cpu.cc revision 9954:72a72649a156
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 for (int tid = 0; tid < numThreads; ++tid) 678 isa[tid]->startup(threadContexts[tid]); 679 680 fetch.startupStage(); 681 decode.startupStage(); 682 iew.startupStage(); 683 rename.startupStage(); 684 commit.startupStage(); 685} 686 687template <class Impl> 688void 689FullO3CPU<Impl>::activateThread(ThreadID tid) 690{ 691 list<ThreadID>::iterator isActive = 692 std::find(activeThreads.begin(), activeThreads.end(), tid); 693 694 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 695 assert(!switchedOut()); 696 697 if (isActive == activeThreads.end()) { 698 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 699 tid); 700 701 activeThreads.push_back(tid); 702 } 703} 704 705template <class Impl> 706void 707FullO3CPU<Impl>::deactivateThread(ThreadID tid) 708{ 709 //Remove From Active List, if Active 710 list<ThreadID>::iterator thread_it = 711 std::find(activeThreads.begin(), activeThreads.end(), tid); 712 713 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 714 assert(!switchedOut()); 715 716 if (thread_it != activeThreads.end()) { 717 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 718 tid); 719 activeThreads.erase(thread_it); 720 } 721} 722 723template <class Impl> 724Counter 725FullO3CPU<Impl>::totalInsts() const 726{ 727 Counter total(0); 728 729 ThreadID size = thread.size(); 730 for (ThreadID i = 0; i < size; i++) 731 total += thread[i]->numInst; 732 733 return total; 734} 735 736template <class Impl> 737Counter 738FullO3CPU<Impl>::totalOps() const 739{ 740 Counter total(0); 741 742 ThreadID size = thread.size(); 743 for (ThreadID i = 0; i < size; i++) 744 total += thread[i]->numOp; 745 746 return total; 747} 748 749template <class Impl> 750void 751FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay) 752{ 753 assert(!switchedOut()); 754 755 // Needs to set each stage to running as well. 756 if (delay){ 757 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 758 "on cycle %d\n", tid, clockEdge(delay)); 759 scheduleActivateThreadEvent(tid, delay); 760 } else { 761 activateThread(tid); 762 } 763 764 // We don't want to wake the CPU if it is drained. In that case, 765 // we just want to flag the thread as active and schedule the tick 766 // event from drainResume() instead. 767 if (getDrainState() == Drainable::Drained) 768 return; 769 770 // If we are time 0 or if the last activation time is in the past, 771 // schedule the next tick and wake up the fetch unit 772 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 773 scheduleTickEvent(delay); 774 775 // Be sure to signal that there's some activity so the CPU doesn't 776 // deschedule itself. 777 activityRec.activity(); 778 fetch.wakeFromQuiesce(); 779 780 Cycles cycles(curCycle() - lastRunningCycle); 781 // @todo: This is an oddity that is only here to match the stats 782 if (cycles != 0) 783 --cycles; 784 quiesceCycles += cycles; 785 786 lastActivatedCycle = curTick(); 787 788 _status = Running; 789 } 790} 791 792template <class Impl> 793bool 794FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 795 Cycles delay) 796{ 797 // Schedule removal of thread data from CPU 798 if (delay){ 799 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 800 "on tick %d\n", tid, clockEdge(delay)); 801 scheduleDeallocateContextEvent(tid, remove, delay); 802 return false; 803 } else { 804 deactivateThread(tid); 805 if (remove) 806 removeThread(tid); 807 return true; 808 } 809} 810 811template <class Impl> 812void 813FullO3CPU<Impl>::suspendContext(ThreadID tid) 814{ 815 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 816 assert(!switchedOut()); 817 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1)); 818 // If this was the last thread then unschedule the tick event. 819 if ((activeThreads.size() == 1 && !deallocated) || 820 activeThreads.size() == 0) 821 unscheduleTickEvent(); 822 823 DPRINTF(Quiesce, "Suspending Context\n"); 824 lastRunningCycle = curCycle(); 825 _status = Idle; 826} 827 828template <class Impl> 829void 830FullO3CPU<Impl>::haltContext(ThreadID tid) 831{ 832 //For now, this is the same as deallocate 833 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 834 assert(!switchedOut()); 835 scheduleDeallocateContext(tid, true, Cycles(1)); 836} 837 838template <class Impl> 839void 840FullO3CPU<Impl>::insertThread(ThreadID tid) 841{ 842 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 843 // Will change now that the PC and thread state is internal to the CPU 844 // and not in the ThreadContext. 845 ThreadContext *src_tc; 846 if (FullSystem) 847 src_tc = system->threadContexts[tid]; 848 else 849 src_tc = tcBase(tid); 850 851 //Bind Int Regs to Rename Map 852 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 853 PhysRegIndex phys_reg = freeList.getIntReg(); 854 855 renameMap[tid].setEntry(ireg,phys_reg); 856 scoreboard.setReg(phys_reg); 857 } 858 859 //Bind Float Regs to Rename Map 860 int max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 861 for (int freg = TheISA::NumIntRegs; freg < max_reg; freg++) { 862 PhysRegIndex phys_reg = freeList.getFloatReg(); 863 864 renameMap[tid].setEntry(freg,phys_reg); 865 scoreboard.setReg(phys_reg); 866 } 867 868 //Bind condition-code Regs to Rename Map 869 max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs + TheISA::NumCCRegs; 870 for (int creg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 871 creg < max_reg; creg++) { 872 PhysRegIndex phys_reg = freeList.getCCReg(); 873 874 renameMap[tid].setEntry(creg,phys_reg); 875 scoreboard.setReg(phys_reg); 876 } 877 878 //Copy Thread Data Into RegFile 879 //this->copyFromTC(tid); 880 881 //Set PC/NPC/NNPC 882 pcState(src_tc->pcState(), tid); 883 884 src_tc->setStatus(ThreadContext::Active); 885 886 activateContext(tid, Cycles(1)); 887 888 //Reset ROB/IQ/LSQ Entries 889 commit.rob->resetEntries(); 890 iew.resetEntries(); 891} 892 893template <class Impl> 894void 895FullO3CPU<Impl>::removeThread(ThreadID tid) 896{ 897 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 898 899 // Copy Thread Data From RegFile 900 // If thread is suspended, it might be re-allocated 901 // this->copyToTC(tid); 902 903 904 // @todo: 2-27-2008: Fix how we free up rename mappings 905 // here to alleviate the case for double-freeing registers 906 // in SMT workloads. 907 908 // Unbind Int Regs from Rename Map 909 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 910 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 911 912 scoreboard.unsetReg(phys_reg); 913 freeList.addReg(phys_reg); 914 } 915 916 // Unbind Float Regs from Rename Map 917 int max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 918 for (int freg = TheISA::NumIntRegs; freg < max_reg; freg++) { 919 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 920 921 scoreboard.unsetReg(phys_reg); 922 freeList.addReg(phys_reg); 923 } 924 925 // Unbind condition-code Regs from Rename Map 926 max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs + TheISA::NumCCRegs; 927 for (int creg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 928 creg < max_reg; creg++) { 929 PhysRegIndex phys_reg = renameMap[tid].lookup(creg); 930 931 scoreboard.unsetReg(phys_reg); 932 freeList.addReg(phys_reg); 933 } 934 935 // Squash Throughout Pipeline 936 DynInstPtr inst = commit.rob->readHeadInst(tid); 937 InstSeqNum squash_seq_num = inst->seqNum; 938 fetch.squash(0, squash_seq_num, inst, tid); 939 decode.squash(tid); 940 rename.squash(squash_seq_num, tid); 941 iew.squash(tid); 942 iew.ldstQueue.squash(squash_seq_num, tid); 943 commit.rob->squash(squash_seq_num, tid); 944 945 946 assert(iew.instQueue.getCount(tid) == 0); 947 assert(iew.ldstQueue.getCount(tid) == 0); 948 949 // Reset ROB/IQ/LSQ Entries 950 951 // Commented out for now. This should be possible to do by 952 // telling all the pipeline stages to drain first, and then 953 // checking until the drain completes. Once the pipeline is 954 // drained, call resetEntries(). - 10-09-06 ktlim 955/* 956 if (activeThreads.size() >= 1) { 957 commit.rob->resetEntries(); 958 iew.resetEntries(); 959 } 960*/ 961} 962 963 964template <class Impl> 965void 966FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 967{ 968 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 969 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 970 tid); 971 972 bool ready = true; 973 974 // Should these all be '<' not '>='? This seems backwards... 975 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 976 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 977 "Phys. Int. Regs.\n", 978 tid); 979 ready = false; 980 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 981 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 982 "Phys. Float. Regs.\n", 983 tid); 984 ready = false; 985 } else if (freeList.numFreeCCRegs() >= TheISA::NumCCRegs) { 986 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 987 "Phys. CC. Regs.\n", 988 tid); 989 ready = false; 990 } else if (commit.rob->numFreeEntries() >= 991 commit.rob->entryAmount(activeThreads.size() + 1)) { 992 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 993 "ROB entries.\n", 994 tid); 995 ready = false; 996 } else if (iew.instQueue.numFreeEntries() >= 997 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 998 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 999 "IQ entries.\n", 1000 tid); 1001 ready = false; 1002 } else if (iew.ldstQueue.numFreeEntries() >= 1003 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 1004 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 1005 "LSQ entries.\n", 1006 tid); 1007 ready = false; 1008 } 1009 1010 if (ready) { 1011 insertThread(tid); 1012 1013 contextSwitch = false; 1014 1015 cpuWaitList.remove(tid); 1016 } else { 1017 suspendContext(tid); 1018 1019 //blocks fetch 1020 contextSwitch = true; 1021 1022 //@todo: dont always add to waitlist 1023 //do waitlist 1024 cpuWaitList.push_back(tid); 1025 } 1026} 1027 1028template <class Impl> 1029Fault 1030FullO3CPU<Impl>::hwrei(ThreadID tid) 1031{ 1032#if THE_ISA == ALPHA_ISA 1033 // Need to clear the lock flag upon returning from an interrupt. 1034 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 1035 1036 this->thread[tid]->kernelStats->hwrei(); 1037 1038 // FIXME: XXX check for interrupts? XXX 1039#endif 1040 return NoFault; 1041} 1042 1043template <class Impl> 1044bool 1045FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1046{ 1047#if THE_ISA == ALPHA_ISA 1048 if (this->thread[tid]->kernelStats) 1049 this->thread[tid]->kernelStats->callpal(palFunc, 1050 this->threadContexts[tid]); 1051 1052 switch (palFunc) { 1053 case PAL::halt: 1054 halt(); 1055 if (--System::numSystemsRunning == 0) 1056 exitSimLoop("all cpus halted"); 1057 break; 1058 1059 case PAL::bpt: 1060 case PAL::bugchk: 1061 if (this->system->breakpoint()) 1062 return false; 1063 break; 1064 } 1065#endif 1066 return true; 1067} 1068 1069template <class Impl> 1070Fault 1071FullO3CPU<Impl>::getInterrupts() 1072{ 1073 // Check if there are any outstanding interrupts 1074 return this->interrupts->getInterrupt(this->threadContexts[0]); 1075} 1076 1077template <class Impl> 1078void 1079FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1080{ 1081 // Check for interrupts here. For now can copy the code that 1082 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1083 // is the one that handles the interrupts. 1084 // @todo: Possibly consolidate the interrupt checking code. 1085 // @todo: Allow other threads to handle interrupts. 1086 1087 assert(interrupt != NoFault); 1088 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1089 1090 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1091 this->trap(interrupt, 0, NULL); 1092} 1093 1094template <class Impl> 1095void 1096FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1097{ 1098 // Pass the thread's TC into the invoke method. 1099 fault->invoke(this->threadContexts[tid], inst); 1100} 1101 1102template <class Impl> 1103void 1104FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1105{ 1106 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1107 1108 DPRINTF(Activity,"Activity: syscall() called.\n"); 1109 1110 // Temporarily increase this by one to account for the syscall 1111 // instruction. 1112 ++(this->thread[tid]->funcExeInst); 1113 1114 // Execute the actual syscall. 1115 this->thread[tid]->syscall(callnum); 1116 1117 // Decrease funcExeInst by one as the normal commit will handle 1118 // incrementing it. 1119 --(this->thread[tid]->funcExeInst); 1120} 1121 1122template <class Impl> 1123void 1124FullO3CPU<Impl>::serializeThread(std::ostream &os, ThreadID tid) 1125{ 1126 thread[tid]->serialize(os); 1127} 1128 1129template <class Impl> 1130void 1131FullO3CPU<Impl>::unserializeThread(Checkpoint *cp, const std::string §ion, 1132 ThreadID tid) 1133{ 1134 thread[tid]->unserialize(cp, section); 1135} 1136 1137template <class Impl> 1138unsigned int 1139FullO3CPU<Impl>::drain(DrainManager *drain_manager) 1140{ 1141 // If the CPU isn't doing anything, then return immediately. 1142 if (switchedOut()) { 1143 setDrainState(Drainable::Drained); 1144 return 0; 1145 } 1146 1147 DPRINTF(Drain, "Draining...\n"); 1148 setDrainState(Drainable::Draining); 1149 1150 // We only need to signal a drain to the commit stage as this 1151 // initiates squashing controls the draining. Once the commit 1152 // stage commits an instruction where it is safe to stop, it'll 1153 // squash the rest of the instructions in the pipeline and force 1154 // the fetch stage to stall. The pipeline will be drained once all 1155 // in-flight instructions have retired. 1156 commit.drain(); 1157 1158 // Wake the CPU and record activity so everything can drain out if 1159 // the CPU was not able to immediately drain. 1160 if (!isDrained()) { 1161 drainManager = drain_manager; 1162 1163 wakeCPU(); 1164 activityRec.activity(); 1165 1166 DPRINTF(Drain, "CPU not drained\n"); 1167 1168 return 1; 1169 } else { 1170 setDrainState(Drainable::Drained); 1171 DPRINTF(Drain, "CPU is already drained\n"); 1172 if (tickEvent.scheduled()) 1173 deschedule(tickEvent); 1174 1175 // Flush out any old data from the time buffers. In 1176 // particular, there might be some data in flight from the 1177 // fetch stage that isn't visible in any of the CPU buffers we 1178 // test in isDrained(). 1179 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1180 timeBuffer.advance(); 1181 fetchQueue.advance(); 1182 decodeQueue.advance(); 1183 renameQueue.advance(); 1184 iewQueue.advance(); 1185 } 1186 1187 drainSanityCheck(); 1188 return 0; 1189 } 1190} 1191 1192template <class Impl> 1193bool 1194FullO3CPU<Impl>::tryDrain() 1195{ 1196 if (!drainManager || !isDrained()) 1197 return false; 1198 1199 if (tickEvent.scheduled()) 1200 deschedule(tickEvent); 1201 1202 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1203 drainManager->signalDrainDone(); 1204 drainManager = NULL; 1205 1206 return true; 1207} 1208 1209template <class Impl> 1210void 1211FullO3CPU<Impl>::drainSanityCheck() const 1212{ 1213 assert(isDrained()); 1214 fetch.drainSanityCheck(); 1215 decode.drainSanityCheck(); 1216 rename.drainSanityCheck(); 1217 iew.drainSanityCheck(); 1218 commit.drainSanityCheck(); 1219} 1220 1221template <class Impl> 1222bool 1223FullO3CPU<Impl>::isDrained() const 1224{ 1225 bool drained(true); 1226 1227 for (ThreadID i = 0; i < thread.size(); ++i) { 1228 if (activateThreadEvent[i].scheduled()) { 1229 DPRINTF(Drain, "CPU not drained, tread %i has a " 1230 "pending activate event\n", i); 1231 drained = false; 1232 } 1233 if (deallocateContextEvent[i].scheduled()) { 1234 DPRINTF(Drain, "CPU not drained, tread %i has a " 1235 "pending deallocate context event\n", i); 1236 drained = false; 1237 } 1238 } 1239 1240 if (!instList.empty() || !removeList.empty()) { 1241 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1242 drained = false; 1243 } 1244 1245 if (!fetch.isDrained()) { 1246 DPRINTF(Drain, "Fetch not drained.\n"); 1247 drained = false; 1248 } 1249 1250 if (!decode.isDrained()) { 1251 DPRINTF(Drain, "Decode not drained.\n"); 1252 drained = false; 1253 } 1254 1255 if (!rename.isDrained()) { 1256 DPRINTF(Drain, "Rename not drained.\n"); 1257 drained = false; 1258 } 1259 1260 if (!iew.isDrained()) { 1261 DPRINTF(Drain, "IEW not drained.\n"); 1262 drained = false; 1263 } 1264 1265 if (!commit.isDrained()) { 1266 DPRINTF(Drain, "Commit not drained.\n"); 1267 drained = false; 1268 } 1269 1270 return drained; 1271} 1272 1273template <class Impl> 1274void 1275FullO3CPU<Impl>::commitDrained(ThreadID tid) 1276{ 1277 fetch.drainStall(tid); 1278} 1279 1280template <class Impl> 1281void 1282FullO3CPU<Impl>::drainResume() 1283{ 1284 setDrainState(Drainable::Running); 1285 if (switchedOut()) 1286 return; 1287 1288 DPRINTF(Drain, "Resuming...\n"); 1289 verifyMemoryMode(); 1290 1291 fetch.drainResume(); 1292 commit.drainResume(); 1293 1294 _status = Idle; 1295 for (ThreadID i = 0; i < thread.size(); i++) { 1296 if (thread[i]->status() == ThreadContext::Active) { 1297 DPRINTF(Drain, "Activating thread: %i\n", i); 1298 activateThread(i); 1299 _status = Running; 1300 } 1301 } 1302 1303 assert(!tickEvent.scheduled()); 1304 if (_status == Running) 1305 schedule(tickEvent, nextCycle()); 1306} 1307 1308template <class Impl> 1309void 1310FullO3CPU<Impl>::switchOut() 1311{ 1312 DPRINTF(O3CPU, "Switching out\n"); 1313 BaseCPU::switchOut(); 1314 1315 activityRec.reset(); 1316 1317 _status = SwitchedOut; 1318 1319 if (checker) 1320 checker->switchOut(); 1321} 1322 1323template <class Impl> 1324void 1325FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1326{ 1327 BaseCPU::takeOverFrom(oldCPU); 1328 1329 fetch.takeOverFrom(); 1330 decode.takeOverFrom(); 1331 rename.takeOverFrom(); 1332 iew.takeOverFrom(); 1333 commit.takeOverFrom(); 1334 1335 assert(!tickEvent.scheduled()); 1336 1337 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1338 if (oldO3CPU) 1339 globalSeqNum = oldO3CPU->globalSeqNum; 1340 1341 lastRunningCycle = curCycle(); 1342 _status = Idle; 1343} 1344 1345template <class Impl> 1346void 1347FullO3CPU<Impl>::verifyMemoryMode() const 1348{ 1349 if (!system->isTimingMode()) { 1350 fatal("The O3 CPU requires the memory system to be in " 1351 "'timing' mode.\n"); 1352 } 1353} 1354 1355template <class Impl> 1356TheISA::MiscReg 1357FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1358{ 1359 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1360} 1361 1362template <class Impl> 1363TheISA::MiscReg 1364FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1365{ 1366 miscRegfileReads++; 1367 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1368} 1369 1370template <class Impl> 1371void 1372FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1373 const TheISA::MiscReg &val, ThreadID tid) 1374{ 1375 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1376} 1377 1378template <class Impl> 1379void 1380FullO3CPU<Impl>::setMiscReg(int misc_reg, 1381 const TheISA::MiscReg &val, ThreadID tid) 1382{ 1383 miscRegfileWrites++; 1384 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1385} 1386 1387template <class Impl> 1388uint64_t 1389FullO3CPU<Impl>::readIntReg(int reg_idx) 1390{ 1391 intRegfileReads++; 1392 return regFile.readIntReg(reg_idx); 1393} 1394 1395template <class Impl> 1396FloatReg 1397FullO3CPU<Impl>::readFloatReg(int reg_idx) 1398{ 1399 fpRegfileReads++; 1400 return regFile.readFloatReg(reg_idx); 1401} 1402 1403template <class Impl> 1404FloatRegBits 1405FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1406{ 1407 fpRegfileReads++; 1408 return regFile.readFloatRegBits(reg_idx); 1409} 1410 1411template <class Impl> 1412CCReg 1413FullO3CPU<Impl>::readCCReg(int reg_idx) 1414{ 1415 ccRegfileReads++; 1416 return regFile.readCCReg(reg_idx); 1417} 1418 1419template <class Impl> 1420void 1421FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1422{ 1423 intRegfileWrites++; 1424 regFile.setIntReg(reg_idx, val); 1425} 1426 1427template <class Impl> 1428void 1429FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1430{ 1431 fpRegfileWrites++; 1432 regFile.setFloatReg(reg_idx, val); 1433} 1434 1435template <class Impl> 1436void 1437FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1438{ 1439 fpRegfileWrites++; 1440 regFile.setFloatRegBits(reg_idx, val); 1441} 1442 1443template <class Impl> 1444void 1445FullO3CPU<Impl>::setCCReg(int reg_idx, CCReg val) 1446{ 1447 ccRegfileWrites++; 1448 regFile.setCCReg(reg_idx, val); 1449} 1450 1451template <class Impl> 1452uint64_t 1453FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1454{ 1455 intRegfileReads++; 1456 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1457 1458 return regFile.readIntReg(phys_reg); 1459} 1460 1461template <class Impl> 1462float 1463FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1464{ 1465 fpRegfileReads++; 1466 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1467 1468 return regFile.readFloatReg(phys_reg); 1469} 1470 1471template <class Impl> 1472uint64_t 1473FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1474{ 1475 fpRegfileReads++; 1476 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1477 1478 return regFile.readFloatRegBits(phys_reg); 1479} 1480 1481template <class Impl> 1482CCReg 1483FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1484{ 1485 ccRegfileReads++; 1486 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1487 1488 return regFile.readCCReg(phys_reg); 1489} 1490 1491template <class Impl> 1492void 1493FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1494{ 1495 intRegfileWrites++; 1496 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1497 1498 regFile.setIntReg(phys_reg, val); 1499} 1500 1501template <class Impl> 1502void 1503FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1504{ 1505 fpRegfileWrites++; 1506 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1507 1508 regFile.setFloatReg(phys_reg, val); 1509} 1510 1511template <class Impl> 1512void 1513FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1514{ 1515 fpRegfileWrites++; 1516 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1517 1518 regFile.setFloatRegBits(phys_reg, val); 1519} 1520 1521template <class Impl> 1522void 1523FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1524{ 1525 ccRegfileWrites++; 1526 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1527 1528 regFile.setCCReg(phys_reg, val); 1529} 1530 1531template <class Impl> 1532TheISA::PCState 1533FullO3CPU<Impl>::pcState(ThreadID tid) 1534{ 1535 return commit.pcState(tid); 1536} 1537 1538template <class Impl> 1539void 1540FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1541{ 1542 commit.pcState(val, tid); 1543} 1544 1545template <class Impl> 1546Addr 1547FullO3CPU<Impl>::instAddr(ThreadID tid) 1548{ 1549 return commit.instAddr(tid); 1550} 1551 1552template <class Impl> 1553Addr 1554FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1555{ 1556 return commit.nextInstAddr(tid); 1557} 1558 1559template <class Impl> 1560MicroPC 1561FullO3CPU<Impl>::microPC(ThreadID tid) 1562{ 1563 return commit.microPC(tid); 1564} 1565 1566template <class Impl> 1567void 1568FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1569{ 1570 this->thread[tid]->noSquashFromTC = true; 1571 this->commit.generateTCEvent(tid); 1572} 1573 1574template <class Impl> 1575typename FullO3CPU<Impl>::ListIt 1576FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1577{ 1578 instList.push_back(inst); 1579 1580 return --(instList.end()); 1581} 1582 1583template <class Impl> 1584void 1585FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1586{ 1587 // Keep an instruction count. 1588 if (!inst->isMicroop() || inst->isLastMicroop()) { 1589 thread[tid]->numInst++; 1590 thread[tid]->numInsts++; 1591 committedInsts[tid]++; 1592 totalCommittedInsts++; 1593 } 1594 thread[tid]->numOp++; 1595 thread[tid]->numOps++; 1596 committedOps[tid]++; 1597 1598 system->totalNumInsts++; 1599 // Check for instruction-count-based events. 1600 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1601 system->instEventQueue.serviceEvents(system->totalNumInsts); 1602} 1603 1604template <class Impl> 1605void 1606FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1607{ 1608 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1609 "[sn:%lli]\n", 1610 inst->threadNumber, inst->pcState(), inst->seqNum); 1611 1612 removeInstsThisCycle = true; 1613 1614 // Remove the front instruction. 1615 removeList.push(inst->getInstListIt()); 1616} 1617 1618template <class Impl> 1619void 1620FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1621{ 1622 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1623 " list.\n", tid); 1624 1625 ListIt end_it; 1626 1627 bool rob_empty = false; 1628 1629 if (instList.empty()) { 1630 return; 1631 } else if (rob.isEmpty(/*tid*/)) { 1632 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1633 end_it = instList.begin(); 1634 rob_empty = true; 1635 } else { 1636 end_it = (rob.readTailInst(tid))->getInstListIt(); 1637 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1638 } 1639 1640 removeInstsThisCycle = true; 1641 1642 ListIt inst_it = instList.end(); 1643 1644 inst_it--; 1645 1646 // Walk through the instruction list, removing any instructions 1647 // that were inserted after the given instruction iterator, end_it. 1648 while (inst_it != end_it) { 1649 assert(!instList.empty()); 1650 1651 squashInstIt(inst_it, tid); 1652 1653 inst_it--; 1654 } 1655 1656 // If the ROB was empty, then we actually need to remove the first 1657 // instruction as well. 1658 if (rob_empty) { 1659 squashInstIt(inst_it, tid); 1660 } 1661} 1662 1663template <class Impl> 1664void 1665FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1666{ 1667 assert(!instList.empty()); 1668 1669 removeInstsThisCycle = true; 1670 1671 ListIt inst_iter = instList.end(); 1672 1673 inst_iter--; 1674 1675 DPRINTF(O3CPU, "Deleting instructions from instruction " 1676 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1677 tid, seq_num, (*inst_iter)->seqNum); 1678 1679 while ((*inst_iter)->seqNum > seq_num) { 1680 1681 bool break_loop = (inst_iter == instList.begin()); 1682 1683 squashInstIt(inst_iter, tid); 1684 1685 inst_iter--; 1686 1687 if (break_loop) 1688 break; 1689 } 1690} 1691 1692template <class Impl> 1693inline void 1694FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1695{ 1696 if ((*instIt)->threadNumber == tid) { 1697 DPRINTF(O3CPU, "Squashing instruction, " 1698 "[tid:%i] [sn:%lli] PC %s\n", 1699 (*instIt)->threadNumber, 1700 (*instIt)->seqNum, 1701 (*instIt)->pcState()); 1702 1703 // Mark it as squashed. 1704 (*instIt)->setSquashed(); 1705 1706 // @todo: Formulate a consistent method for deleting 1707 // instructions from the instruction list 1708 // Remove the instruction from the list. 1709 removeList.push(instIt); 1710 } 1711} 1712 1713template <class Impl> 1714void 1715FullO3CPU<Impl>::cleanUpRemovedInsts() 1716{ 1717 while (!removeList.empty()) { 1718 DPRINTF(O3CPU, "Removing instruction, " 1719 "[tid:%i] [sn:%lli] PC %s\n", 1720 (*removeList.front())->threadNumber, 1721 (*removeList.front())->seqNum, 1722 (*removeList.front())->pcState()); 1723 1724 instList.erase(removeList.front()); 1725 1726 removeList.pop(); 1727 } 1728 1729 removeInstsThisCycle = false; 1730} 1731/* 1732template <class Impl> 1733void 1734FullO3CPU<Impl>::removeAllInsts() 1735{ 1736 instList.clear(); 1737} 1738*/ 1739template <class Impl> 1740void 1741FullO3CPU<Impl>::dumpInsts() 1742{ 1743 int num = 0; 1744 1745 ListIt inst_list_it = instList.begin(); 1746 1747 cprintf("Dumping Instruction List\n"); 1748 1749 while (inst_list_it != instList.end()) { 1750 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1751 "Squashed:%i\n\n", 1752 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1753 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1754 (*inst_list_it)->isSquashed()); 1755 inst_list_it++; 1756 ++num; 1757 } 1758} 1759/* 1760template <class Impl> 1761void 1762FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1763{ 1764 iew.wakeDependents(inst); 1765} 1766*/ 1767template <class Impl> 1768void 1769FullO3CPU<Impl>::wakeCPU() 1770{ 1771 if (activityRec.active() || tickEvent.scheduled()) { 1772 DPRINTF(Activity, "CPU already running.\n"); 1773 return; 1774 } 1775 1776 DPRINTF(Activity, "Waking up CPU\n"); 1777 1778 Cycles cycles(curCycle() - lastRunningCycle); 1779 // @todo: This is an oddity that is only here to match the stats 1780 if (cycles != 0) 1781 --cycles; 1782 idleCycles += cycles; 1783 numCycles += cycles; 1784 1785 schedule(tickEvent, clockEdge()); 1786} 1787 1788template <class Impl> 1789void 1790FullO3CPU<Impl>::wakeup() 1791{ 1792 if (this->thread[0]->status() != ThreadContext::Suspended) 1793 return; 1794 1795 this->wakeCPU(); 1796 1797 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1798 this->threadContexts[0]->activate(); 1799} 1800 1801template <class Impl> 1802ThreadID 1803FullO3CPU<Impl>::getFreeTid() 1804{ 1805 for (ThreadID tid = 0; tid < numThreads; tid++) { 1806 if (!tids[tid]) { 1807 tids[tid] = true; 1808 return tid; 1809 } 1810 } 1811 1812 return InvalidThreadID; 1813} 1814 1815template <class Impl> 1816void 1817FullO3CPU<Impl>::doContextSwitch() 1818{ 1819 if (contextSwitch) { 1820 1821 //ADD CODE TO DEACTIVE THREAD HERE (???) 1822 1823 ThreadID size = cpuWaitList.size(); 1824 for (ThreadID tid = 0; tid < size; tid++) { 1825 activateWhenReady(tid); 1826 } 1827 1828 if (cpuWaitList.size() == 0) 1829 contextSwitch = true; 1830 } 1831} 1832 1833template <class Impl> 1834void 1835FullO3CPU<Impl>::updateThreadPriority() 1836{ 1837 if (activeThreads.size() > 1) { 1838 //DEFAULT TO ROUND ROBIN SCHEME 1839 //e.g. Move highest priority to end of thread list 1840 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1841 1842 unsigned high_thread = *list_begin; 1843 1844 activeThreads.erase(list_begin); 1845 1846 activeThreads.push_back(high_thread); 1847 } 1848} 1849 1850// Forward declaration of FullO3CPU. 1851template class FullO3CPU<O3CPUImpl>; 1852