensure createInterruptController() is called.\n", name()); 458 } 459 460 for (ThreadID tid = 0; tid < this->numThreads; tid++) 461 this->thread[tid]->setFuncExeInst(0); 462} 463 464template <class Impl> 465FullO3CPU<Impl>::~FullO3CPU() 466{ 467} 468 469template <class Impl> 470void 471FullO3CPU<Impl>::regStats() 472{ 473 BaseO3CPU::regStats(); 474 475 // Register any of the O3CPU's stats here. 476 timesIdled 477 .name(name() + ".timesIdled") 478 .desc("Number of times that the entire CPU went into an idle state and" 479 " unscheduled itself") 480 .prereq(timesIdled); 481 482 idleCycles 483 .name(name() + ".idleCycles") 484 .desc("Total number of cycles that the CPU has spent unscheduled due " 485 "to idling") 486 .prereq(idleCycles); 487 488 quiesceCycles 489 .name(name() + ".quiesceCycles") 490 .desc("Total number of cycles that CPU has spent quiesced or waiting " 491 "for an interrupt") 492 .prereq(quiesceCycles); 493 494 // Number of Instructions simulated 495 // -------------------------------- 496 // Should probably be in Base CPU but need templated 497 // MaxThreads so put in here instead 498 committedInsts 499 .init(numThreads) 500 .name(name() + ".committedInsts") 501 .desc("Number of Instructions Simulated"); 502 503 committedOps 504 .init(numThreads) 505 .name(name() + ".committedOps") 506 .desc("Number of Ops (including micro ops) Simulated"); 507 508 totalCommittedInsts 509 .name(name() + ".committedInsts_total") 510 .desc("Number of Instructions Simulated"); 511 512 cpi 513 .name(name() + ".cpi") 514 .desc("CPI: Cycles Per Instruction") 515 .precision(6); 516 cpi = numCycles / committedInsts; 517 518 totalCpi 519 .name(name() + ".cpi_total") 520 .desc("CPI: Total CPI of All Threads") 521 .precision(6); 522 totalCpi = numCycles / totalCommittedInsts; 523 524 ipc 525 .name(name() + ".ipc") 526 .desc("IPC: Instructions Per Cycle") 527 .precision(6); 528 ipc = committedInsts / numCycles; 529 530 totalIpc 531 .name(name() + ".ipc_total") 532 .desc("IPC: Total IPC of All Threads") 533 .precision(6); 534 totalIpc = totalCommittedInsts / numCycles; 535 536 this->fetch.regStats(); 537 this->decode.regStats(); 538 this->rename.regStats(); 539 this->iew.regStats(); 540 this->commit.regStats(); 541 this->rob.regStats(); 542 543 intRegfileReads 544 .name(name() + ".int_regfile_reads") 545 .desc("number of integer regfile reads") 546 .prereq(intRegfileReads); 547 548 intRegfileWrites 549 .name(name() + ".int_regfile_writes") 550 .desc("number of integer regfile writes") 551 .prereq(intRegfileWrites); 552 553 fpRegfileReads 554 .name(name() + ".fp_regfile_reads") 555 .desc("number of floating regfile reads") 556 .prereq(fpRegfileReads); 557 558 fpRegfileWrites 559 .name(name() + ".fp_regfile_writes") 560 .desc("number of floating regfile writes") 561 .prereq(fpRegfileWrites); 562 563 ccRegfileReads 564 .name(name() + ".cc_regfile_reads") 565 .desc("number of cc regfile reads") 566 .prereq(ccRegfileReads); 567 568 ccRegfileWrites 569 .name(name() + ".cc_regfile_writes") 570 .desc("number of cc regfile writes") 571 .prereq(ccRegfileWrites); 572 573 miscRegfileReads 574 .name(name() + ".misc_regfile_reads") 575 .desc("number of misc regfile reads") 576 .prereq(miscRegfileReads); 577 578 miscRegfileWrites 579 .name(name() + ".misc_regfile_writes") 580 .desc("number of misc regfile writes") 581 .prereq(miscRegfileWrites); 582} 583 584template <class Impl> 585void 586FullO3CPU<Impl>::tick() 587{ 588 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 589 assert(!switchedOut()); 590 assert(getDrainState() != Drainable::Drained); 591 592 ++numCycles; 593 594// activity = false; 595 596 //Tick each of the stages 597 fetch.tick(); 598 599 decode.tick(); 600 601 rename.tick(); 602 603 iew.tick(); 604 605 commit.tick(); 606 607 if (!FullSystem) 608 doContextSwitch(); 609 610 // Now advance the time buffers 611 timeBuffer.advance(); 612 613 fetchQueue.advance(); 614 decodeQueue.advance(); 615 renameQueue.advance(); 616 iewQueue.advance(); 617 618 activityRec.advance(); 619 620 if (removeInstsThisCycle) { 621 cleanUpRemovedInsts(); 622 } 623 624 if (!tickEvent.scheduled()) { 625 if (_status == SwitchedOut) { 626 DPRINTF(O3CPU, "Switched out!\n"); 627 // increment stat 628 lastRunningCycle = curCycle(); 629 } else if (!activityRec.active() || _status == Idle) { 630 DPRINTF(O3CPU, "Idle!\n"); 631 lastRunningCycle = curCycle(); 632 timesIdled++; 633 } else { 634 schedule(tickEvent, clockEdge(Cycles(1))); 635 DPRINTF(O3CPU, "Scheduling next tick!\n"); 636 } 637 } 638 639 if (!FullSystem) 640 updateThreadPriority(); 641 642 tryDrain(); 643} 644 645template <class Impl> 646void 647FullO3CPU<Impl>::init() 648{ 649 BaseCPU::init(); 650 651 for (ThreadID tid = 0; tid < numThreads; ++tid) { 652 // Set noSquashFromTC so that the CPU doesn't squash when initially 653 // setting up registers. 654 thread[tid]->noSquashFromTC = true; 655 // Initialise the ThreadContext's memory proxies 656 thread[tid]->initMemProxies(thread[tid]->getTC()); 657 } 658 659 if (FullSystem && !params()->switched_out) { 660 for (ThreadID tid = 0; tid < numThreads; tid++) { 661 ThreadContext *src_tc = threadContexts[tid]; 662 TheISA::initCPU(src_tc, src_tc->contextId()); 663 } 664 } 665 666 // Clear noSquashFromTC. 667 for (int tid = 0; tid < numThreads; ++tid) 668 thread[tid]->noSquashFromTC = false; 669 670 commit.setThreads(thread); 671} 672 673template <class Impl> 674void 675FullO3CPU<Impl>::startup() 676{ 677 BaseCPU::startup(); 678 for (int tid = 0; tid < numThreads; ++tid) 679 isa[tid]->startup(threadContexts[tid]); 680 681 fetch.startupStage(); 682 decode.startupStage(); 683 iew.startupStage(); 684 rename.startupStage(); 685 commit.startupStage(); 686} 687 688template <class Impl> 689void 690FullO3CPU<Impl>::activateThread(ThreadID tid) 691{ 692 list<ThreadID>::iterator isActive = 693 std::find(activeThreads.begin(), activeThreads.end(), tid); 694 695 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 696 assert(!switchedOut()); 697 698 if (isActive == activeThreads.end()) { 699 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 700 tid); 701 702 activeThreads.push_back(tid); 703 } 704} 705 706template <class Impl> 707void 708FullO3CPU<Impl>::deactivateThread(ThreadID tid) 709{ 710 //Remove From Active List, if Active 711 list<ThreadID>::iterator thread_it = 712 std::find(activeThreads.begin(), activeThreads.end(), tid); 713 714 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 715 assert(!switchedOut()); 716 717 if (thread_it != activeThreads.end()) { 718 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 719 tid); 720 activeThreads.erase(thread_it); 721 } 722} 723 724template <class Impl> 725Counter 726FullO3CPU<Impl>::totalInsts() const 727{ 728 Counter total(0); 729 730 ThreadID size = thread.size(); 731 for (ThreadID i = 0; i < size; i++) 732 total += thread[i]->numInst; 733 734 return total; 735} 736 737template <class Impl> 738Counter 739FullO3CPU<Impl>::totalOps() const 740{ 741 Counter total(0); 742 743 ThreadID size = thread.size(); 744 for (ThreadID i = 0; i < size; i++) 745 total += thread[i]->numOp; 746 747 return total; 748} 749 750template <class Impl> 751void 752FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay) 753{ 754 assert(!switchedOut()); 755 756 // Needs to set each stage to running as well. 757 if (delay){ 758 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 759 "on cycle %d\n", tid, clockEdge(delay)); 760 scheduleActivateThreadEvent(tid, delay); 761 } else { 762 activateThread(tid); 763 } 764 765 // We don't want to wake the CPU if it is drained. In that case, 766 // we just want to flag the thread as active and schedule the tick 767 // event from drainResume() instead. 768 if (getDrainState() == Drainable::Drained) 769 return; 770 771 // If we are time 0 or if the last activation time is in the past, 772 // schedule the next tick and wake up the fetch unit 773 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 774 scheduleTickEvent(delay); 775 776 // Be sure to signal that there's some activity so the CPU doesn't 777 // deschedule itself. 778 activityRec.activity(); 779 fetch.wakeFromQuiesce(); 780 781 Cycles cycles(curCycle() - lastRunningCycle); 782 // @todo: This is an oddity that is only here to match the stats 783 if (cycles != 0) 784 --cycles; 785 quiesceCycles += cycles; 786 787 lastActivatedCycle = curTick(); 788 789 _status = Running; 790 } 791} 792 793template <class Impl> 794bool 795FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 796 Cycles delay) 797{ 798 // Schedule removal of thread data from CPU 799 if (delay){ 800 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 801 "on tick %d\n", tid, clockEdge(delay)); 802 scheduleDeallocateContextEvent(tid, remove, delay); 803 return false; 804 } else { 805 deactivateThread(tid); 806 if (remove) 807 removeThread(tid); 808 return true; 809 } 810} 811 812template <class Impl> 813void 814FullO3CPU<Impl>::suspendContext(ThreadID tid) 815{ 816 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 817 assert(!switchedOut()); 818 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1)); 819 // If this was the last thread then unschedule the tick event. 820 if ((activeThreads.size() == 1 && !deallocated) || 821 activeThreads.size() == 0) 822 unscheduleTickEvent(); 823 824 DPRINTF(Quiesce, "Suspending Context\n"); 825 lastRunningCycle = curCycle(); 826 _status = Idle; 827} 828 829template <class Impl> 830void 831FullO3CPU<Impl>::haltContext(ThreadID tid) 832{ 833 //For now, this is the same as deallocate 834 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 835 assert(!switchedOut()); 836 scheduleDeallocateContext(tid, true, Cycles(1)); 837} 838 839template <class Impl> 840void 841FullO3CPU<Impl>::insertThread(ThreadID tid) 842{ 843 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 844 // Will change now that the PC and thread state is internal to the CPU 845 // and not in the ThreadContext. 846 ThreadContext *src_tc; 847 if (FullSystem) 848 src_tc = system->threadContexts[tid]; 849 else 850 src_tc = tcBase(tid); 851 852 //Bind Int Regs to Rename Map 853 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 854 PhysRegIndex phys_reg = freeList.getIntReg(); 855 856 renameMap[tid].setEntry(ireg,phys_reg); 857 scoreboard.setReg(phys_reg); 858 } 859 860 //Bind Float Regs to Rename Map 861 int max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 862 for (int freg = TheISA::NumIntRegs; freg < max_reg; freg++) { 863 PhysRegIndex phys_reg = freeList.getFloatReg(); 864 865 renameMap[tid].setEntry(freg,phys_reg); 866 scoreboard.setReg(phys_reg); 867 } 868 869 //Bind condition-code Regs to Rename Map 870 max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs + TheISA::NumCCRegs; 871 for (int creg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 872 creg < max_reg; creg++) { 873 PhysRegIndex phys_reg = freeList.getCCReg(); 874 875 renameMap[tid].setEntry(creg,phys_reg); 876 scoreboard.setReg(phys_reg); 877 } 878 879 //Copy Thread Data Into RegFile 880 //this->copyFromTC(tid); 881 882 //Set PC/NPC/NNPC 883 pcState(src_tc->pcState(), tid); 884 885 src_tc->setStatus(ThreadContext::Active); 886 887 activateContext(tid, Cycles(1)); 888 889 //Reset ROB/IQ/LSQ Entries 890 commit.rob->resetEntries(); 891 iew.resetEntries(); 892} 893 894template <class Impl> 895void 896FullO3CPU<Impl>::removeThread(ThreadID tid) 897{ 898 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 899 900 // Copy Thread Data From RegFile 901 // If thread is suspended, it might be re-allocated 902 // this->copyToTC(tid); 903 904 905 // @todo: 2-27-2008: Fix how we free up rename mappings 906 // here to alleviate the case for double-freeing registers 907 // in SMT workloads. 908 909 // Unbind Int Regs from Rename Map 910 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 911 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 912 913 scoreboard.unsetReg(phys_reg); 914 freeList.addReg(phys_reg); 915 } 916 917 // Unbind Float Regs from Rename Map 918 int max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 919 for (int freg = TheISA::NumIntRegs; freg < max_reg; freg++) { 920 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 921 922 scoreboard.unsetReg(phys_reg); 923 freeList.addReg(phys_reg); 924 } 925 926 // Unbind condition-code Regs from Rename Map 927 max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs + TheISA::NumCCRegs; 928 for (int creg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 929 creg < max_reg; creg++) { 930 PhysRegIndex phys_reg = renameMap[tid].lookup(creg); 931 932 scoreboard.unsetReg(phys_reg); 933 freeList.addReg(phys_reg); 934 } 935 936 // Squash Throughout Pipeline 937 DynInstPtr inst = commit.rob->readHeadInst(tid); 938 InstSeqNum squash_seq_num = inst->seqNum; 939 fetch.squash(0, squash_seq_num, inst, tid); 940 decode.squash(tid); 941 rename.squash(squash_seq_num, tid); 942 iew.squash(tid); 943 iew.ldstQueue.squash(squash_seq_num, tid); 944 commit.rob->squash(squash_seq_num, tid); 945 946 947 assert(iew.instQueue.getCount(tid) == 0); 948 assert(iew.ldstQueue.getCount(tid) == 0); 949 950 // Reset ROB/IQ/LSQ Entries 951 952 // Commented out for now. This should be possible to do by 953 // telling all the pipeline stages to drain first, and then 954 // checking until the drain completes. Once the pipeline is 955 // drained, call resetEntries(). - 10-09-06 ktlim 956/* 957 if (activeThreads.size() >= 1) { 958 commit.rob->resetEntries(); 959 iew.resetEntries(); 960 } 961*/ 962} 963 964 965template <class Impl> 966void 967FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 968{ 969 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 970 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 971 tid); 972 973 bool ready = true; 974 975 // Should these all be '<' not '>='? This seems backwards... 976 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 977 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 978 "Phys. Int. Regs.\n", 979 tid); 980 ready = false; 981 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 982 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 983 "Phys. Float. Regs.\n", 984 tid); 985 ready = false; 986 } else if (freeList.numFreeCCRegs() >= TheISA::NumCCRegs) { 987 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 988 "Phys. CC. Regs.\n", 989 tid); 990 ready = false; 991 } else if (commit.rob->numFreeEntries() >= 992 commit.rob->entryAmount(activeThreads.size() + 1)) { 993 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 994 "ROB entries.\n", 995 tid); 996 ready = false; 997 } else if (iew.instQueue.numFreeEntries() >= 998 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 999 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 1000 "IQ entries.\n", 1001 tid); 1002 ready = false; 1003 } else if (iew.ldstQueue.numFreeEntries() >= 1004 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 1005 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 1006 "LSQ entries.\n", 1007 tid); 1008 ready = false; 1009 } 1010 1011 if (ready) { 1012 insertThread(tid); 1013 1014 contextSwitch = false; 1015 1016 cpuWaitList.remove(tid); 1017 } else { 1018 suspendContext(tid); 1019 1020 //blocks fetch 1021 contextSwitch = true; 1022 1023 //@todo: dont always add to waitlist 1024 //do waitlist 1025 cpuWaitList.push_back(tid); 1026 } 1027} 1028 1029template <class Impl> 1030Fault 1031FullO3CPU<Impl>::hwrei(ThreadID tid) 1032{ 1033#if THE_ISA == ALPHA_ISA 1034 // Need to clear the lock flag upon returning from an interrupt. 1035 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 1036 1037 this->thread[tid]->kernelStats->hwrei(); 1038 1039 // FIXME: XXX check for interrupts? XXX 1040#endif 1041 return NoFault; 1042} 1043 1044template <class Impl> 1045bool 1046FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1047{ 1048#if THE_ISA == ALPHA_ISA 1049 if (this->thread[tid]->kernelStats) 1050 this->thread[tid]->kernelStats->callpal(palFunc, 1051 this->threadContexts[tid]); 1052 1053 switch (palFunc) { 1054 case PAL::halt: 1055 halt(); 1056 if (--System::numSystemsRunning == 0) 1057 exitSimLoop("all cpus halted"); 1058 break; 1059 1060 case PAL::bpt: 1061 case PAL::bugchk: 1062 if (this->system->breakpoint()) 1063 return false; 1064 break; 1065 } 1066#endif 1067 return true; 1068} 1069 1070template <class Impl> 1071Fault 1072FullO3CPU<Impl>::getInterrupts() 1073{ 1074 // Check if there are any outstanding interrupts 1075 return this->interrupts->getInterrupt(this->threadContexts[0]); 1076} 1077 1078template <class Impl> 1079void 1080FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1081{ 1082 // Check for interrupts here. For now can copy the code that 1083 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1084 // is the one that handles the interrupts. 1085 // @todo: Possibly consolidate the interrupt checking code. 1086 // @todo: Allow other threads to handle interrupts. 1087 1088 assert(interrupt != NoFault); 1089 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1090 1091 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1092 this->trap(interrupt, 0, NULL); 1093} 1094 1095template <class Impl> 1096void 1097FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1098{ 1099 // Pass the thread's TC into the invoke method. 1100 fault->invoke(this->threadContexts[tid], inst); 1101} 1102 1103template <class Impl> 1104void 1105FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1106{ 1107 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1108 1109 DPRINTF(Activity,"Activity: syscall() called.\n"); 1110 1111 // Temporarily increase this by one to account for the syscall 1112 // instruction. 1113 ++(this->thread[tid]->funcExeInst); 1114 1115 // Execute the actual syscall. 1116 this->thread[tid]->syscall(callnum); 1117 1118 // Decrease funcExeInst by one as the normal commit will handle 1119 // incrementing it. 1120 --(this->thread[tid]->funcExeInst); 1121} 1122 1123template <class Impl> 1124void 1125FullO3CPU<Impl>::serializeThread(std::ostream &os, ThreadID tid) 1126{ 1127 thread[tid]->serialize(os); 1128} 1129 1130template <class Impl> 1131void 1132FullO3CPU<Impl>::unserializeThread(Checkpoint *cp, const std::string §ion, 1133 ThreadID tid) 1134{ 1135 thread[tid]->unserialize(cp, section); 1136} 1137 1138template <class Impl> 1139unsigned int 1140FullO3CPU<Impl>::drain(DrainManager *drain_manager) 1141{ 1142 // If the CPU isn't doing anything, then return immediately. 1143 if (switchedOut()) { 1144 setDrainState(Drainable::Drained); 1145 return 0; 1146 } 1147 1148 DPRINTF(Drain, "Draining...\n"); 1149 setDrainState(Drainable::Draining); 1150 1151 // We only need to signal a drain to the commit stage as this 1152 // initiates squashing controls the draining. Once the commit 1153 // stage commits an instruction where it is safe to stop, it'll 1154 // squash the rest of the instructions in the pipeline and force 1155 // the fetch stage to stall. The pipeline will be drained once all 1156 // in-flight instructions have retired. 1157 commit.drain(); 1158 1159 // Wake the CPU and record activity so everything can drain out if 1160 // the CPU was not able to immediately drain. 1161 if (!isDrained()) { 1162 drainManager = drain_manager; 1163 1164 wakeCPU(); 1165 activityRec.activity(); 1166 1167 DPRINTF(Drain, "CPU not drained\n"); 1168 1169 return 1; 1170 } else { 1171 setDrainState(Drainable::Drained); 1172 DPRINTF(Drain, "CPU is already drained\n"); 1173 if (tickEvent.scheduled()) 1174 deschedule(tickEvent); 1175 1176 // Flush out any old data from the time buffers. In 1177 // particular, there might be some data in flight from the 1178 // fetch stage that isn't visible in any of the CPU buffers we 1179 // test in isDrained(). 1180 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1181 timeBuffer.advance(); 1182 fetchQueue.advance(); 1183 decodeQueue.advance(); 1184 renameQueue.advance(); 1185 iewQueue.advance(); 1186 } 1187 1188 drainSanityCheck(); 1189 return 0; 1190 } 1191} 1192 1193template <class Impl> 1194bool 1195FullO3CPU<Impl>::tryDrain() 1196{ 1197 if (!drainManager || !isDrained()) 1198 return false; 1199 1200 if (tickEvent.scheduled()) 1201 deschedule(tickEvent); 1202 1203 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1204 drainManager->signalDrainDone(); 1205 drainManager = NULL; 1206 1207 return true; 1208} 1209 1210template <class Impl> 1211void 1212FullO3CPU<Impl>::drainSanityCheck() const 1213{ 1214 assert(isDrained()); 1215 fetch.drainSanityCheck(); 1216 decode.drainSanityCheck(); 1217 rename.drainSanityCheck(); 1218 iew.drainSanityCheck(); 1219 commit.drainSanityCheck(); 1220} 1221 1222template <class Impl> 1223bool 1224FullO3CPU<Impl>::isDrained() const 1225{ 1226 bool drained(true); 1227 1228 for (ThreadID i = 0; i < thread.size(); ++i) { 1229 if (activateThreadEvent[i].scheduled()) { 1230 DPRINTF(Drain, "CPU not drained, tread %i has a " 1231 "pending activate event\n", i); 1232 drained = false; 1233 } 1234 if (deallocateContextEvent[i].scheduled()) { 1235 DPRINTF(Drain, "CPU not drained, tread %i has a " 1236 "pending deallocate context event\n", i); 1237 drained = false; 1238 } 1239 } 1240 1241 if (!instList.empty() || !removeList.empty()) { 1242 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1243 drained = false; 1244 } 1245 1246 if (!fetch.isDrained()) { 1247 DPRINTF(Drain, "Fetch not drained.\n"); 1248 drained = false; 1249 } 1250 1251 if (!decode.isDrained()) { 1252 DPRINTF(Drain, "Decode not drained.\n"); 1253 drained = false; 1254 } 1255 1256 if (!rename.isDrained()) { 1257 DPRINTF(Drain, "Rename not drained.\n"); 1258 drained = false; 1259 } 1260 1261 if (!iew.isDrained()) { 1262 DPRINTF(Drain, "IEW not drained.\n"); 1263 drained = false; 1264 } 1265 1266 if (!commit.isDrained()) { 1267 DPRINTF(Drain, "Commit not drained.\n"); 1268 drained = false; 1269 } 1270 1271 return drained; 1272} 1273 1274template <class Impl> 1275void 1276FullO3CPU<Impl>::commitDrained(ThreadID tid) 1277{ 1278 fetch.drainStall(tid); 1279} 1280 1281template <class Impl> 1282void 1283FullO3CPU<Impl>::drainResume() 1284{ 1285 setDrainState(Drainable::Running); 1286 if (switchedOut()) 1287 return; 1288 1289 DPRINTF(Drain, "Resuming...\n"); 1290 verifyMemoryMode(); 1291 1292 fetch.drainResume(); 1293 commit.drainResume(); 1294 1295 _status = Idle; 1296 for (ThreadID i = 0; i < thread.size(); i++) { 1297 if (thread[i]->status() == ThreadContext::Active) { 1298 DPRINTF(Drain, "Activating thread: %i\n", i); 1299 activateThread(i); 1300 _status = Running; 1301 } 1302 } 1303 1304 assert(!tickEvent.scheduled()); 1305 if (_status == Running) 1306 schedule(tickEvent, nextCycle()); 1307} 1308 1309template <class Impl> 1310void 1311FullO3CPU<Impl>::switchOut() 1312{ 1313 DPRINTF(O3CPU, "Switching out\n"); 1314 BaseCPU::switchOut(); 1315 1316 activityRec.reset(); 1317 1318 _status = SwitchedOut; 1319 1320 if (checker) 1321 checker->switchOut(); 1322} 1323 1324template <class Impl> 1325void 1326FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1327{ 1328 BaseCPU::takeOverFrom(oldCPU); 1329 1330 fetch.takeOverFrom(); 1331 decode.takeOverFrom(); 1332 rename.takeOverFrom(); 1333 iew.takeOverFrom(); 1334 commit.takeOverFrom(); 1335 1336 assert(!tickEvent.scheduled()); 1337 1338 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1339 if (oldO3CPU) 1340 globalSeqNum = oldO3CPU->globalSeqNum; 1341 1342 lastRunningCycle = curCycle(); 1343 _status = Idle; 1344} 1345 1346template <class Impl> 1347void 1348FullO3CPU<Impl>::verifyMemoryMode() const 1349{ 1350 if (!system->isTimingMode()) { 1351 fatal("The O3 CPU requires the memory system to be in " 1352 "'timing' mode.\n"); 1353 } 1354} 1355 1356template <class Impl> 1357TheISA::MiscReg 1358FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1359{ 1360 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1361} 1362 1363template <class Impl> 1364TheISA::MiscReg 1365FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1366{ 1367 miscRegfileReads++; 1368 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1369} 1370 1371template <class Impl> 1372void 1373FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1374 const TheISA::MiscReg &val, ThreadID tid) 1375{ 1376 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1377} 1378 1379template <class Impl> 1380void 1381FullO3CPU<Impl>::setMiscReg(int misc_reg, 1382 const TheISA::MiscReg &val, ThreadID tid) 1383{ 1384 miscRegfileWrites++; 1385 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1386} 1387 1388template <class Impl> 1389uint64_t 1390FullO3CPU<Impl>::readIntReg(int reg_idx) 1391{ 1392 intRegfileReads++; 1393 return regFile.readIntReg(reg_idx); 1394} 1395 1396template <class Impl> 1397FloatReg 1398FullO3CPU<Impl>::readFloatReg(int reg_idx) 1399{ 1400 fpRegfileReads++; 1401 return regFile.readFloatReg(reg_idx); 1402} 1403 1404template <class Impl> 1405FloatRegBits 1406FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1407{ 1408 fpRegfileReads++; 1409 return regFile.readFloatRegBits(reg_idx); 1410} 1411 1412template <class Impl> 1413CCReg 1414FullO3CPU<Impl>::readCCReg(int reg_idx) 1415{ 1416 ccRegfileReads++; 1417 return regFile.readCCReg(reg_idx); 1418} 1419 1420template <class Impl> 1421void 1422FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1423{ 1424 intRegfileWrites++; 1425 regFile.setIntReg(reg_idx, val); 1426} 1427 1428template <class Impl> 1429void 1430FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1431{ 1432 fpRegfileWrites++; 1433 regFile.setFloatReg(reg_idx, val); 1434} 1435 1436template <class Impl> 1437void 1438FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1439{ 1440 fpRegfileWrites++; 1441 regFile.setFloatRegBits(reg_idx, val); 1442} 1443 1444template <class Impl> 1445void 1446FullO3CPU<Impl>::setCCReg(int reg_idx, CCReg val) 1447{ 1448 ccRegfileWrites++; 1449 regFile.setCCReg(reg_idx, val); 1450} 1451 1452template <class Impl> 1453uint64_t 1454FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1455{ 1456 intRegfileReads++; 1457 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1458 1459 return regFile.readIntReg(phys_reg); 1460} 1461 1462template <class Impl> 1463float 1464FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1465{ 1466 fpRegfileReads++; 1467 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1468 1469 return regFile.readFloatReg(phys_reg); 1470} 1471 1472template <class Impl> 1473uint64_t 1474FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1475{ 1476 fpRegfileReads++; 1477 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1478 1479 return regFile.readFloatRegBits(phys_reg); 1480} 1481 1482template <class Impl> 1483CCReg 1484FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1485{ 1486 ccRegfileReads++; 1487 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1488 1489 return regFile.readCCReg(phys_reg); 1490} 1491 1492template <class Impl> 1493void 1494FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1495{ 1496 intRegfileWrites++; 1497 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1498 1499 regFile.setIntReg(phys_reg, val); 1500} 1501 1502template <class Impl> 1503void 1504FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1505{ 1506 fpRegfileWrites++; 1507 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1508 1509 regFile.setFloatReg(phys_reg, val); 1510} 1511 1512template <class Impl> 1513void 1514FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1515{ 1516 fpRegfileWrites++; 1517 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1518 1519 regFile.setFloatRegBits(phys_reg, val); 1520} 1521 1522template <class Impl> 1523void 1524FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1525{ 1526 ccRegfileWrites++; 1527 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1528 1529 regFile.setCCReg(phys_reg, val); 1530} 1531 1532template <class Impl> 1533TheISA::PCState 1534FullO3CPU<Impl>::pcState(ThreadID tid) 1535{ 1536 return commit.pcState(tid); 1537} 1538 1539template <class Impl> 1540void 1541FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1542{ 1543 commit.pcState(val, tid); 1544} 1545 1546template <class Impl> 1547Addr 1548FullO3CPU<Impl>::instAddr(ThreadID tid) 1549{ 1550 return commit.instAddr(tid); 1551} 1552 1553template <class Impl> 1554Addr 1555FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1556{ 1557 return commit.nextInstAddr(tid); 1558} 1559 1560template <class Impl> 1561MicroPC 1562FullO3CPU<Impl>::microPC(ThreadID tid) 1563{ 1564 return commit.microPC(tid); 1565} 1566 1567template <class Impl> 1568void 1569FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1570{ 1571 this->thread[tid]->noSquashFromTC = true; 1572 this->commit.generateTCEvent(tid); 1573} 1574 1575template <class Impl> 1576typename FullO3CPU<Impl>::ListIt 1577FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1578{ 1579 instList.push_back(inst); 1580 1581 return --(instList.end()); 1582} 1583 1584template <class Impl> 1585void 1586FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1587{ 1588 // Keep an instruction count. 1589 if (!inst->isMicroop() || inst->isLastMicroop()) { 1590 thread[tid]->numInst++; 1591 thread[tid]->numInsts++; 1592 committedInsts[tid]++; 1593 totalCommittedInsts++; 1594 } 1595 thread[tid]->numOp++; 1596 thread[tid]->numOps++; 1597 committedOps[tid]++; 1598 1599 system->totalNumInsts++; 1600 // Check for instruction-count-based events. 1601 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1602 system->instEventQueue.serviceEvents(system->totalNumInsts); 1603} 1604 1605template <class Impl> 1606void 1607FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1608{ 1609 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1610 "[sn:%lli]\n", 1611 inst->threadNumber, inst->pcState(), inst->seqNum); 1612 1613 removeInstsThisCycle = true; 1614 1615 // Remove the front instruction. 1616 removeList.push(inst->getInstListIt()); 1617} 1618 1619template <class Impl> 1620void 1621FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1622{ 1623 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1624 " list.\n", tid); 1625 1626 ListIt end_it; 1627 1628 bool rob_empty = false; 1629 1630 if (instList.empty()) { 1631 return; 1632 } else if (rob.isEmpty(/*tid*/)) { 1633 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1634 end_it = instList.begin(); 1635 rob_empty = true; 1636 } else { 1637 end_it = (rob.readTailInst(tid))->getInstListIt(); 1638 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1639 } 1640 1641 removeInstsThisCycle = true; 1642 1643 ListIt inst_it = instList.end(); 1644 1645 inst_it--; 1646 1647 // Walk through the instruction list, removing any instructions 1648 // that were inserted after the given instruction iterator, end_it. 1649 while (inst_it != end_it) { 1650 assert(!instList.empty()); 1651 1652 squashInstIt(inst_it, tid); 1653 1654 inst_it--; 1655 } 1656 1657 // If the ROB was empty, then we actually need to remove the first 1658 // instruction as well. 1659 if (rob_empty) { 1660 squashInstIt(inst_it, tid); 1661 } 1662} 1663 1664template <class Impl> 1665void 1666FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1667{ 1668 assert(!instList.empty()); 1669 1670 removeInstsThisCycle = true; 1671 1672 ListIt inst_iter = instList.end(); 1673 1674 inst_iter--; 1675 1676 DPRINTF(O3CPU, "Deleting instructions from instruction " 1677 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1678 tid, seq_num, (*inst_iter)->seqNum); 1679 1680 while ((*inst_iter)->seqNum > seq_num) { 1681 1682 bool break_loop = (inst_iter == instList.begin()); 1683 1684 squashInstIt(inst_iter, tid); 1685 1686 inst_iter--; 1687 1688 if (break_loop) 1689 break; 1690 } 1691} 1692 1693template <class Impl> 1694inline void 1695FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1696{ 1697 if ((*instIt)->threadNumber == tid) { 1698 DPRINTF(O3CPU, "Squashing instruction, " 1699 "[tid:%i] [sn:%lli] PC %s\n", 1700 (*instIt)->threadNumber, 1701 (*instIt)->seqNum, 1702 (*instIt)->pcState()); 1703 1704 // Mark it as squashed. 1705 (*instIt)->setSquashed(); 1706 1707 // @todo: Formulate a consistent method for deleting 1708 // instructions from the instruction list 1709 // Remove the instruction from the list. 1710 removeList.push(instIt); 1711 } 1712} 1713 1714template <class Impl> 1715void 1716FullO3CPU<Impl>::cleanUpRemovedInsts() 1717{ 1718 while (!removeList.empty()) { 1719 DPRINTF(O3CPU, "Removing instruction, " 1720 "[tid:%i] [sn:%lli] PC %s\n", 1721 (*removeList.front())->threadNumber, 1722 (*removeList.front())->seqNum, 1723 (*removeList.front())->pcState()); 1724 1725 instList.erase(removeList.front()); 1726 1727 removeList.pop(); 1728 } 1729 1730 removeInstsThisCycle = false; 1731} 1732/* 1733template <class Impl> 1734void 1735FullO3CPU<Impl>::removeAllInsts() 1736{ 1737 instList.clear(); 1738} 1739*/ 1740template <class Impl> 1741void 1742FullO3CPU<Impl>::dumpInsts() 1743{ 1744 int num = 0; 1745 1746 ListIt inst_list_it = instList.begin(); 1747 1748 cprintf("Dumping Instruction List\n"); 1749 1750 while (inst_list_it != instList.end()) { 1751 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1752 "Squashed:%i\n\n", 1753 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1754 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1755 (*inst_list_it)->isSquashed()); 1756 inst_list_it++; 1757 ++num; 1758 } 1759} 1760/* 1761template <class Impl> 1762void 1763FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1764{ 1765 iew.wakeDependents(inst); 1766} 1767*/ 1768template <class Impl> 1769void 1770FullO3CPU<Impl>::wakeCPU() 1771{ 1772 if (activityRec.active() || tickEvent.scheduled()) { 1773 DPRINTF(Activity, "CPU already running.\n"); 1774 return; 1775 } 1776 1777 DPRINTF(Activity, "Waking up CPU\n"); 1778 1779 Cycles cycles(curCycle() - lastRunningCycle); 1780 // @todo: This is an oddity that is only here to match the stats 1781 if (cycles != 0) 1782 --cycles; 1783 idleCycles += cycles; 1784 numCycles += cycles; 1785 1786 schedule(tickEvent, clockEdge()); 1787} 1788 1789template <class Impl> 1790void 1791FullO3CPU<Impl>::wakeup() 1792{ 1793 if (this->thread[0]->status() != ThreadContext::Suspended) 1794 return; 1795 1796 this->wakeCPU(); 1797 1798 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1799 this->threadContexts[0]->activate(); 1800} 1801 1802template <class Impl> 1803ThreadID 1804FullO3CPU<Impl>::getFreeTid() 1805{ 1806 for (ThreadID tid = 0; tid < numThreads; tid++) { 1807 if (!tids[tid]) { 1808 tids[tid] = true; 1809 return tid; 1810 } 1811 } 1812 1813 return InvalidThreadID; 1814} 1815 1816template <class Impl> 1817void 1818FullO3CPU<Impl>::doContextSwitch() 1819{ 1820 if (contextSwitch) { 1821 1822 //ADD CODE TO DEACTIVE THREAD HERE (???) 1823 1824 ThreadID size = cpuWaitList.size(); 1825 for (ThreadID tid = 0; tid < size; tid++) { 1826 activateWhenReady(tid); 1827 } 1828 1829 if (cpuWaitList.size() == 0) 1830 contextSwitch = true; 1831 } 1832} 1833 1834template <class Impl> 1835void 1836FullO3CPU<Impl>::updateThreadPriority() 1837{ 1838 if (activeThreads.size() > 1) { 1839 //DEFAULT TO ROUND ROBIN SCHEME 1840 //e.g. Move highest priority to end of thread list 1841 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1842 1843 unsigned high_thread = *list_begin; 1844 1845 activeThreads.erase(list_begin); 1846 1847 activeThreads.push_back(high_thread); 1848 } 1849} 1850 1851// Forward declaration of FullO3CPU. 1852template class FullO3CPU<O3CPUImpl>; 1853