cpu.cc revision 5714:76abee886def
1/* 2 * Copyright (c) 2004-2006 The Regents of The University of Michigan 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are 7 * met: redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer; 9 * redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution; 12 * neither the name of the copyright holders nor the names of its 13 * contributors may be used to endorse or promote products derived from 14 * this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 * 28 * Authors: Kevin Lim 29 * Korey Sewell 30 */ 31 32#include "config/full_system.hh" 33#include "config/use_checker.hh" 34 35#include "cpu/activity.hh" 36#include "cpu/simple_thread.hh" 37#include "cpu/thread_context.hh" 38#include "cpu/o3/isa_specific.hh" 39#include "cpu/o3/cpu.hh" 40#include "cpu/o3/thread_context.hh" 41#include "enums/MemoryMode.hh" 42#include "sim/core.hh" 43#include "sim/stat_control.hh" 44 45#if FULL_SYSTEM 46#include "cpu/quiesce_event.hh" 47#include "sim/system.hh" 48#else 49#include "sim/process.hh" 50#endif 51 52#if USE_CHECKER 53#include "cpu/checker/cpu.hh" 54#endif 55 56#if THE_ISA == ALPHA_ISA 57#include "arch/alpha/osfpal.hh" 58#endif 59 60class BaseCPUParams; 61 62using namespace TheISA; 63 64BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 65 : BaseCPU(params) 66{ 67} 68 69void 70BaseO3CPU::regStats() 71{ 72 BaseCPU::regStats(); 73} 74 75template <class Impl> 76FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 77 : Event(CPU_Tick_Pri), cpu(c) 78{ 79} 80 81template <class Impl> 82void 83FullO3CPU<Impl>::TickEvent::process() 84{ 85 cpu->tick(); 86} 87 88template <class Impl> 89const char * 90FullO3CPU<Impl>::TickEvent::description() const 91{ 92 return "FullO3CPU tick"; 93} 94 95template <class Impl> 96FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 97 : Event(CPU_Switch_Pri) 98{ 99} 100 101template <class Impl> 102void 103FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 104 FullO3CPU<Impl> *thread_cpu) 105{ 106 tid = thread_num; 107 cpu = thread_cpu; 108} 109 110template <class Impl> 111void 112FullO3CPU<Impl>::ActivateThreadEvent::process() 113{ 114 cpu->activateThread(tid); 115} 116 117template <class Impl> 118const char * 119FullO3CPU<Impl>::ActivateThreadEvent::description() const 120{ 121 return "FullO3CPU \"Activate Thread\""; 122} 123 124template <class Impl> 125FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 126 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 127{ 128} 129 130template <class Impl> 131void 132FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 133 FullO3CPU<Impl> *thread_cpu) 134{ 135 tid = thread_num; 136 cpu = thread_cpu; 137 remove = false; 138} 139 140template <class Impl> 141void 142FullO3CPU<Impl>::DeallocateContextEvent::process() 143{ 144 cpu->deactivateThread(tid); 145 if (remove) 146 cpu->removeThread(tid); 147} 148 149template <class Impl> 150const char * 151FullO3CPU<Impl>::DeallocateContextEvent::description() const 152{ 153 return "FullO3CPU \"Deallocate Context\""; 154} 155 156template <class Impl> 157FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 158 : BaseO3CPU(params), 159 itb(params->itb), 160 dtb(params->dtb), 161 tickEvent(this), 162 removeInstsThisCycle(false), 163 fetch(this, params), 164 decode(this, params), 165 rename(this, params), 166 iew(this, params), 167 commit(this, params), 168 169 regFile(this, params->numPhysIntRegs, 170 params->numPhysFloatRegs), 171 172 freeList(params->numThreads, 173 TheISA::NumIntRegs, params->numPhysIntRegs, 174 TheISA::NumFloatRegs, params->numPhysFloatRegs), 175 176 rob(this, 177 params->numROBEntries, params->squashWidth, 178 params->smtROBPolicy, params->smtROBThreshold, 179 params->numThreads), 180 181 scoreboard(params->numThreads, 182 TheISA::NumIntRegs, params->numPhysIntRegs, 183 TheISA::NumFloatRegs, params->numPhysFloatRegs, 184 TheISA::NumMiscRegs * number_of_threads, 185 TheISA::ZeroReg), 186 187 timeBuffer(params->backComSize, params->forwardComSize), 188 fetchQueue(params->backComSize, params->forwardComSize), 189 decodeQueue(params->backComSize, params->forwardComSize), 190 renameQueue(params->backComSize, params->forwardComSize), 191 iewQueue(params->backComSize, params->forwardComSize), 192 activityRec(NumStages, 193 params->backComSize + params->forwardComSize, 194 params->activity), 195 196 globalSeqNum(1), 197#if FULL_SYSTEM 198 system(params->system), 199 physmem(system->physmem), 200#endif // FULL_SYSTEM 201 drainCount(0), 202 deferRegistration(params->defer_registration), 203 numThreads(number_of_threads) 204{ 205 if (!deferRegistration) { 206 _status = Running; 207 } else { 208 _status = Idle; 209 } 210 211#if USE_CHECKER 212 if (params->checker) { 213 BaseCPU *temp_checker = params->checker; 214 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 215#if FULL_SYSTEM 216 checker->setSystem(params->system); 217#endif 218 } else { 219 checker = NULL; 220 } 221#endif // USE_CHECKER 222 223#if !FULL_SYSTEM 224 thread.resize(number_of_threads); 225 tids.resize(number_of_threads); 226#endif 227 228 // The stages also need their CPU pointer setup. However this 229 // must be done at the upper level CPU because they have pointers 230 // to the upper level CPU, and not this FullO3CPU. 231 232 // Set up Pointers to the activeThreads list for each stage 233 fetch.setActiveThreads(&activeThreads); 234 decode.setActiveThreads(&activeThreads); 235 rename.setActiveThreads(&activeThreads); 236 iew.setActiveThreads(&activeThreads); 237 commit.setActiveThreads(&activeThreads); 238 239 // Give each of the stages the time buffer they will use. 240 fetch.setTimeBuffer(&timeBuffer); 241 decode.setTimeBuffer(&timeBuffer); 242 rename.setTimeBuffer(&timeBuffer); 243 iew.setTimeBuffer(&timeBuffer); 244 commit.setTimeBuffer(&timeBuffer); 245 246 // Also setup each of the stages' queues. 247 fetch.setFetchQueue(&fetchQueue); 248 decode.setFetchQueue(&fetchQueue); 249 commit.setFetchQueue(&fetchQueue); 250 decode.setDecodeQueue(&decodeQueue); 251 rename.setDecodeQueue(&decodeQueue); 252 rename.setRenameQueue(&renameQueue); 253 iew.setRenameQueue(&renameQueue); 254 iew.setIEWQueue(&iewQueue); 255 commit.setIEWQueue(&iewQueue); 256 commit.setRenameQueue(&renameQueue); 257 258 commit.setIEWStage(&iew); 259 rename.setIEWStage(&iew); 260 rename.setCommitStage(&commit); 261 262#if !FULL_SYSTEM 263 int active_threads = params->workload.size(); 264 265 if (active_threads > Impl::MaxThreads) { 266 panic("Workload Size too large. Increase the 'MaxThreads'" 267 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or " 268 "edit your workload size."); 269 } 270#else 271 int active_threads = 1; 272#endif 273 274 //Make Sure That this a Valid Architeture 275 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 276 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 277 278 rename.setScoreboard(&scoreboard); 279 iew.setScoreboard(&scoreboard); 280 281 // Setup the rename map for whichever stages need it. 282 PhysRegIndex lreg_idx = 0; 283 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 284 285 for (int tid=0; tid < numThreads; tid++) { 286 bool bindRegs = (tid <= active_threads - 1); 287 288 commitRenameMap[tid].init(TheISA::NumIntRegs, 289 params->numPhysIntRegs, 290 lreg_idx, //Index for Logical. Regs 291 292 TheISA::NumFloatRegs, 293 params->numPhysFloatRegs, 294 freg_idx, //Index for Float Regs 295 296 TheISA::NumMiscRegs, 297 298 TheISA::ZeroReg, 299 TheISA::ZeroReg, 300 301 tid, 302 false); 303 304 renameMap[tid].init(TheISA::NumIntRegs, 305 params->numPhysIntRegs, 306 lreg_idx, //Index for Logical. Regs 307 308 TheISA::NumFloatRegs, 309 params->numPhysFloatRegs, 310 freg_idx, //Index for Float Regs 311 312 TheISA::NumMiscRegs, 313 314 TheISA::ZeroReg, 315 TheISA::ZeroReg, 316 317 tid, 318 bindRegs); 319 320 activateThreadEvent[tid].init(tid, this); 321 deallocateContextEvent[tid].init(tid, this); 322 } 323 324 rename.setRenameMap(renameMap); 325 commit.setRenameMap(commitRenameMap); 326 327 // Give renameMap & rename stage access to the freeList; 328 for (int i=0; i < numThreads; i++) { 329 renameMap[i].setFreeList(&freeList); 330 } 331 rename.setFreeList(&freeList); 332 333 // Setup the ROB for whichever stages need it. 334 commit.setROB(&rob); 335 336 lastRunningCycle = curTick; 337 338 lastActivatedCycle = -1; 339 340 // Give renameMap & rename stage access to the freeList; 341 //for (int i=0; i < numThreads; i++) { 342 //globalSeqNum[i] = 1; 343 //} 344 345 contextSwitch = false; 346 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 347 348 // Setup any thread state. 349 this->thread.resize(this->numThreads); 350 351 for (int i = 0; i < this->numThreads; ++i) { 352#if FULL_SYSTEM 353 // SMT is not supported in FS mode yet. 354 assert(this->numThreads == 1); 355 this->thread[i] = new Thread(this, 0); 356 this->thread[i]->setStatus(ThreadContext::Suspended); 357#else 358 if (i < params->workload.size()) { 359 DPRINTF(O3CPU, "Workload[%i] process is %#x", 360 i, this->thread[i]); 361 this->thread[i] = new typename FullO3CPU<Impl>::Thread( 362 (typename Impl::O3CPU *)(this), 363 i, params->workload[i], i); 364 365 this->thread[i]->setStatus(ThreadContext::Suspended); 366 367 //usedTids[i] = true; 368 //threadMap[i] = i; 369 } else { 370 //Allocate Empty thread so M5 can use later 371 //when scheduling threads to CPU 372 Process* dummy_proc = NULL; 373 374 this->thread[i] = new typename FullO3CPU<Impl>::Thread( 375 (typename Impl::O3CPU *)(this), 376 i, dummy_proc, i); 377 //usedTids[i] = false; 378 } 379#endif // !FULL_SYSTEM 380 381 ThreadContext *tc; 382 383 // Setup the TC that will serve as the interface to the threads/CPU. 384 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 385 386 tc = o3_tc; 387 388 // If we're using a checker, then the TC should be the 389 // CheckerThreadContext. 390#if USE_CHECKER 391 if (params->checker) { 392 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 393 o3_tc, this->checker); 394 } 395#endif 396 397 o3_tc->cpu = (typename Impl::O3CPU *)(this); 398 assert(o3_tc->cpu); 399 o3_tc->thread = this->thread[i]; 400 401#if FULL_SYSTEM 402 // Setup quiesce event. 403 this->thread[i]->quiesceEvent = new EndQuiesceEvent(tc); 404#endif 405 // Give the thread the TC. 406 this->thread[i]->tc = tc; 407 408 // Add the TC to the CPU's list of TC's. 409 this->threadContexts.push_back(tc); 410 } 411 412 for (int i=0; i < this->numThreads; i++) { 413 this->thread[i]->setFuncExeInst(0); 414 } 415 416 lockAddr = 0; 417 lockFlag = false; 418} 419 420#if !FULL_SYSTEM 421 422template <class Impl> 423TheISA::IntReg 424FullO3CPU<Impl>::getSyscallArg(int i, int tid) 425{ 426 assert(i < TheISA::NumArgumentRegs); 427 TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), 428 TheISA::ArgumentReg[i]); 429 TheISA::IntReg val = this->readArchIntReg(idx, tid); 430#if THE_ISA == SPARC_ISA 431 if (bits(this->readMiscRegNoEffect(SparcISA::MISCREG_PSTATE, tid), 3, 3)) 432 val = bits(val, 31, 0); 433#endif 434 return val; 435} 436 437template <class Impl> 438void 439FullO3CPU<Impl>::setSyscallArg(int i, TheISA::IntReg val, int tid) 440{ 441 assert(i < TheISA::NumArgumentRegs); 442 TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), 443 TheISA::ArgumentReg[i]); 444 this->setArchIntReg(idx, val, tid); 445} 446#endif 447 448template <class Impl> 449FullO3CPU<Impl>::~FullO3CPU() 450{ 451} 452 453template <class Impl> 454void 455FullO3CPU<Impl>::regStats() 456{ 457 BaseO3CPU::regStats(); 458 459 // Register any of the O3CPU's stats here. 460 timesIdled 461 .name(name() + ".timesIdled") 462 .desc("Number of times that the entire CPU went into an idle state and" 463 " unscheduled itself") 464 .prereq(timesIdled); 465 466 idleCycles 467 .name(name() + ".idleCycles") 468 .desc("Total number of cycles that the CPU has spent unscheduled due " 469 "to idling") 470 .prereq(idleCycles); 471 472 // Number of Instructions simulated 473 // -------------------------------- 474 // Should probably be in Base CPU but need templated 475 // MaxThreads so put in here instead 476 committedInsts 477 .init(numThreads) 478 .name(name() + ".committedInsts") 479 .desc("Number of Instructions Simulated"); 480 481 totalCommittedInsts 482 .name(name() + ".committedInsts_total") 483 .desc("Number of Instructions Simulated"); 484 485 cpi 486 .name(name() + ".cpi") 487 .desc("CPI: Cycles Per Instruction") 488 .precision(6); 489 cpi = numCycles / committedInsts; 490 491 totalCpi 492 .name(name() + ".cpi_total") 493 .desc("CPI: Total CPI of All Threads") 494 .precision(6); 495 totalCpi = numCycles / totalCommittedInsts; 496 497 ipc 498 .name(name() + ".ipc") 499 .desc("IPC: Instructions Per Cycle") 500 .precision(6); 501 ipc = committedInsts / numCycles; 502 503 totalIpc 504 .name(name() + ".ipc_total") 505 .desc("IPC: Total IPC of All Threads") 506 .precision(6); 507 totalIpc = totalCommittedInsts / numCycles; 508 509 this->fetch.regStats(); 510 this->decode.regStats(); 511 this->rename.regStats(); 512 this->iew.regStats(); 513 this->commit.regStats(); 514} 515 516template <class Impl> 517Port * 518FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 519{ 520 if (if_name == "dcache_port") 521 return iew.getDcachePort(); 522 else if (if_name == "icache_port") 523 return fetch.getIcachePort(); 524 else 525 panic("No Such Port\n"); 526} 527 528template <class Impl> 529void 530FullO3CPU<Impl>::tick() 531{ 532 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 533 534 ++numCycles; 535 536// activity = false; 537 538 //Tick each of the stages 539 fetch.tick(); 540 541 decode.tick(); 542 543 rename.tick(); 544 545 iew.tick(); 546 547 commit.tick(); 548 549#if !FULL_SYSTEM 550 doContextSwitch(); 551#endif 552 553 // Now advance the time buffers 554 timeBuffer.advance(); 555 556 fetchQueue.advance(); 557 decodeQueue.advance(); 558 renameQueue.advance(); 559 iewQueue.advance(); 560 561 activityRec.advance(); 562 563 if (removeInstsThisCycle) { 564 cleanUpRemovedInsts(); 565 } 566 567 if (!tickEvent.scheduled()) { 568 if (_status == SwitchedOut || 569 getState() == SimObject::Drained) { 570 DPRINTF(O3CPU, "Switched out!\n"); 571 // increment stat 572 lastRunningCycle = curTick; 573 } else if (!activityRec.active() || _status == Idle) { 574 DPRINTF(O3CPU, "Idle!\n"); 575 lastRunningCycle = curTick; 576 timesIdled++; 577 } else { 578 schedule(tickEvent, nextCycle(curTick + ticks(1))); 579 DPRINTF(O3CPU, "Scheduling next tick!\n"); 580 } 581 } 582 583#if !FULL_SYSTEM 584 updateThreadPriority(); 585#endif 586} 587 588template <class Impl> 589void 590FullO3CPU<Impl>::init() 591{ 592 BaseCPU::init(); 593 594 // Set inSyscall so that the CPU doesn't squash when initially 595 // setting up registers. 596 for (int i = 0; i < number_of_threads; ++i) 597 thread[i]->inSyscall = true; 598 599 for (int tid=0; tid < number_of_threads; tid++) { 600#if FULL_SYSTEM 601 ThreadContext *src_tc = threadContexts[tid]; 602#else 603 ThreadContext *src_tc = thread[tid]->getTC(); 604#endif 605 // Threads start in the Suspended State 606 if (src_tc->status() != ThreadContext::Suspended) { 607 continue; 608 } 609 610#if FULL_SYSTEM 611 TheISA::initCPU(src_tc, src_tc->contextId()); 612#endif 613 } 614 615 // Clear inSyscall. 616 for (int i = 0; i < number_of_threads; ++i) 617 thread[i]->inSyscall = false; 618 619 // Initialize stages. 620 fetch.initStage(); 621 iew.initStage(); 622 rename.initStage(); 623 commit.initStage(); 624 625 commit.setThreads(thread); 626} 627 628template <class Impl> 629void 630FullO3CPU<Impl>::activateThread(unsigned tid) 631{ 632 std::list<unsigned>::iterator isActive = 633 std::find(activeThreads.begin(), activeThreads.end(), tid); 634 635 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 636 637 if (isActive == activeThreads.end()) { 638 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 639 tid); 640 641 activeThreads.push_back(tid); 642 } 643} 644 645template <class Impl> 646void 647FullO3CPU<Impl>::deactivateThread(unsigned tid) 648{ 649 //Remove From Active List, if Active 650 std::list<unsigned>::iterator thread_it = 651 std::find(activeThreads.begin(), activeThreads.end(), tid); 652 653 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 654 655 if (thread_it != activeThreads.end()) { 656 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 657 tid); 658 activeThreads.erase(thread_it); 659 } 660} 661 662template <class Impl> 663void 664FullO3CPU<Impl>::activateContext(int tid, int delay) 665{ 666 // Needs to set each stage to running as well. 667 if (delay){ 668 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 669 "on cycle %d\n", tid, curTick + ticks(delay)); 670 scheduleActivateThreadEvent(tid, delay); 671 } else { 672 activateThread(tid); 673 } 674 675 if (lastActivatedCycle < curTick) { 676 scheduleTickEvent(delay); 677 678 // Be sure to signal that there's some activity so the CPU doesn't 679 // deschedule itself. 680 activityRec.activity(); 681 fetch.wakeFromQuiesce(); 682 683 lastActivatedCycle = curTick; 684 685 _status = Running; 686 } 687} 688 689template <class Impl> 690bool 691FullO3CPU<Impl>::deallocateContext(int tid, bool remove, int delay) 692{ 693 // Schedule removal of thread data from CPU 694 if (delay){ 695 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 696 "on cycle %d\n", tid, curTick + ticks(delay)); 697 scheduleDeallocateContextEvent(tid, remove, delay); 698 return false; 699 } else { 700 deactivateThread(tid); 701 if (remove) 702 removeThread(tid); 703 return true; 704 } 705} 706 707template <class Impl> 708void 709FullO3CPU<Impl>::suspendContext(int tid) 710{ 711 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 712 bool deallocated = deallocateContext(tid, false, 1); 713 // If this was the last thread then unschedule the tick event. 714 if ((activeThreads.size() == 1 && !deallocated) || 715 activeThreads.size() == 0) 716 unscheduleTickEvent(); 717 _status = Idle; 718} 719 720template <class Impl> 721void 722FullO3CPU<Impl>::haltContext(int tid) 723{ 724 //For now, this is the same as deallocate 725 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 726 deallocateContext(tid, true, 1); 727} 728 729template <class Impl> 730void 731FullO3CPU<Impl>::insertThread(unsigned tid) 732{ 733 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 734 // Will change now that the PC and thread state is internal to the CPU 735 // and not in the ThreadContext. 736#if FULL_SYSTEM 737 ThreadContext *src_tc = system->threadContexts[tid]; 738#else 739 ThreadContext *src_tc = tcBase(tid); 740#endif 741 742 //Bind Int Regs to Rename Map 743 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 744 PhysRegIndex phys_reg = freeList.getIntReg(); 745 746 renameMap[tid].setEntry(ireg,phys_reg); 747 scoreboard.setReg(phys_reg); 748 } 749 750 //Bind Float Regs to Rename Map 751 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 752 PhysRegIndex phys_reg = freeList.getFloatReg(); 753 754 renameMap[tid].setEntry(freg,phys_reg); 755 scoreboard.setReg(phys_reg); 756 } 757 758 //Copy Thread Data Into RegFile 759 //this->copyFromTC(tid); 760 761 //Set PC/NPC/NNPC 762 setPC(src_tc->readPC(), tid); 763 setNextPC(src_tc->readNextPC(), tid); 764 setNextNPC(src_tc->readNextNPC(), tid); 765 766 src_tc->setStatus(ThreadContext::Active); 767 768 activateContext(tid,1); 769 770 //Reset ROB/IQ/LSQ Entries 771 commit.rob->resetEntries(); 772 iew.resetEntries(); 773} 774 775template <class Impl> 776void 777FullO3CPU<Impl>::removeThread(unsigned tid) 778{ 779 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 780 781 // Copy Thread Data From RegFile 782 // If thread is suspended, it might be re-allocated 783 // this->copyToTC(tid); 784 785 786 // @todo: 2-27-2008: Fix how we free up rename mappings 787 // here to alleviate the case for double-freeing registers 788 // in SMT workloads. 789 790 // Unbind Int Regs from Rename Map 791 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 792 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 793 794 scoreboard.unsetReg(phys_reg); 795 freeList.addReg(phys_reg); 796 } 797 798 // Unbind Float Regs from Rename Map 799 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 800 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 801 802 scoreboard.unsetReg(phys_reg); 803 freeList.addReg(phys_reg); 804 } 805 806 // Squash Throughout Pipeline 807 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; 808 fetch.squash(0, sizeof(TheISA::MachInst), 0, squash_seq_num, tid); 809 decode.squash(tid); 810 rename.squash(squash_seq_num, tid); 811 iew.squash(tid); 812 iew.ldstQueue.squash(squash_seq_num, tid); 813 commit.rob->squash(squash_seq_num, tid); 814 815 816 assert(iew.instQueue.getCount(tid) == 0); 817 assert(iew.ldstQueue.getCount(tid) == 0); 818 819 // Reset ROB/IQ/LSQ Entries 820 821 // Commented out for now. This should be possible to do by 822 // telling all the pipeline stages to drain first, and then 823 // checking until the drain completes. Once the pipeline is 824 // drained, call resetEntries(). - 10-09-06 ktlim 825/* 826 if (activeThreads.size() >= 1) { 827 commit.rob->resetEntries(); 828 iew.resetEntries(); 829 } 830*/ 831} 832 833 834template <class Impl> 835void 836FullO3CPU<Impl>::activateWhenReady(int tid) 837{ 838 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 839 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 840 tid); 841 842 bool ready = true; 843 844 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 845 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 846 "Phys. Int. Regs.\n", 847 tid); 848 ready = false; 849 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 850 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 851 "Phys. Float. Regs.\n", 852 tid); 853 ready = false; 854 } else if (commit.rob->numFreeEntries() >= 855 commit.rob->entryAmount(activeThreads.size() + 1)) { 856 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 857 "ROB entries.\n", 858 tid); 859 ready = false; 860 } else if (iew.instQueue.numFreeEntries() >= 861 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 862 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 863 "IQ entries.\n", 864 tid); 865 ready = false; 866 } else if (iew.ldstQueue.numFreeEntries() >= 867 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 868 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 869 "LSQ entries.\n", 870 tid); 871 ready = false; 872 } 873 874 if (ready) { 875 insertThread(tid); 876 877 contextSwitch = false; 878 879 cpuWaitList.remove(tid); 880 } else { 881 suspendContext(tid); 882 883 //blocks fetch 884 contextSwitch = true; 885 886 //@todo: dont always add to waitlist 887 //do waitlist 888 cpuWaitList.push_back(tid); 889 } 890} 891 892#if FULL_SYSTEM 893template <class Impl> 894void 895FullO3CPU<Impl>::postInterrupt(int int_num, int index) 896{ 897 BaseCPU::postInterrupt(int_num, index); 898 899 if (this->thread[0]->status() == ThreadContext::Suspended) { 900 DPRINTF(IPI,"Suspended Processor awoke\n"); 901 this->threadContexts[0]->activate(); 902 } 903} 904 905template <class Impl> 906Fault 907FullO3CPU<Impl>::hwrei(unsigned tid) 908{ 909#if THE_ISA == ALPHA_ISA 910 // Need to clear the lock flag upon returning from an interrupt. 911 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 912 913 this->thread[tid]->kernelStats->hwrei(); 914 915 // FIXME: XXX check for interrupts? XXX 916#endif 917 return NoFault; 918} 919 920template <class Impl> 921bool 922FullO3CPU<Impl>::simPalCheck(int palFunc, unsigned tid) 923{ 924#if THE_ISA == ALPHA_ISA 925 if (this->thread[tid]->kernelStats) 926 this->thread[tid]->kernelStats->callpal(palFunc, 927 this->threadContexts[tid]); 928 929 switch (palFunc) { 930 case PAL::halt: 931 halt(); 932 if (--System::numSystemsRunning == 0) 933 exitSimLoop("all cpus halted"); 934 break; 935 936 case PAL::bpt: 937 case PAL::bugchk: 938 if (this->system->breakpoint()) 939 return false; 940 break; 941 } 942#endif 943 return true; 944} 945 946template <class Impl> 947Fault 948FullO3CPU<Impl>::getInterrupts() 949{ 950 // Check if there are any outstanding interrupts 951 return this->interrupts->getInterrupt(this->threadContexts[0]); 952} 953 954template <class Impl> 955void 956FullO3CPU<Impl>::processInterrupts(Fault interrupt) 957{ 958 // Check for interrupts here. For now can copy the code that 959 // exists within isa_fullsys_traits.hh. Also assume that thread 0 960 // is the one that handles the interrupts. 961 // @todo: Possibly consolidate the interrupt checking code. 962 // @todo: Allow other threads to handle interrupts. 963 964 assert(interrupt != NoFault); 965 this->interrupts->updateIntrInfo(this->threadContexts[0]); 966 967 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 968 this->trap(interrupt, 0); 969} 970 971template <class Impl> 972void 973FullO3CPU<Impl>::updateMemPorts() 974{ 975 // Update all ThreadContext's memory ports (Functional/Virtual 976 // Ports) 977 for (int i = 0; i < thread.size(); ++i) 978 thread[i]->connectMemPorts(thread[i]->getTC()); 979} 980#endif 981 982template <class Impl> 983void 984FullO3CPU<Impl>::trap(Fault fault, unsigned tid) 985{ 986 // Pass the thread's TC into the invoke method. 987 fault->invoke(this->threadContexts[tid]); 988} 989 990#if !FULL_SYSTEM 991 992template <class Impl> 993void 994FullO3CPU<Impl>::syscall(int64_t callnum, int tid) 995{ 996 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 997 998 DPRINTF(Activity,"Activity: syscall() called.\n"); 999 1000 // Temporarily increase this by one to account for the syscall 1001 // instruction. 1002 ++(this->thread[tid]->funcExeInst); 1003 1004 // Execute the actual syscall. 1005 this->thread[tid]->syscall(callnum); 1006 1007 // Decrease funcExeInst by one as the normal commit will handle 1008 // incrementing it. 1009 --(this->thread[tid]->funcExeInst); 1010} 1011 1012template <class Impl> 1013void 1014FullO3CPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) 1015{ 1016 TheISA::setSyscallReturn(return_value, this->tcBase(tid)); 1017} 1018 1019#endif 1020 1021template <class Impl> 1022void 1023FullO3CPU<Impl>::serialize(std::ostream &os) 1024{ 1025 SimObject::State so_state = SimObject::getState(); 1026 SERIALIZE_ENUM(so_state); 1027 BaseCPU::serialize(os); 1028 nameOut(os, csprintf("%s.tickEvent", name())); 1029 tickEvent.serialize(os); 1030 1031 // Use SimpleThread's ability to checkpoint to make it easier to 1032 // write out the registers. Also make this static so it doesn't 1033 // get instantiated multiple times (causes a panic in statistics). 1034 static SimpleThread temp; 1035 1036 for (int i = 0; i < thread.size(); i++) { 1037 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1038 temp.copyTC(thread[i]->getTC()); 1039 temp.serialize(os); 1040 } 1041} 1042 1043template <class Impl> 1044void 1045FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1046{ 1047 SimObject::State so_state; 1048 UNSERIALIZE_ENUM(so_state); 1049 BaseCPU::unserialize(cp, section); 1050 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1051 1052 // Use SimpleThread's ability to checkpoint to make it easier to 1053 // read in the registers. Also make this static so it doesn't 1054 // get instantiated multiple times (causes a panic in statistics). 1055 static SimpleThread temp; 1056 1057 for (int i = 0; i < thread.size(); i++) { 1058 temp.copyTC(thread[i]->getTC()); 1059 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1060 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1061 } 1062} 1063 1064template <class Impl> 1065unsigned int 1066FullO3CPU<Impl>::drain(Event *drain_event) 1067{ 1068 DPRINTF(O3CPU, "Switching out\n"); 1069 1070 // If the CPU isn't doing anything, then return immediately. 1071 if (_status == Idle || _status == SwitchedOut) { 1072 return 0; 1073 } 1074 1075 drainCount = 0; 1076 fetch.drain(); 1077 decode.drain(); 1078 rename.drain(); 1079 iew.drain(); 1080 commit.drain(); 1081 1082 // Wake the CPU and record activity so everything can drain out if 1083 // the CPU was not able to immediately drain. 1084 if (getState() != SimObject::Drained) { 1085 // A bit of a hack...set the drainEvent after all the drain() 1086 // calls have been made, that way if all of the stages drain 1087 // immediately, the signalDrained() function knows not to call 1088 // process on the drain event. 1089 drainEvent = drain_event; 1090 1091 wakeCPU(); 1092 activityRec.activity(); 1093 1094 return 1; 1095 } else { 1096 return 0; 1097 } 1098} 1099 1100template <class Impl> 1101void 1102FullO3CPU<Impl>::resume() 1103{ 1104 fetch.resume(); 1105 decode.resume(); 1106 rename.resume(); 1107 iew.resume(); 1108 commit.resume(); 1109 1110 changeState(SimObject::Running); 1111 1112 if (_status == SwitchedOut || _status == Idle) 1113 return; 1114 1115#if FULL_SYSTEM 1116 assert(system->getMemoryMode() == Enums::timing); 1117#endif 1118 1119 if (!tickEvent.scheduled()) 1120 schedule(tickEvent, nextCycle()); 1121 _status = Running; 1122} 1123 1124template <class Impl> 1125void 1126FullO3CPU<Impl>::signalDrained() 1127{ 1128 if (++drainCount == NumStages) { 1129 if (tickEvent.scheduled()) 1130 tickEvent.squash(); 1131 1132 changeState(SimObject::Drained); 1133 1134 BaseCPU::switchOut(); 1135 1136 if (drainEvent) { 1137 drainEvent->process(); 1138 drainEvent = NULL; 1139 } 1140 } 1141 assert(drainCount <= 5); 1142} 1143 1144template <class Impl> 1145void 1146FullO3CPU<Impl>::switchOut() 1147{ 1148 fetch.switchOut(); 1149 rename.switchOut(); 1150 iew.switchOut(); 1151 commit.switchOut(); 1152 instList.clear(); 1153 while (!removeList.empty()) { 1154 removeList.pop(); 1155 } 1156 1157 _status = SwitchedOut; 1158#if USE_CHECKER 1159 if (checker) 1160 checker->switchOut(); 1161#endif 1162 if (tickEvent.scheduled()) 1163 tickEvent.squash(); 1164} 1165 1166template <class Impl> 1167void 1168FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1169{ 1170 // Flush out any old data from the time buffers. 1171 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1172 timeBuffer.advance(); 1173 fetchQueue.advance(); 1174 decodeQueue.advance(); 1175 renameQueue.advance(); 1176 iewQueue.advance(); 1177 } 1178 1179 activityRec.reset(); 1180 1181 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort()); 1182 1183 fetch.takeOverFrom(); 1184 decode.takeOverFrom(); 1185 rename.takeOverFrom(); 1186 iew.takeOverFrom(); 1187 commit.takeOverFrom(); 1188 1189 assert(!tickEvent.scheduled()); 1190 1191 // @todo: Figure out how to properly select the tid to put onto 1192 // the active threads list. 1193 int tid = 0; 1194 1195 std::list<unsigned>::iterator isActive = 1196 std::find(activeThreads.begin(), activeThreads.end(), tid); 1197 1198 if (isActive == activeThreads.end()) { 1199 //May Need to Re-code this if the delay variable is the delay 1200 //needed for thread to activate 1201 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1202 tid); 1203 1204 activeThreads.push_back(tid); 1205 } 1206 1207 // Set all statuses to active, schedule the CPU's tick event. 1208 // @todo: Fix up statuses so this is handled properly 1209 for (int i = 0; i < threadContexts.size(); ++i) { 1210 ThreadContext *tc = threadContexts[i]; 1211 if (tc->status() == ThreadContext::Active && _status != Running) { 1212 _status = Running; 1213 schedule(tickEvent, nextCycle()); 1214 } 1215 } 1216 if (!tickEvent.scheduled()) 1217 schedule(tickEvent, nextCycle()); 1218} 1219 1220template <class Impl> 1221TheISA::MiscReg 1222FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, unsigned tid) 1223{ 1224 return this->regFile.readMiscRegNoEffect(misc_reg, tid); 1225} 1226 1227template <class Impl> 1228TheISA::MiscReg 1229FullO3CPU<Impl>::readMiscReg(int misc_reg, unsigned tid) 1230{ 1231 return this->regFile.readMiscReg(misc_reg, tid); 1232} 1233 1234template <class Impl> 1235void 1236FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1237 const TheISA::MiscReg &val, unsigned tid) 1238{ 1239 this->regFile.setMiscRegNoEffect(misc_reg, val, tid); 1240} 1241 1242template <class Impl> 1243void 1244FullO3CPU<Impl>::setMiscReg(int misc_reg, 1245 const TheISA::MiscReg &val, unsigned tid) 1246{ 1247 this->regFile.setMiscReg(misc_reg, val, tid); 1248} 1249 1250template <class Impl> 1251uint64_t 1252FullO3CPU<Impl>::readIntReg(int reg_idx) 1253{ 1254 return regFile.readIntReg(reg_idx); 1255} 1256 1257template <class Impl> 1258FloatReg 1259FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 1260{ 1261 return regFile.readFloatReg(reg_idx, width); 1262} 1263 1264template <class Impl> 1265FloatReg 1266FullO3CPU<Impl>::readFloatReg(int reg_idx) 1267{ 1268 return regFile.readFloatReg(reg_idx); 1269} 1270 1271template <class Impl> 1272FloatRegBits 1273FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1274{ 1275 return regFile.readFloatRegBits(reg_idx, width); 1276} 1277 1278template <class Impl> 1279FloatRegBits 1280FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1281{ 1282 return regFile.readFloatRegBits(reg_idx); 1283} 1284 1285template <class Impl> 1286void 1287FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1288{ 1289 regFile.setIntReg(reg_idx, val); 1290} 1291 1292template <class Impl> 1293void 1294FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1295{ 1296 regFile.setFloatReg(reg_idx, val, width); 1297} 1298 1299template <class Impl> 1300void 1301FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1302{ 1303 regFile.setFloatReg(reg_idx, val); 1304} 1305 1306template <class Impl> 1307void 1308FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1309{ 1310 regFile.setFloatRegBits(reg_idx, val, width); 1311} 1312 1313template <class Impl> 1314void 1315FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1316{ 1317 regFile.setFloatRegBits(reg_idx, val); 1318} 1319 1320template <class Impl> 1321uint64_t 1322FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1323{ 1324 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1325 1326 return regFile.readIntReg(phys_reg); 1327} 1328 1329template <class Impl> 1330float 1331FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 1332{ 1333 int idx = reg_idx + TheISA::FP_Base_DepTag; 1334 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1335 1336 return regFile.readFloatReg(phys_reg); 1337} 1338 1339template <class Impl> 1340double 1341FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 1342{ 1343 int idx = reg_idx + TheISA::FP_Base_DepTag; 1344 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1345 1346 return regFile.readFloatReg(phys_reg, 64); 1347} 1348 1349template <class Impl> 1350uint64_t 1351FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1352{ 1353 int idx = reg_idx + TheISA::FP_Base_DepTag; 1354 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1355 1356 return regFile.readFloatRegBits(phys_reg); 1357} 1358 1359template <class Impl> 1360void 1361FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1362{ 1363 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1364 1365 regFile.setIntReg(phys_reg, val); 1366} 1367 1368template <class Impl> 1369void 1370FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 1371{ 1372 int idx = reg_idx + TheISA::FP_Base_DepTag; 1373 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1374 1375 regFile.setFloatReg(phys_reg, val); 1376} 1377 1378template <class Impl> 1379void 1380FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1381{ 1382 int idx = reg_idx + TheISA::FP_Base_DepTag; 1383 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1384 1385 regFile.setFloatReg(phys_reg, val, 64); 1386} 1387 1388template <class Impl> 1389void 1390FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1391{ 1392 int idx = reg_idx + TheISA::FP_Base_DepTag; 1393 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1394 1395 regFile.setFloatRegBits(phys_reg, val); 1396} 1397 1398template <class Impl> 1399uint64_t 1400FullO3CPU<Impl>::readPC(unsigned tid) 1401{ 1402 return commit.readPC(tid); 1403} 1404 1405template <class Impl> 1406void 1407FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1408{ 1409 commit.setPC(new_PC, tid); 1410} 1411 1412template <class Impl> 1413uint64_t 1414FullO3CPU<Impl>::readMicroPC(unsigned tid) 1415{ 1416 return commit.readMicroPC(tid); 1417} 1418 1419template <class Impl> 1420void 1421FullO3CPU<Impl>::setMicroPC(Addr new_PC,unsigned tid) 1422{ 1423 commit.setMicroPC(new_PC, tid); 1424} 1425 1426template <class Impl> 1427uint64_t 1428FullO3CPU<Impl>::readNextPC(unsigned tid) 1429{ 1430 return commit.readNextPC(tid); 1431} 1432 1433template <class Impl> 1434void 1435FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1436{ 1437 commit.setNextPC(val, tid); 1438} 1439 1440template <class Impl> 1441uint64_t 1442FullO3CPU<Impl>::readNextNPC(unsigned tid) 1443{ 1444 return commit.readNextNPC(tid); 1445} 1446 1447template <class Impl> 1448void 1449FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1450{ 1451 commit.setNextNPC(val, tid); 1452} 1453 1454template <class Impl> 1455uint64_t 1456FullO3CPU<Impl>::readNextMicroPC(unsigned tid) 1457{ 1458 return commit.readNextMicroPC(tid); 1459} 1460 1461template <class Impl> 1462void 1463FullO3CPU<Impl>::setNextMicroPC(Addr new_PC,unsigned tid) 1464{ 1465 commit.setNextMicroPC(new_PC, tid); 1466} 1467 1468template <class Impl> 1469void 1470FullO3CPU<Impl>::squashFromTC(unsigned tid) 1471{ 1472 this->thread[tid]->inSyscall = true; 1473 this->commit.generateTCEvent(tid); 1474} 1475 1476template <class Impl> 1477typename FullO3CPU<Impl>::ListIt 1478FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1479{ 1480 instList.push_back(inst); 1481 1482 return --(instList.end()); 1483} 1484 1485template <class Impl> 1486void 1487FullO3CPU<Impl>::instDone(unsigned tid) 1488{ 1489 // Keep an instruction count. 1490 thread[tid]->numInst++; 1491 thread[tid]->numInsts++; 1492 committedInsts[tid]++; 1493 totalCommittedInsts++; 1494 1495 // Check for instruction-count-based events. 1496 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1497} 1498 1499template <class Impl> 1500void 1501FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1502{ 1503 removeInstsThisCycle = true; 1504 1505 removeList.push(inst->getInstListIt()); 1506} 1507 1508template <class Impl> 1509void 1510FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1511{ 1512 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1513 "[sn:%lli]\n", 1514 inst->threadNumber, inst->readPC(), inst->seqNum); 1515 1516 removeInstsThisCycle = true; 1517 1518 // Remove the front instruction. 1519 removeList.push(inst->getInstListIt()); 1520} 1521 1522template <class Impl> 1523void 1524FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 1525{ 1526 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1527 " list.\n", tid); 1528 1529 ListIt end_it; 1530 1531 bool rob_empty = false; 1532 1533 if (instList.empty()) { 1534 return; 1535 } else if (rob.isEmpty(/*tid*/)) { 1536 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1537 end_it = instList.begin(); 1538 rob_empty = true; 1539 } else { 1540 end_it = (rob.readTailInst(tid))->getInstListIt(); 1541 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1542 } 1543 1544 removeInstsThisCycle = true; 1545 1546 ListIt inst_it = instList.end(); 1547 1548 inst_it--; 1549 1550 // Walk through the instruction list, removing any instructions 1551 // that were inserted after the given instruction iterator, end_it. 1552 while (inst_it != end_it) { 1553 assert(!instList.empty()); 1554 1555 squashInstIt(inst_it, tid); 1556 1557 inst_it--; 1558 } 1559 1560 // If the ROB was empty, then we actually need to remove the first 1561 // instruction as well. 1562 if (rob_empty) { 1563 squashInstIt(inst_it, tid); 1564 } 1565} 1566 1567template <class Impl> 1568void 1569FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1570 unsigned tid) 1571{ 1572 assert(!instList.empty()); 1573 1574 removeInstsThisCycle = true; 1575 1576 ListIt inst_iter = instList.end(); 1577 1578 inst_iter--; 1579 1580 DPRINTF(O3CPU, "Deleting instructions from instruction " 1581 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1582 tid, seq_num, (*inst_iter)->seqNum); 1583 1584 while ((*inst_iter)->seqNum > seq_num) { 1585 1586 bool break_loop = (inst_iter == instList.begin()); 1587 1588 squashInstIt(inst_iter, tid); 1589 1590 inst_iter--; 1591 1592 if (break_loop) 1593 break; 1594 } 1595} 1596 1597template <class Impl> 1598inline void 1599FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1600{ 1601 if ((*instIt)->threadNumber == tid) { 1602 DPRINTF(O3CPU, "Squashing instruction, " 1603 "[tid:%i] [sn:%lli] PC %#x\n", 1604 (*instIt)->threadNumber, 1605 (*instIt)->seqNum, 1606 (*instIt)->readPC()); 1607 1608 // Mark it as squashed. 1609 (*instIt)->setSquashed(); 1610 1611 // @todo: Formulate a consistent method for deleting 1612 // instructions from the instruction list 1613 // Remove the instruction from the list. 1614 removeList.push(instIt); 1615 } 1616} 1617 1618template <class Impl> 1619void 1620FullO3CPU<Impl>::cleanUpRemovedInsts() 1621{ 1622 while (!removeList.empty()) { 1623 DPRINTF(O3CPU, "Removing instruction, " 1624 "[tid:%i] [sn:%lli] PC %#x\n", 1625 (*removeList.front())->threadNumber, 1626 (*removeList.front())->seqNum, 1627 (*removeList.front())->readPC()); 1628 1629 instList.erase(removeList.front()); 1630 1631 removeList.pop(); 1632 } 1633 1634 removeInstsThisCycle = false; 1635} 1636/* 1637template <class Impl> 1638void 1639FullO3CPU<Impl>::removeAllInsts() 1640{ 1641 instList.clear(); 1642} 1643*/ 1644template <class Impl> 1645void 1646FullO3CPU<Impl>::dumpInsts() 1647{ 1648 int num = 0; 1649 1650 ListIt inst_list_it = instList.begin(); 1651 1652 cprintf("Dumping Instruction List\n"); 1653 1654 while (inst_list_it != instList.end()) { 1655 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1656 "Squashed:%i\n\n", 1657 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1658 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1659 (*inst_list_it)->isSquashed()); 1660 inst_list_it++; 1661 ++num; 1662 } 1663} 1664/* 1665template <class Impl> 1666void 1667FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1668{ 1669 iew.wakeDependents(inst); 1670} 1671*/ 1672template <class Impl> 1673void 1674FullO3CPU<Impl>::wakeCPU() 1675{ 1676 if (activityRec.active() || tickEvent.scheduled()) { 1677 DPRINTF(Activity, "CPU already running.\n"); 1678 return; 1679 } 1680 1681 DPRINTF(Activity, "Waking up CPU\n"); 1682 1683 idleCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1684 numCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1685 1686 schedule(tickEvent, nextCycle()); 1687} 1688 1689template <class Impl> 1690int 1691FullO3CPU<Impl>::getFreeTid() 1692{ 1693 for (int i=0; i < numThreads; i++) { 1694 if (!tids[i]) { 1695 tids[i] = true; 1696 return i; 1697 } 1698 } 1699 1700 return -1; 1701} 1702 1703template <class Impl> 1704void 1705FullO3CPU<Impl>::doContextSwitch() 1706{ 1707 if (contextSwitch) { 1708 1709 //ADD CODE TO DEACTIVE THREAD HERE (???) 1710 1711 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1712 activateWhenReady(tid); 1713 } 1714 1715 if (cpuWaitList.size() == 0) 1716 contextSwitch = true; 1717 } 1718} 1719 1720template <class Impl> 1721void 1722FullO3CPU<Impl>::updateThreadPriority() 1723{ 1724 if (activeThreads.size() > 1) 1725 { 1726 //DEFAULT TO ROUND ROBIN SCHEME 1727 //e.g. Move highest priority to end of thread list 1728 std::list<unsigned>::iterator list_begin = activeThreads.begin(); 1729 std::list<unsigned>::iterator list_end = activeThreads.end(); 1730 1731 unsigned high_thread = *list_begin; 1732 1733 activeThreads.erase(list_begin); 1734 1735 activeThreads.push_back(high_thread); 1736 } 1737} 1738 1739// Forward declaration of FullO3CPU. 1740template class FullO3CPU<O3CPUImpl>; 1741