cpu.cc revision 9436:4a0223da4924
1/* 2 * Copyright (c) 2011-2012 ARM Limited 3 * All rights reserved 4 * 5 * The license below extends only to copyright in the software and shall 6 * not be construed as granting a license to any other intellectual 7 * property including but not limited to intellectual property relating 8 * to a hardware implementation of the functionality of the software 9 * licensed hereunder. You may use the software subject to the license 10 * terms below provided that you ensure that this notice is replicated 11 * unmodified and in its entirety in all distributions of the software, 12 * modified or unmodified, in source code or in binary form. 13 * 14 * Copyright (c) 2004-2006 The Regents of The University of Michigan 15 * Copyright (c) 2011 Regents of the University of California 16 * All rights reserved. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions are 20 * met: redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer; 22 * redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution; 25 * neither the name of the copyright holders nor the names of its 26 * contributors may be used to endorse or promote products derived from 27 * this software without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40 * 41 * Authors: Kevin Lim 42 * Korey Sewell 43 * Rick Strong 44 */ 45 46#include "arch/kernel_stats.hh" 47#include "config/the_isa.hh" 48#include "cpu/checker/cpu.hh" 49#include "cpu/checker/thread_context.hh" 50#include "cpu/o3/cpu.hh" 51#include "cpu/o3/isa_specific.hh" 52#include "cpu/o3/thread_context.hh" 53#include "cpu/activity.hh" 54#include "cpu/quiesce_event.hh" 55#include "cpu/simple_thread.hh" 56#include "cpu/thread_context.hh" 57#include "debug/Activity.hh" 58#include "debug/Drain.hh" 59#include "debug/O3CPU.hh" 60#include "debug/Quiesce.hh" 61#include "enums/MemoryMode.hh" 62#include "sim/core.hh" 63#include "sim/full_system.hh" 64#include "sim/process.hh" 65#include "sim/stat_control.hh" 66#include "sim/system.hh" 67 68#if THE_ISA == ALPHA_ISA 69#include "arch/alpha/osfpal.hh" 70#include "debug/Activity.hh" 71#endif 72 73struct BaseCPUParams; 74 75using namespace TheISA; 76using namespace std; 77 78BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 79 : BaseCPU(params) 80{ 81} 82 83void 84BaseO3CPU::regStats() 85{ 86 BaseCPU::regStats(); 87} 88 89template<class Impl> 90bool 91FullO3CPU<Impl>::IcachePort::recvTimingResp(PacketPtr pkt) 92{ 93 DPRINTF(O3CPU, "Fetch unit received timing\n"); 94 // We shouldn't ever get a block in ownership state 95 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 96 fetch->processCacheCompletion(pkt); 97 98 return true; 99} 100 101template<class Impl> 102void 103FullO3CPU<Impl>::IcachePort::recvRetry() 104{ 105 fetch->recvRetry(); 106} 107 108template <class Impl> 109bool 110FullO3CPU<Impl>::DcachePort::recvTimingResp(PacketPtr pkt) 111{ 112 return lsq->recvTimingResp(pkt); 113} 114 115template <class Impl> 116void 117FullO3CPU<Impl>::DcachePort::recvTimingSnoopReq(PacketPtr pkt) 118{ 119 lsq->recvTimingSnoopReq(pkt); 120} 121 122template <class Impl> 123void 124FullO3CPU<Impl>::DcachePort::recvRetry() 125{ 126 lsq->recvRetry(); 127} 128 129template <class Impl> 130FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 131 : Event(CPU_Tick_Pri), cpu(c) 132{ 133} 134 135template <class Impl> 136void 137FullO3CPU<Impl>::TickEvent::process() 138{ 139 cpu->tick(); 140} 141 142template <class Impl> 143const char * 144FullO3CPU<Impl>::TickEvent::description() const 145{ 146 return "FullO3CPU tick"; 147} 148 149template <class Impl> 150FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 151 : Event(CPU_Switch_Pri) 152{ 153} 154 155template <class Impl> 156void 157FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 158 FullO3CPU<Impl> *thread_cpu) 159{ 160 tid = thread_num; 161 cpu = thread_cpu; 162} 163 164template <class Impl> 165void 166FullO3CPU<Impl>::ActivateThreadEvent::process() 167{ 168 cpu->activateThread(tid); 169} 170 171template <class Impl> 172const char * 173FullO3CPU<Impl>::ActivateThreadEvent::description() const 174{ 175 return "FullO3CPU \"Activate Thread\""; 176} 177 178template <class Impl> 179FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 180 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 181{ 182} 183 184template <class Impl> 185void 186FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 187 FullO3CPU<Impl> *thread_cpu) 188{ 189 tid = thread_num; 190 cpu = thread_cpu; 191 remove = false; 192} 193 194template <class Impl> 195void 196FullO3CPU<Impl>::DeallocateContextEvent::process() 197{ 198 cpu->deactivateThread(tid); 199 if (remove) 200 cpu->removeThread(tid); 201} 202 203template <class Impl> 204const char * 205FullO3CPU<Impl>::DeallocateContextEvent::description() const 206{ 207 return "FullO3CPU \"Deallocate Context\""; 208} 209 210template <class Impl> 211FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 212 : BaseO3CPU(params), 213 itb(params->itb), 214 dtb(params->dtb), 215 tickEvent(this), 216#ifndef NDEBUG 217 instcount(0), 218#endif 219 removeInstsThisCycle(false), 220 fetch(this, params), 221 decode(this, params), 222 rename(this, params), 223 iew(this, params), 224 commit(this, params), 225 226 regFile(this, params->numPhysIntRegs, 227 params->numPhysFloatRegs), 228 229 freeList(params->numThreads, 230 TheISA::NumIntRegs, params->numPhysIntRegs, 231 TheISA::NumFloatRegs, params->numPhysFloatRegs), 232 233 rob(this, 234 params->numROBEntries, params->squashWidth, 235 params->smtROBPolicy, params->smtROBThreshold, 236 params->numThreads), 237 238 scoreboard(params->numThreads, 239 TheISA::NumIntRegs, params->numPhysIntRegs, 240 TheISA::NumFloatRegs, params->numPhysFloatRegs, 241 TheISA::NumMiscRegs * numThreads, 242 TheISA::ZeroReg), 243 244 isa(numThreads, NULL), 245 246 icachePort(&fetch, this), 247 dcachePort(&iew.ldstQueue, this), 248 249 timeBuffer(params->backComSize, params->forwardComSize), 250 fetchQueue(params->backComSize, params->forwardComSize), 251 decodeQueue(params->backComSize, params->forwardComSize), 252 renameQueue(params->backComSize, params->forwardComSize), 253 iewQueue(params->backComSize, params->forwardComSize), 254 activityRec(name(), NumStages, 255 params->backComSize + params->forwardComSize, 256 params->activity), 257 258 globalSeqNum(1), 259 system(params->system), 260 drainCount(0), 261 lastRunningCycle(curCycle()) 262{ 263 if (!params->switched_out) { 264 _status = Running; 265 } else { 266 _status = SwitchedOut; 267 } 268 269 if (params->checker) { 270 BaseCPU *temp_checker = params->checker; 271 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 272 checker->setIcachePort(&icachePort); 273 checker->setSystem(params->system); 274 } else { 275 checker = NULL; 276 } 277 278 if (!FullSystem) { 279 thread.resize(numThreads); 280 tids.resize(numThreads); 281 } 282 283 // The stages also need their CPU pointer setup. However this 284 // must be done at the upper level CPU because they have pointers 285 // to the upper level CPU, and not this FullO3CPU. 286 287 // Set up Pointers to the activeThreads list for each stage 288 fetch.setActiveThreads(&activeThreads); 289 decode.setActiveThreads(&activeThreads); 290 rename.setActiveThreads(&activeThreads); 291 iew.setActiveThreads(&activeThreads); 292 commit.setActiveThreads(&activeThreads); 293 294 // Give each of the stages the time buffer they will use. 295 fetch.setTimeBuffer(&timeBuffer); 296 decode.setTimeBuffer(&timeBuffer); 297 rename.setTimeBuffer(&timeBuffer); 298 iew.setTimeBuffer(&timeBuffer); 299 commit.setTimeBuffer(&timeBuffer); 300 301 // Also setup each of the stages' queues. 302 fetch.setFetchQueue(&fetchQueue); 303 decode.setFetchQueue(&fetchQueue); 304 commit.setFetchQueue(&fetchQueue); 305 decode.setDecodeQueue(&decodeQueue); 306 rename.setDecodeQueue(&decodeQueue); 307 rename.setRenameQueue(&renameQueue); 308 iew.setRenameQueue(&renameQueue); 309 iew.setIEWQueue(&iewQueue); 310 commit.setIEWQueue(&iewQueue); 311 commit.setRenameQueue(&renameQueue); 312 313 commit.setIEWStage(&iew); 314 rename.setIEWStage(&iew); 315 rename.setCommitStage(&commit); 316 317 ThreadID active_threads; 318 if (FullSystem) { 319 active_threads = 1; 320 } else { 321 active_threads = params->workload.size(); 322 323 if (active_threads > Impl::MaxThreads) { 324 panic("Workload Size too large. Increase the 'MaxThreads' " 325 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 326 "or edit your workload size."); 327 } 328 } 329 330 //Make Sure That this a Valid Architeture 331 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 332 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 333 334 rename.setScoreboard(&scoreboard); 335 iew.setScoreboard(&scoreboard); 336 337 // Setup the rename map for whichever stages need it. 338 PhysRegIndex lreg_idx = 0; 339 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 340 341 for (ThreadID tid = 0; tid < numThreads; tid++) { 342 bool bindRegs = (tid <= active_threads - 1); 343 344 isa[tid] = params->isa[tid]; 345 346 commitRenameMap[tid].init(TheISA::NumIntRegs, 347 params->numPhysIntRegs, 348 lreg_idx, //Index for Logical. Regs 349 350 TheISA::NumFloatRegs, 351 params->numPhysFloatRegs, 352 freg_idx, //Index for Float Regs 353 354 TheISA::NumMiscRegs, 355 356 TheISA::ZeroReg, 357 TheISA::ZeroReg, 358 359 tid, 360 false); 361 362 renameMap[tid].init(TheISA::NumIntRegs, 363 params->numPhysIntRegs, 364 lreg_idx, //Index for Logical. Regs 365 366 TheISA::NumFloatRegs, 367 params->numPhysFloatRegs, 368 freg_idx, //Index for Float Regs 369 370 TheISA::NumMiscRegs, 371 372 TheISA::ZeroReg, 373 TheISA::ZeroReg, 374 375 tid, 376 bindRegs); 377 378 activateThreadEvent[tid].init(tid, this); 379 deallocateContextEvent[tid].init(tid, this); 380 } 381 382 rename.setRenameMap(renameMap); 383 commit.setRenameMap(commitRenameMap); 384 385 // Give renameMap & rename stage access to the freeList; 386 for (ThreadID tid = 0; tid < numThreads; tid++) 387 renameMap[tid].setFreeList(&freeList); 388 rename.setFreeList(&freeList); 389 390 // Setup the ROB for whichever stages need it. 391 commit.setROB(&rob); 392 393 lastActivatedCycle = 0; 394#if 0 395 // Give renameMap & rename stage access to the freeList; 396 for (ThreadID tid = 0; tid < numThreads; tid++) 397 globalSeqNum[tid] = 1; 398#endif 399 400 contextSwitch = false; 401 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 402 403 // Setup any thread state. 404 this->thread.resize(this->numThreads); 405 406 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 407 if (FullSystem) { 408 // SMT is not supported in FS mode yet. 409 assert(this->numThreads == 1); 410 this->thread[tid] = new Thread(this, 0, NULL); 411 } else { 412 if (tid < params->workload.size()) { 413 DPRINTF(O3CPU, "Workload[%i] process is %#x", 414 tid, this->thread[tid]); 415 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 416 (typename Impl::O3CPU *)(this), 417 tid, params->workload[tid]); 418 419 //usedTids[tid] = true; 420 //threadMap[tid] = tid; 421 } else { 422 //Allocate Empty thread so M5 can use later 423 //when scheduling threads to CPU 424 Process* dummy_proc = NULL; 425 426 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 427 (typename Impl::O3CPU *)(this), 428 tid, dummy_proc); 429 //usedTids[tid] = false; 430 } 431 } 432 433 ThreadContext *tc; 434 435 // Setup the TC that will serve as the interface to the threads/CPU. 436 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 437 438 tc = o3_tc; 439 440 // If we're using a checker, then the TC should be the 441 // CheckerThreadContext. 442 if (params->checker) { 443 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 444 o3_tc, this->checker); 445 } 446 447 o3_tc->cpu = (typename Impl::O3CPU *)(this); 448 assert(o3_tc->cpu); 449 o3_tc->thread = this->thread[tid]; 450 451 if (FullSystem) { 452 // Setup quiesce event. 453 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 454 } 455 // Give the thread the TC. 456 this->thread[tid]->tc = tc; 457 458 // Add the TC to the CPU's list of TC's. 459 this->threadContexts.push_back(tc); 460 } 461 462 // FullO3CPU always requires an interrupt controller. 463 if (!params->switched_out && !interrupts) { 464 fatal("FullO3CPU %s has no interrupt controller.\n" 465 "Ensure createInterruptController() is called.\n", name()); 466 } 467 468 for (ThreadID tid = 0; tid < this->numThreads; tid++) 469 this->thread[tid]->setFuncExeInst(0); 470} 471 472template <class Impl> 473FullO3CPU<Impl>::~FullO3CPU() 474{ 475} 476 477template <class Impl> 478void 479FullO3CPU<Impl>::regStats() 480{ 481 BaseO3CPU::regStats(); 482 483 // Register any of the O3CPU's stats here. 484 timesIdled 485 .name(name() + ".timesIdled") 486 .desc("Number of times that the entire CPU went into an idle state and" 487 " unscheduled itself") 488 .prereq(timesIdled); 489 490 idleCycles 491 .name(name() + ".idleCycles") 492 .desc("Total number of cycles that the CPU has spent unscheduled due " 493 "to idling") 494 .prereq(idleCycles); 495 496 quiesceCycles 497 .name(name() + ".quiesceCycles") 498 .desc("Total number of cycles that CPU has spent quiesced or waiting " 499 "for an interrupt") 500 .prereq(quiesceCycles); 501 502 // Number of Instructions simulated 503 // -------------------------------- 504 // Should probably be in Base CPU but need templated 505 // MaxThreads so put in here instead 506 committedInsts 507 .init(numThreads) 508 .name(name() + ".committedInsts") 509 .desc("Number of Instructions Simulated"); 510 511 committedOps 512 .init(numThreads) 513 .name(name() + ".committedOps") 514 .desc("Number of Ops (including micro ops) Simulated"); 515 516 totalCommittedInsts 517 .name(name() + ".committedInsts_total") 518 .desc("Number of Instructions Simulated"); 519 520 cpi 521 .name(name() + ".cpi") 522 .desc("CPI: Cycles Per Instruction") 523 .precision(6); 524 cpi = numCycles / committedInsts; 525 526 totalCpi 527 .name(name() + ".cpi_total") 528 .desc("CPI: Total CPI of All Threads") 529 .precision(6); 530 totalCpi = numCycles / totalCommittedInsts; 531 532 ipc 533 .name(name() + ".ipc") 534 .desc("IPC: Instructions Per Cycle") 535 .precision(6); 536 ipc = committedInsts / numCycles; 537 538 totalIpc 539 .name(name() + ".ipc_total") 540 .desc("IPC: Total IPC of All Threads") 541 .precision(6); 542 totalIpc = totalCommittedInsts / numCycles; 543 544 this->fetch.regStats(); 545 this->decode.regStats(); 546 this->rename.regStats(); 547 this->iew.regStats(); 548 this->commit.regStats(); 549 this->rob.regStats(); 550 551 intRegfileReads 552 .name(name() + ".int_regfile_reads") 553 .desc("number of integer regfile reads") 554 .prereq(intRegfileReads); 555 556 intRegfileWrites 557 .name(name() + ".int_regfile_writes") 558 .desc("number of integer regfile writes") 559 .prereq(intRegfileWrites); 560 561 fpRegfileReads 562 .name(name() + ".fp_regfile_reads") 563 .desc("number of floating regfile reads") 564 .prereq(fpRegfileReads); 565 566 fpRegfileWrites 567 .name(name() + ".fp_regfile_writes") 568 .desc("number of floating regfile writes") 569 .prereq(fpRegfileWrites); 570 571 miscRegfileReads 572 .name(name() + ".misc_regfile_reads") 573 .desc("number of misc regfile reads") 574 .prereq(miscRegfileReads); 575 576 miscRegfileWrites 577 .name(name() + ".misc_regfile_writes") 578 .desc("number of misc regfile writes") 579 .prereq(miscRegfileWrites); 580} 581 582template <class Impl> 583void 584FullO3CPU<Impl>::tick() 585{ 586 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 587 588 ++numCycles; 589 590// activity = false; 591 592 //Tick each of the stages 593 fetch.tick(); 594 595 decode.tick(); 596 597 rename.tick(); 598 599 iew.tick(); 600 601 commit.tick(); 602 603 if (!FullSystem) 604 doContextSwitch(); 605 606 // Now advance the time buffers 607 timeBuffer.advance(); 608 609 fetchQueue.advance(); 610 decodeQueue.advance(); 611 renameQueue.advance(); 612 iewQueue.advance(); 613 614 activityRec.advance(); 615 616 if (removeInstsThisCycle) { 617 cleanUpRemovedInsts(); 618 } 619 620 if (!tickEvent.scheduled()) { 621 if (_status == SwitchedOut || 622 getDrainState() == Drainable::Drained) { 623 DPRINTF(O3CPU, "Switched out!\n"); 624 // increment stat 625 lastRunningCycle = curCycle(); 626 } else if (!activityRec.active() || _status == Idle) { 627 DPRINTF(O3CPU, "Idle!\n"); 628 lastRunningCycle = curCycle(); 629 timesIdled++; 630 } else { 631 schedule(tickEvent, clockEdge(Cycles(1))); 632 DPRINTF(O3CPU, "Scheduling next tick!\n"); 633 } 634 } 635 636 if (!FullSystem) 637 updateThreadPriority(); 638} 639 640template <class Impl> 641void 642FullO3CPU<Impl>::init() 643{ 644 BaseCPU::init(); 645 646 if (!params()->switched_out && 647 system->getMemoryMode() != Enums::timing) { 648 fatal("The O3 CPU requires the memory system to be in " 649 "'timing' mode.\n"); 650 } 651 652 for (ThreadID tid = 0; tid < numThreads; ++tid) { 653 // Set noSquashFromTC so that the CPU doesn't squash when initially 654 // setting up registers. 655 thread[tid]->noSquashFromTC = true; 656 // Initialise the ThreadContext's memory proxies 657 thread[tid]->initMemProxies(thread[tid]->getTC()); 658 } 659 660 // this CPU could still be unconnected if we are restoring from a 661 // checkpoint and this CPU is to be switched in, thus we can only 662 // do this here if the instruction port is actually connected, if 663 // not we have to do it as part of takeOverFrom 664 if (icachePort.isConnected()) 665 fetch.setIcache(); 666 667 if (FullSystem && !params()->switched_out) { 668 for (ThreadID tid = 0; tid < numThreads; tid++) { 669 ThreadContext *src_tc = threadContexts[tid]; 670 TheISA::initCPU(src_tc, src_tc->contextId()); 671 } 672 } 673 674 // Clear noSquashFromTC. 675 for (int tid = 0; tid < numThreads; ++tid) 676 thread[tid]->noSquashFromTC = false; 677 678 commit.setThreads(thread); 679} 680 681template <class Impl> 682void 683FullO3CPU<Impl>::startup() 684{ 685 fetch.startupStage(); 686 iew.startupStage(); 687 rename.startupStage(); 688 commit.startupStage(); 689} 690 691template <class Impl> 692void 693FullO3CPU<Impl>::activateThread(ThreadID tid) 694{ 695 list<ThreadID>::iterator isActive = 696 std::find(activeThreads.begin(), activeThreads.end(), tid); 697 698 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 699 700 if (isActive == activeThreads.end()) { 701 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 702 tid); 703 704 activeThreads.push_back(tid); 705 } 706} 707 708template <class Impl> 709void 710FullO3CPU<Impl>::deactivateThread(ThreadID tid) 711{ 712 //Remove From Active List, if Active 713 list<ThreadID>::iterator thread_it = 714 std::find(activeThreads.begin(), activeThreads.end(), tid); 715 716 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 717 718 if (thread_it != activeThreads.end()) { 719 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 720 tid); 721 activeThreads.erase(thread_it); 722 } 723} 724 725template <class Impl> 726Counter 727FullO3CPU<Impl>::totalInsts() const 728{ 729 Counter total(0); 730 731 ThreadID size = thread.size(); 732 for (ThreadID i = 0; i < size; i++) 733 total += thread[i]->numInst; 734 735 return total; 736} 737 738template <class Impl> 739Counter 740FullO3CPU<Impl>::totalOps() const 741{ 742 Counter total(0); 743 744 ThreadID size = thread.size(); 745 for (ThreadID i = 0; i < size; i++) 746 total += thread[i]->numOp; 747 748 return total; 749} 750 751template <class Impl> 752void 753FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay) 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 // If we are time 0 or if the last activation time is in the past, 765 // schedule the next tick and wake up the fetch unit 766 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 767 scheduleTickEvent(delay); 768 769 // Be sure to signal that there's some activity so the CPU doesn't 770 // deschedule itself. 771 activityRec.activity(); 772 fetch.wakeFromQuiesce(); 773 774 Cycles cycles(curCycle() - lastRunningCycle); 775 // @todo: This is an oddity that is only here to match the stats 776 if (cycles != 0) 777 --cycles; 778 quiesceCycles += cycles; 779 780 lastActivatedCycle = curTick(); 781 782 _status = Running; 783 } 784} 785 786template <class Impl> 787bool 788FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 789 Cycles delay) 790{ 791 // Schedule removal of thread data from CPU 792 if (delay){ 793 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 794 "on tick %d\n", tid, clockEdge(delay)); 795 scheduleDeallocateContextEvent(tid, remove, delay); 796 return false; 797 } else { 798 deactivateThread(tid); 799 if (remove) 800 removeThread(tid); 801 return true; 802 } 803} 804 805template <class Impl> 806void 807FullO3CPU<Impl>::suspendContext(ThreadID tid) 808{ 809 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 810 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1)); 811 // If this was the last thread then unschedule the tick event. 812 if ((activeThreads.size() == 1 && !deallocated) || 813 activeThreads.size() == 0) 814 unscheduleTickEvent(); 815 816 DPRINTF(Quiesce, "Suspending Context\n"); 817 lastRunningCycle = curCycle(); 818 _status = Idle; 819} 820 821template <class Impl> 822void 823FullO3CPU<Impl>::haltContext(ThreadID tid) 824{ 825 //For now, this is the same as deallocate 826 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 827 scheduleDeallocateContext(tid, true, Cycles(1)); 828} 829 830template <class Impl> 831void 832FullO3CPU<Impl>::insertThread(ThreadID tid) 833{ 834 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 835 // Will change now that the PC and thread state is internal to the CPU 836 // and not in the ThreadContext. 837 ThreadContext *src_tc; 838 if (FullSystem) 839 src_tc = system->threadContexts[tid]; 840 else 841 src_tc = tcBase(tid); 842 843 //Bind Int Regs to Rename Map 844 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 845 PhysRegIndex phys_reg = freeList.getIntReg(); 846 847 renameMap[tid].setEntry(ireg,phys_reg); 848 scoreboard.setReg(phys_reg); 849 } 850 851 //Bind Float Regs to Rename Map 852 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 853 PhysRegIndex phys_reg = freeList.getFloatReg(); 854 855 renameMap[tid].setEntry(freg,phys_reg); 856 scoreboard.setReg(phys_reg); 857 } 858 859 //Copy Thread Data Into RegFile 860 //this->copyFromTC(tid); 861 862 //Set PC/NPC/NNPC 863 pcState(src_tc->pcState(), tid); 864 865 src_tc->setStatus(ThreadContext::Active); 866 867 activateContext(tid, Cycles(1)); 868 869 //Reset ROB/IQ/LSQ Entries 870 commit.rob->resetEntries(); 871 iew.resetEntries(); 872} 873 874template <class Impl> 875void 876FullO3CPU<Impl>::removeThread(ThreadID tid) 877{ 878 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 879 880 // Copy Thread Data From RegFile 881 // If thread is suspended, it might be re-allocated 882 // this->copyToTC(tid); 883 884 885 // @todo: 2-27-2008: Fix how we free up rename mappings 886 // here to alleviate the case for double-freeing registers 887 // in SMT workloads. 888 889 // Unbind Int Regs from Rename Map 890 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 891 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 892 893 scoreboard.unsetReg(phys_reg); 894 freeList.addReg(phys_reg); 895 } 896 897 // Unbind Float Regs from Rename Map 898 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 899 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 900 901 scoreboard.unsetReg(phys_reg); 902 freeList.addReg(phys_reg); 903 } 904 905 // Squash Throughout Pipeline 906 DynInstPtr inst = commit.rob->readHeadInst(tid); 907 InstSeqNum squash_seq_num = inst->seqNum; 908 fetch.squash(0, squash_seq_num, inst, tid); 909 decode.squash(tid); 910 rename.squash(squash_seq_num, tid); 911 iew.squash(tid); 912 iew.ldstQueue.squash(squash_seq_num, tid); 913 commit.rob->squash(squash_seq_num, tid); 914 915 916 assert(iew.instQueue.getCount(tid) == 0); 917 assert(iew.ldstQueue.getCount(tid) == 0); 918 919 // Reset ROB/IQ/LSQ Entries 920 921 // Commented out for now. This should be possible to do by 922 // telling all the pipeline stages to drain first, and then 923 // checking until the drain completes. Once the pipeline is 924 // drained, call resetEntries(). - 10-09-06 ktlim 925/* 926 if (activeThreads.size() >= 1) { 927 commit.rob->resetEntries(); 928 iew.resetEntries(); 929 } 930*/ 931} 932 933 934template <class Impl> 935void 936FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 937{ 938 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 939 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 940 tid); 941 942 bool ready = true; 943 944 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 945 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 946 "Phys. Int. Regs.\n", 947 tid); 948 ready = false; 949 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 950 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 951 "Phys. Float. Regs.\n", 952 tid); 953 ready = false; 954 } else if (commit.rob->numFreeEntries() >= 955 commit.rob->entryAmount(activeThreads.size() + 1)) { 956 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 957 "ROB entries.\n", 958 tid); 959 ready = false; 960 } else if (iew.instQueue.numFreeEntries() >= 961 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 962 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 963 "IQ entries.\n", 964 tid); 965 ready = false; 966 } else if (iew.ldstQueue.numFreeEntries() >= 967 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 968 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 969 "LSQ entries.\n", 970 tid); 971 ready = false; 972 } 973 974 if (ready) { 975 insertThread(tid); 976 977 contextSwitch = false; 978 979 cpuWaitList.remove(tid); 980 } else { 981 suspendContext(tid); 982 983 //blocks fetch 984 contextSwitch = true; 985 986 //@todo: dont always add to waitlist 987 //do waitlist 988 cpuWaitList.push_back(tid); 989 } 990} 991 992template <class Impl> 993Fault 994FullO3CPU<Impl>::hwrei(ThreadID tid) 995{ 996#if THE_ISA == ALPHA_ISA 997 // Need to clear the lock flag upon returning from an interrupt. 998 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 999 1000 this->thread[tid]->kernelStats->hwrei(); 1001 1002 // FIXME: XXX check for interrupts? XXX 1003#endif 1004 return NoFault; 1005} 1006 1007template <class Impl> 1008bool 1009FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1010{ 1011#if THE_ISA == ALPHA_ISA 1012 if (this->thread[tid]->kernelStats) 1013 this->thread[tid]->kernelStats->callpal(palFunc, 1014 this->threadContexts[tid]); 1015 1016 switch (palFunc) { 1017 case PAL::halt: 1018 halt(); 1019 if (--System::numSystemsRunning == 0) 1020 exitSimLoop("all cpus halted"); 1021 break; 1022 1023 case PAL::bpt: 1024 case PAL::bugchk: 1025 if (this->system->breakpoint()) 1026 return false; 1027 break; 1028 } 1029#endif 1030 return true; 1031} 1032 1033template <class Impl> 1034Fault 1035FullO3CPU<Impl>::getInterrupts() 1036{ 1037 // Check if there are any outstanding interrupts 1038 return this->interrupts->getInterrupt(this->threadContexts[0]); 1039} 1040 1041template <class Impl> 1042void 1043FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1044{ 1045 // Check for interrupts here. For now can copy the code that 1046 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1047 // is the one that handles the interrupts. 1048 // @todo: Possibly consolidate the interrupt checking code. 1049 // @todo: Allow other threads to handle interrupts. 1050 1051 assert(interrupt != NoFault); 1052 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1053 1054 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1055 this->trap(interrupt, 0, NULL); 1056} 1057 1058template <class Impl> 1059void 1060FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1061{ 1062 // Pass the thread's TC into the invoke method. 1063 fault->invoke(this->threadContexts[tid], inst); 1064} 1065 1066template <class Impl> 1067void 1068FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1069{ 1070 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1071 1072 DPRINTF(Activity,"Activity: syscall() called.\n"); 1073 1074 // Temporarily increase this by one to account for the syscall 1075 // instruction. 1076 ++(this->thread[tid]->funcExeInst); 1077 1078 // Execute the actual syscall. 1079 this->thread[tid]->syscall(callnum); 1080 1081 // Decrease funcExeInst by one as the normal commit will handle 1082 // incrementing it. 1083 --(this->thread[tid]->funcExeInst); 1084} 1085 1086template <class Impl> 1087void 1088FullO3CPU<Impl>::serialize(std::ostream &os) 1089{ 1090 Drainable::State so_state(getDrainState()); 1091 SERIALIZE_ENUM(so_state); 1092 BaseCPU::serialize(os); 1093 nameOut(os, csprintf("%s.tickEvent", name())); 1094 tickEvent.serialize(os); 1095 1096 for (ThreadID i = 0; i < thread.size(); i++) { 1097 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1098 thread[i]->serialize(os); 1099 } 1100} 1101 1102template <class Impl> 1103void 1104FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1105{ 1106 Drainable::State so_state; 1107 UNSERIALIZE_ENUM(so_state); 1108 BaseCPU::unserialize(cp, section); 1109 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1110 1111 for (ThreadID i = 0; i < thread.size(); i++) { 1112 thread[i]->unserialize(cp, 1113 csprintf("%s.xc.%i", section, i)); 1114 if (thread[i]->status() == ThreadContext::Active) 1115 activateThread(i); 1116 } 1117} 1118 1119template <class Impl> 1120unsigned int 1121FullO3CPU<Impl>::drain(DrainManager *drain_manager) 1122{ 1123 DPRINTF(O3CPU, "Switching out\n"); 1124 1125 // If the CPU isn't doing anything, then return immediately. 1126 if (_status == SwitchedOut) 1127 return 0; 1128 1129 drainCount = 0; 1130 fetch.drain(); 1131 decode.drain(); 1132 rename.drain(); 1133 iew.drain(); 1134 commit.drain(); 1135 1136 // Wake the CPU and record activity so everything can drain out if 1137 // the CPU was not able to immediately drain. 1138 if (getDrainState() != Drainable::Drained) { 1139 // A bit of a hack...set the drainManager after all the drain() 1140 // calls have been made, that way if all of the stages drain 1141 // immediately, the signalDrained() function knows not to call 1142 // process on the drain event. 1143 drainManager = drain_manager; 1144 1145 wakeCPU(); 1146 activityRec.activity(); 1147 1148 DPRINTF(Drain, "CPU not drained\n"); 1149 1150 return 1; 1151 } else { 1152 return 0; 1153 } 1154} 1155 1156template <class Impl> 1157void 1158FullO3CPU<Impl>::drainResume() 1159{ 1160 fetch.resume(); 1161 decode.resume(); 1162 rename.resume(); 1163 iew.resume(); 1164 commit.resume(); 1165 1166 setDrainState(Drainable::Running); 1167 1168 if (_status == SwitchedOut) 1169 return; 1170 1171 if (system->getMemoryMode() != Enums::timing) { 1172 fatal("The O3 CPU requires the memory system to be in " 1173 "'timing' mode.\n"); 1174 } 1175 1176 if (!tickEvent.scheduled()) 1177 schedule(tickEvent, nextCycle()); 1178 _status = Running; 1179} 1180 1181template <class Impl> 1182void 1183FullO3CPU<Impl>::signalDrained() 1184{ 1185 if (++drainCount == NumStages) { 1186 if (tickEvent.scheduled()) 1187 tickEvent.squash(); 1188 1189 setDrainState(Drainable::Drained); 1190 1191 if (drainManager) { 1192 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1193 drainManager->signalDrainDone(); 1194 drainManager = NULL; 1195 } 1196 } 1197 assert(drainCount <= 5); 1198} 1199 1200template <class Impl> 1201void 1202FullO3CPU<Impl>::switchOut() 1203{ 1204 BaseCPU::switchOut(); 1205 1206 fetch.switchOut(); 1207 rename.switchOut(); 1208 iew.switchOut(); 1209 commit.switchOut(); 1210 instList.clear(); 1211 while (!removeList.empty()) { 1212 removeList.pop(); 1213 } 1214 1215 _status = SwitchedOut; 1216 1217 if (checker) 1218 checker->switchOut(); 1219 1220 if (tickEvent.scheduled()) 1221 tickEvent.squash(); 1222} 1223 1224template <class Impl> 1225void 1226FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1227{ 1228 // Flush out any old data from the time buffers. 1229 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1230 timeBuffer.advance(); 1231 fetchQueue.advance(); 1232 decodeQueue.advance(); 1233 renameQueue.advance(); 1234 iewQueue.advance(); 1235 } 1236 1237 activityRec.reset(); 1238 1239 BaseCPU::takeOverFrom(oldCPU); 1240 1241 fetch.takeOverFrom(); 1242 decode.takeOverFrom(); 1243 rename.takeOverFrom(); 1244 iew.takeOverFrom(); 1245 commit.takeOverFrom(); 1246 1247 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1248 1249 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1250 if (oldO3CPU) 1251 globalSeqNum = oldO3CPU->globalSeqNum; 1252 1253 // @todo: Figure out how to properly select the tid to put onto 1254 // the active threads list. 1255 ThreadID tid = 0; 1256 1257 list<ThreadID>::iterator isActive = 1258 std::find(activeThreads.begin(), activeThreads.end(), tid); 1259 1260 if (isActive == activeThreads.end()) { 1261 //May Need to Re-code this if the delay variable is the delay 1262 //needed for thread to activate 1263 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1264 tid); 1265 1266 activeThreads.push_back(tid); 1267 } 1268 1269 // Set all statuses to active, schedule the CPU's tick event. 1270 // @todo: Fix up statuses so this is handled properly 1271 ThreadID size = threadContexts.size(); 1272 for (ThreadID i = 0; i < size; ++i) { 1273 ThreadContext *tc = threadContexts[i]; 1274 if (tc->status() == ThreadContext::Active && _status != Running) { 1275 _status = Running; 1276 reschedule(tickEvent, nextCycle(), true); 1277 } 1278 } 1279 if (!tickEvent.scheduled()) 1280 schedule(tickEvent, nextCycle()); 1281 1282 lastRunningCycle = curCycle(); 1283} 1284 1285template <class Impl> 1286TheISA::MiscReg 1287FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1288{ 1289 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1290} 1291 1292template <class Impl> 1293TheISA::MiscReg 1294FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1295{ 1296 miscRegfileReads++; 1297 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1298} 1299 1300template <class Impl> 1301void 1302FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1303 const TheISA::MiscReg &val, ThreadID tid) 1304{ 1305 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1306} 1307 1308template <class Impl> 1309void 1310FullO3CPU<Impl>::setMiscReg(int misc_reg, 1311 const TheISA::MiscReg &val, ThreadID tid) 1312{ 1313 miscRegfileWrites++; 1314 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1315} 1316 1317template <class Impl> 1318uint64_t 1319FullO3CPU<Impl>::readIntReg(int reg_idx) 1320{ 1321 intRegfileReads++; 1322 return regFile.readIntReg(reg_idx); 1323} 1324 1325template <class Impl> 1326FloatReg 1327FullO3CPU<Impl>::readFloatReg(int reg_idx) 1328{ 1329 fpRegfileReads++; 1330 return regFile.readFloatReg(reg_idx); 1331} 1332 1333template <class Impl> 1334FloatRegBits 1335FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1336{ 1337 fpRegfileReads++; 1338 return regFile.readFloatRegBits(reg_idx); 1339} 1340 1341template <class Impl> 1342void 1343FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1344{ 1345 intRegfileWrites++; 1346 regFile.setIntReg(reg_idx, val); 1347} 1348 1349template <class Impl> 1350void 1351FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1352{ 1353 fpRegfileWrites++; 1354 regFile.setFloatReg(reg_idx, val); 1355} 1356 1357template <class Impl> 1358void 1359FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1360{ 1361 fpRegfileWrites++; 1362 regFile.setFloatRegBits(reg_idx, val); 1363} 1364 1365template <class Impl> 1366uint64_t 1367FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1368{ 1369 intRegfileReads++; 1370 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1371 1372 return regFile.readIntReg(phys_reg); 1373} 1374 1375template <class Impl> 1376float 1377FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1378{ 1379 fpRegfileReads++; 1380 int idx = reg_idx + TheISA::NumIntRegs; 1381 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1382 1383 return regFile.readFloatReg(phys_reg); 1384} 1385 1386template <class Impl> 1387uint64_t 1388FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1389{ 1390 fpRegfileReads++; 1391 int idx = reg_idx + TheISA::NumIntRegs; 1392 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1393 1394 return regFile.readFloatRegBits(phys_reg); 1395} 1396 1397template <class Impl> 1398void 1399FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1400{ 1401 intRegfileWrites++; 1402 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1403 1404 regFile.setIntReg(phys_reg, val); 1405} 1406 1407template <class Impl> 1408void 1409FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1410{ 1411 fpRegfileWrites++; 1412 int idx = reg_idx + TheISA::NumIntRegs; 1413 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1414 1415 regFile.setFloatReg(phys_reg, val); 1416} 1417 1418template <class Impl> 1419void 1420FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1421{ 1422 fpRegfileWrites++; 1423 int idx = reg_idx + TheISA::NumIntRegs; 1424 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1425 1426 regFile.setFloatRegBits(phys_reg, val); 1427} 1428 1429template <class Impl> 1430TheISA::PCState 1431FullO3CPU<Impl>::pcState(ThreadID tid) 1432{ 1433 return commit.pcState(tid); 1434} 1435 1436template <class Impl> 1437void 1438FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1439{ 1440 commit.pcState(val, tid); 1441} 1442 1443template <class Impl> 1444Addr 1445FullO3CPU<Impl>::instAddr(ThreadID tid) 1446{ 1447 return commit.instAddr(tid); 1448} 1449 1450template <class Impl> 1451Addr 1452FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1453{ 1454 return commit.nextInstAddr(tid); 1455} 1456 1457template <class Impl> 1458MicroPC 1459FullO3CPU<Impl>::microPC(ThreadID tid) 1460{ 1461 return commit.microPC(tid); 1462} 1463 1464template <class Impl> 1465void 1466FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1467{ 1468 this->thread[tid]->noSquashFromTC = true; 1469 this->commit.generateTCEvent(tid); 1470} 1471 1472template <class Impl> 1473typename FullO3CPU<Impl>::ListIt 1474FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1475{ 1476 instList.push_back(inst); 1477 1478 return --(instList.end()); 1479} 1480 1481template <class Impl> 1482void 1483FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1484{ 1485 // Keep an instruction count. 1486 if (!inst->isMicroop() || inst->isLastMicroop()) { 1487 thread[tid]->numInst++; 1488 thread[tid]->numInsts++; 1489 committedInsts[tid]++; 1490 totalCommittedInsts++; 1491 } 1492 thread[tid]->numOp++; 1493 thread[tid]->numOps++; 1494 committedOps[tid]++; 1495 1496 system->totalNumInsts++; 1497 // Check for instruction-count-based events. 1498 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1499 system->instEventQueue.serviceEvents(system->totalNumInsts); 1500} 1501 1502template <class Impl> 1503void 1504FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1505{ 1506 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1507 "[sn:%lli]\n", 1508 inst->threadNumber, inst->pcState(), inst->seqNum); 1509 1510 removeInstsThisCycle = true; 1511 1512 // Remove the front instruction. 1513 removeList.push(inst->getInstListIt()); 1514} 1515 1516template <class Impl> 1517void 1518FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1519{ 1520 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1521 " list.\n", tid); 1522 1523 ListIt end_it; 1524 1525 bool rob_empty = false; 1526 1527 if (instList.empty()) { 1528 return; 1529 } else if (rob.isEmpty(/*tid*/)) { 1530 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1531 end_it = instList.begin(); 1532 rob_empty = true; 1533 } else { 1534 end_it = (rob.readTailInst(tid))->getInstListIt(); 1535 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1536 } 1537 1538 removeInstsThisCycle = true; 1539 1540 ListIt inst_it = instList.end(); 1541 1542 inst_it--; 1543 1544 // Walk through the instruction list, removing any instructions 1545 // that were inserted after the given instruction iterator, end_it. 1546 while (inst_it != end_it) { 1547 assert(!instList.empty()); 1548 1549 squashInstIt(inst_it, tid); 1550 1551 inst_it--; 1552 } 1553 1554 // If the ROB was empty, then we actually need to remove the first 1555 // instruction as well. 1556 if (rob_empty) { 1557 squashInstIt(inst_it, tid); 1558 } 1559} 1560 1561template <class Impl> 1562void 1563FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1564{ 1565 assert(!instList.empty()); 1566 1567 removeInstsThisCycle = true; 1568 1569 ListIt inst_iter = instList.end(); 1570 1571 inst_iter--; 1572 1573 DPRINTF(O3CPU, "Deleting instructions from instruction " 1574 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1575 tid, seq_num, (*inst_iter)->seqNum); 1576 1577 while ((*inst_iter)->seqNum > seq_num) { 1578 1579 bool break_loop = (inst_iter == instList.begin()); 1580 1581 squashInstIt(inst_iter, tid); 1582 1583 inst_iter--; 1584 1585 if (break_loop) 1586 break; 1587 } 1588} 1589 1590template <class Impl> 1591inline void 1592FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1593{ 1594 if ((*instIt)->threadNumber == tid) { 1595 DPRINTF(O3CPU, "Squashing instruction, " 1596 "[tid:%i] [sn:%lli] PC %s\n", 1597 (*instIt)->threadNumber, 1598 (*instIt)->seqNum, 1599 (*instIt)->pcState()); 1600 1601 // Mark it as squashed. 1602 (*instIt)->setSquashed(); 1603 1604 // @todo: Formulate a consistent method for deleting 1605 // instructions from the instruction list 1606 // Remove the instruction from the list. 1607 removeList.push(instIt); 1608 } 1609} 1610 1611template <class Impl> 1612void 1613FullO3CPU<Impl>::cleanUpRemovedInsts() 1614{ 1615 while (!removeList.empty()) { 1616 DPRINTF(O3CPU, "Removing instruction, " 1617 "[tid:%i] [sn:%lli] PC %s\n", 1618 (*removeList.front())->threadNumber, 1619 (*removeList.front())->seqNum, 1620 (*removeList.front())->pcState()); 1621 1622 instList.erase(removeList.front()); 1623 1624 removeList.pop(); 1625 } 1626 1627 removeInstsThisCycle = false; 1628} 1629/* 1630template <class Impl> 1631void 1632FullO3CPU<Impl>::removeAllInsts() 1633{ 1634 instList.clear(); 1635} 1636*/ 1637template <class Impl> 1638void 1639FullO3CPU<Impl>::dumpInsts() 1640{ 1641 int num = 0; 1642 1643 ListIt inst_list_it = instList.begin(); 1644 1645 cprintf("Dumping Instruction List\n"); 1646 1647 while (inst_list_it != instList.end()) { 1648 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1649 "Squashed:%i\n\n", 1650 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1651 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1652 (*inst_list_it)->isSquashed()); 1653 inst_list_it++; 1654 ++num; 1655 } 1656} 1657/* 1658template <class Impl> 1659void 1660FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1661{ 1662 iew.wakeDependents(inst); 1663} 1664*/ 1665template <class Impl> 1666void 1667FullO3CPU<Impl>::wakeCPU() 1668{ 1669 if (activityRec.active() || tickEvent.scheduled()) { 1670 DPRINTF(Activity, "CPU already running.\n"); 1671 return; 1672 } 1673 1674 DPRINTF(Activity, "Waking up CPU\n"); 1675 1676 Cycles cycles(curCycle() - lastRunningCycle); 1677 // @todo: This is an oddity that is only here to match the stats 1678 if (cycles != 0) 1679 --cycles; 1680 idleCycles += cycles; 1681 numCycles += cycles; 1682 1683 schedule(tickEvent, nextCycle()); 1684} 1685 1686template <class Impl> 1687void 1688FullO3CPU<Impl>::wakeup() 1689{ 1690 if (this->thread[0]->status() != ThreadContext::Suspended) 1691 return; 1692 1693 this->wakeCPU(); 1694 1695 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1696 this->threadContexts[0]->activate(); 1697} 1698 1699template <class Impl> 1700ThreadID 1701FullO3CPU<Impl>::getFreeTid() 1702{ 1703 for (ThreadID tid = 0; tid < numThreads; tid++) { 1704 if (!tids[tid]) { 1705 tids[tid] = true; 1706 return tid; 1707 } 1708 } 1709 1710 return InvalidThreadID; 1711} 1712 1713template <class Impl> 1714void 1715FullO3CPU<Impl>::doContextSwitch() 1716{ 1717 if (contextSwitch) { 1718 1719 //ADD CODE TO DEACTIVE THREAD HERE (???) 1720 1721 ThreadID size = cpuWaitList.size(); 1722 for (ThreadID tid = 0; tid < size; tid++) { 1723 activateWhenReady(tid); 1724 } 1725 1726 if (cpuWaitList.size() == 0) 1727 contextSwitch = true; 1728 } 1729} 1730 1731template <class Impl> 1732void 1733FullO3CPU<Impl>::updateThreadPriority() 1734{ 1735 if (activeThreads.size() > 1) { 1736 //DEFAULT TO ROUND ROBIN SCHEME 1737 //e.g. Move highest priority to end of thread list 1738 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1739 1740 unsigned high_thread = *list_begin; 1741 1742 activeThreads.erase(list_begin); 1743 1744 activeThreads.push_back(high_thread); 1745 } 1746} 1747 1748// Forward declaration of FullO3CPU. 1749template class FullO3CPU<O3CPUImpl>; 1750