cpu.cc revision 5639:67cc7f0427e7
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), cpu_id(0) 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 this->thread[i]->setCpuId(params->cpu_id); 408 409 // Add the TC to the CPU's list of TC's. 410 this->threadContexts.push_back(tc); 411 } 412 413 for (int i=0; i < this->numThreads; i++) { 414 this->thread[i]->setFuncExeInst(0); 415 } 416 417 lockAddr = 0; 418 lockFlag = false; 419} 420 421#if !FULL_SYSTEM 422 423template <class Impl> 424TheISA::IntReg 425FullO3CPU<Impl>::getSyscallArg(int i, int tid) 426{ 427 assert(i < TheISA::NumArgumentRegs); 428 TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), 429 TheISA::ArgumentReg[i]); 430 TheISA::IntReg val = this->readArchIntReg(idx, tid); 431#if THE_ISA == SPARC_ISA 432 if (bits(this->readMiscRegNoEffect(SparcISA::MISCREG_PSTATE, tid), 3, 3)) 433 val = bits(val, 31, 0); 434#endif 435 return val; 436} 437 438template <class Impl> 439void 440FullO3CPU<Impl>::setSyscallArg(int i, TheISA::IntReg val, int tid) 441{ 442 assert(i < TheISA::NumArgumentRegs); 443 TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), 444 TheISA::ArgumentReg[i]); 445 this->setArchIntReg(idx, val, tid); 446} 447#endif 448 449template <class Impl> 450FullO3CPU<Impl>::~FullO3CPU() 451{ 452} 453 454template <class Impl> 455void 456FullO3CPU<Impl>::regStats() 457{ 458 BaseO3CPU::regStats(); 459 460 // Register any of the O3CPU's stats here. 461 timesIdled 462 .name(name() + ".timesIdled") 463 .desc("Number of times that the entire CPU went into an idle state and" 464 " unscheduled itself") 465 .prereq(timesIdled); 466 467 idleCycles 468 .name(name() + ".idleCycles") 469 .desc("Total number of cycles that the CPU has spent unscheduled due " 470 "to idling") 471 .prereq(idleCycles); 472 473 // Number of Instructions simulated 474 // -------------------------------- 475 // Should probably be in Base CPU but need templated 476 // MaxThreads so put in here instead 477 committedInsts 478 .init(numThreads) 479 .name(name() + ".committedInsts") 480 .desc("Number of Instructions Simulated"); 481 482 totalCommittedInsts 483 .name(name() + ".committedInsts_total") 484 .desc("Number of Instructions Simulated"); 485 486 cpi 487 .name(name() + ".cpi") 488 .desc("CPI: Cycles Per Instruction") 489 .precision(6); 490 cpi = numCycles / committedInsts; 491 492 totalCpi 493 .name(name() + ".cpi_total") 494 .desc("CPI: Total CPI of All Threads") 495 .precision(6); 496 totalCpi = numCycles / totalCommittedInsts; 497 498 ipc 499 .name(name() + ".ipc") 500 .desc("IPC: Instructions Per Cycle") 501 .precision(6); 502 ipc = committedInsts / numCycles; 503 504 totalIpc 505 .name(name() + ".ipc_total") 506 .desc("IPC: Total IPC of All Threads") 507 .precision(6); 508 totalIpc = totalCommittedInsts / numCycles; 509 510 this->fetch.regStats(); 511 this->decode.regStats(); 512 this->rename.regStats(); 513 this->iew.regStats(); 514 this->commit.regStats(); 515} 516 517template <class Impl> 518Port * 519FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 520{ 521 if (if_name == "dcache_port") 522 return iew.getDcachePort(); 523 else if (if_name == "icache_port") 524 return fetch.getIcachePort(); 525 else 526 panic("No Such Port\n"); 527} 528 529template <class Impl> 530void 531FullO3CPU<Impl>::tick() 532{ 533 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 534 535 ++numCycles; 536 537// activity = false; 538 539 //Tick each of the stages 540 fetch.tick(); 541 542 decode.tick(); 543 544 rename.tick(); 545 546 iew.tick(); 547 548 commit.tick(); 549 550#if !FULL_SYSTEM 551 doContextSwitch(); 552#endif 553 554 // Now advance the time buffers 555 timeBuffer.advance(); 556 557 fetchQueue.advance(); 558 decodeQueue.advance(); 559 renameQueue.advance(); 560 iewQueue.advance(); 561 562 activityRec.advance(); 563 564 if (removeInstsThisCycle) { 565 cleanUpRemovedInsts(); 566 } 567 568 if (!tickEvent.scheduled()) { 569 if (_status == SwitchedOut || 570 getState() == SimObject::Drained) { 571 DPRINTF(O3CPU, "Switched out!\n"); 572 // increment stat 573 lastRunningCycle = curTick; 574 } else if (!activityRec.active() || _status == Idle) { 575 DPRINTF(O3CPU, "Idle!\n"); 576 lastRunningCycle = curTick; 577 timesIdled++; 578 } else { 579 schedule(tickEvent, nextCycle(curTick + ticks(1))); 580 DPRINTF(O3CPU, "Scheduling next tick!\n"); 581 } 582 } 583 584#if !FULL_SYSTEM 585 updateThreadPriority(); 586#endif 587} 588 589template <class Impl> 590void 591FullO3CPU<Impl>::init() 592{ 593 if (!deferRegistration) { 594 registerThreadContexts(); 595 } 596 597 // Set inSyscall so that the CPU doesn't squash when initially 598 // setting up registers. 599 for (int i = 0; i < number_of_threads; ++i) 600 thread[i]->inSyscall = true; 601 602 for (int tid=0; tid < number_of_threads; tid++) { 603#if FULL_SYSTEM 604 ThreadContext *src_tc = threadContexts[tid]; 605#else 606 ThreadContext *src_tc = thread[tid]->getTC(); 607#endif 608 // Threads start in the Suspended State 609 if (src_tc->status() != ThreadContext::Suspended) { 610 continue; 611 } 612 613#if FULL_SYSTEM 614 TheISA::initCPU(src_tc, src_tc->readCpuId()); 615#endif 616 } 617 618 // Clear inSyscall. 619 for (int i = 0; i < number_of_threads; ++i) 620 thread[i]->inSyscall = false; 621 622 // Initialize stages. 623 fetch.initStage(); 624 iew.initStage(); 625 rename.initStage(); 626 commit.initStage(); 627 628 commit.setThreads(thread); 629} 630 631template <class Impl> 632void 633FullO3CPU<Impl>::activateThread(unsigned tid) 634{ 635 std::list<unsigned>::iterator isActive = 636 std::find(activeThreads.begin(), activeThreads.end(), tid); 637 638 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 639 640 if (isActive == activeThreads.end()) { 641 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 642 tid); 643 644 activeThreads.push_back(tid); 645 } 646} 647 648template <class Impl> 649void 650FullO3CPU<Impl>::deactivateThread(unsigned tid) 651{ 652 //Remove From Active List, if Active 653 std::list<unsigned>::iterator thread_it = 654 std::find(activeThreads.begin(), activeThreads.end(), tid); 655 656 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 657 658 if (thread_it != activeThreads.end()) { 659 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 660 tid); 661 activeThreads.erase(thread_it); 662 } 663} 664 665template <class Impl> 666void 667FullO3CPU<Impl>::activateContext(int tid, int delay) 668{ 669 // Needs to set each stage to running as well. 670 if (delay){ 671 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 672 "on cycle %d\n", tid, curTick + ticks(delay)); 673 scheduleActivateThreadEvent(tid, delay); 674 } else { 675 activateThread(tid); 676 } 677 678 if (lastActivatedCycle < curTick) { 679 scheduleTickEvent(delay); 680 681 // Be sure to signal that there's some activity so the CPU doesn't 682 // deschedule itself. 683 activityRec.activity(); 684 fetch.wakeFromQuiesce(); 685 686 lastActivatedCycle = curTick; 687 688 _status = Running; 689 } 690} 691 692template <class Impl> 693bool 694FullO3CPU<Impl>::deallocateContext(int tid, bool remove, int delay) 695{ 696 // Schedule removal of thread data from CPU 697 if (delay){ 698 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 699 "on cycle %d\n", tid, curTick + ticks(delay)); 700 scheduleDeallocateContextEvent(tid, remove, delay); 701 return false; 702 } else { 703 deactivateThread(tid); 704 if (remove) 705 removeThread(tid); 706 return true; 707 } 708} 709 710template <class Impl> 711void 712FullO3CPU<Impl>::suspendContext(int tid) 713{ 714 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 715 bool deallocated = deallocateContext(tid, false, 1); 716 // If this was the last thread then unschedule the tick event. 717 if ((activeThreads.size() == 1 && !deallocated) || 718 activeThreads.size() == 0) 719 unscheduleTickEvent(); 720 _status = Idle; 721} 722 723template <class Impl> 724void 725FullO3CPU<Impl>::haltContext(int tid) 726{ 727 //For now, this is the same as deallocate 728 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 729 deallocateContext(tid, true, 1); 730} 731 732template <class Impl> 733void 734FullO3CPU<Impl>::insertThread(unsigned tid) 735{ 736 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 737 // Will change now that the PC and thread state is internal to the CPU 738 // and not in the ThreadContext. 739#if FULL_SYSTEM 740 ThreadContext *src_tc = system->threadContexts[tid]; 741#else 742 ThreadContext *src_tc = tcBase(tid); 743#endif 744 745 //Bind Int Regs to Rename Map 746 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 747 PhysRegIndex phys_reg = freeList.getIntReg(); 748 749 renameMap[tid].setEntry(ireg,phys_reg); 750 scoreboard.setReg(phys_reg); 751 } 752 753 //Bind Float Regs to Rename Map 754 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 755 PhysRegIndex phys_reg = freeList.getFloatReg(); 756 757 renameMap[tid].setEntry(freg,phys_reg); 758 scoreboard.setReg(phys_reg); 759 } 760 761 //Copy Thread Data Into RegFile 762 //this->copyFromTC(tid); 763 764 //Set PC/NPC/NNPC 765 setPC(src_tc->readPC(), tid); 766 setNextPC(src_tc->readNextPC(), tid); 767 setNextNPC(src_tc->readNextNPC(), tid); 768 769 src_tc->setStatus(ThreadContext::Active); 770 771 activateContext(tid,1); 772 773 //Reset ROB/IQ/LSQ Entries 774 commit.rob->resetEntries(); 775 iew.resetEntries(); 776} 777 778template <class Impl> 779void 780FullO3CPU<Impl>::removeThread(unsigned tid) 781{ 782 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 783 784 // Copy Thread Data From RegFile 785 // If thread is suspended, it might be re-allocated 786 // this->copyToTC(tid); 787 788 789 // @todo: 2-27-2008: Fix how we free up rename mappings 790 // here to alleviate the case for double-freeing registers 791 // in SMT workloads. 792 793 // Unbind Int Regs from Rename Map 794 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 795 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 796 797 scoreboard.unsetReg(phys_reg); 798 freeList.addReg(phys_reg); 799 } 800 801 // Unbind Float Regs from Rename Map 802 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 803 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 804 805 scoreboard.unsetReg(phys_reg); 806 freeList.addReg(phys_reg); 807 } 808 809 // Squash Throughout Pipeline 810 InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; 811 fetch.squash(0, sizeof(TheISA::MachInst), 0, squash_seq_num, tid); 812 decode.squash(tid); 813 rename.squash(squash_seq_num, tid); 814 iew.squash(tid); 815 iew.ldstQueue.squash(squash_seq_num, tid); 816 commit.rob->squash(squash_seq_num, tid); 817 818 819 assert(iew.instQueue.getCount(tid) == 0); 820 assert(iew.ldstQueue.getCount(tid) == 0); 821 822 // Reset ROB/IQ/LSQ Entries 823 824 // Commented out for now. This should be possible to do by 825 // telling all the pipeline stages to drain first, and then 826 // checking until the drain completes. Once the pipeline is 827 // drained, call resetEntries(). - 10-09-06 ktlim 828/* 829 if (activeThreads.size() >= 1) { 830 commit.rob->resetEntries(); 831 iew.resetEntries(); 832 } 833*/ 834} 835 836 837template <class Impl> 838void 839FullO3CPU<Impl>::activateWhenReady(int tid) 840{ 841 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 842 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 843 tid); 844 845 bool ready = true; 846 847 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 848 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 849 "Phys. Int. Regs.\n", 850 tid); 851 ready = false; 852 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 853 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 854 "Phys. Float. Regs.\n", 855 tid); 856 ready = false; 857 } else if (commit.rob->numFreeEntries() >= 858 commit.rob->entryAmount(activeThreads.size() + 1)) { 859 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 860 "ROB entries.\n", 861 tid); 862 ready = false; 863 } else if (iew.instQueue.numFreeEntries() >= 864 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 865 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 866 "IQ entries.\n", 867 tid); 868 ready = false; 869 } else if (iew.ldstQueue.numFreeEntries() >= 870 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 871 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 872 "LSQ entries.\n", 873 tid); 874 ready = false; 875 } 876 877 if (ready) { 878 insertThread(tid); 879 880 contextSwitch = false; 881 882 cpuWaitList.remove(tid); 883 } else { 884 suspendContext(tid); 885 886 //blocks fetch 887 contextSwitch = true; 888 889 //@todo: dont always add to waitlist 890 //do waitlist 891 cpuWaitList.push_back(tid); 892 } 893} 894 895#if FULL_SYSTEM 896template <class Impl> 897void 898FullO3CPU<Impl>::post_interrupt(int int_num, int index) 899{ 900 BaseCPU::post_interrupt(int_num, index); 901 902 if (this->thread[0]->status() == ThreadContext::Suspended) { 903 DPRINTF(IPI,"Suspended Processor awoke\n"); 904 this->threadContexts[0]->activate(); 905 } 906} 907 908template <class Impl> 909bool 910FullO3CPU<Impl>::simPalCheck(int palFunc, unsigned tid) 911{ 912#if THE_ISA == ALPHA_ISA 913 if (this->thread[tid]->kernelStats) 914 this->thread[tid]->kernelStats->callpal(palFunc, 915 this->threadContexts[tid]); 916 917 switch (palFunc) { 918 case PAL::halt: 919 halt(); 920 if (--System::numSystemsRunning == 0) 921 exitSimLoop("all cpus halted"); 922 break; 923 924 case PAL::bpt: 925 case PAL::bugchk: 926 if (this->system->breakpoint()) 927 return false; 928 break; 929 } 930#endif 931 return true; 932} 933 934template <class Impl> 935Fault 936FullO3CPU<Impl>::getInterrupts() 937{ 938 // Check if there are any outstanding interrupts 939 return this->interrupts.getInterrupt(this->threadContexts[0]); 940} 941 942template <class Impl> 943void 944FullO3CPU<Impl>::processInterrupts(Fault interrupt) 945{ 946 // Check for interrupts here. For now can copy the code that 947 // exists within isa_fullsys_traits.hh. Also assume that thread 0 948 // is the one that handles the interrupts. 949 // @todo: Possibly consolidate the interrupt checking code. 950 // @todo: Allow other threads to handle interrupts. 951 952 assert(interrupt != NoFault); 953 this->interrupts.updateIntrInfo(this->threadContexts[0]); 954 955 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 956 this->trap(interrupt, 0); 957} 958 959template <class Impl> 960void 961FullO3CPU<Impl>::updateMemPorts() 962{ 963 // Update all ThreadContext's memory ports (Functional/Virtual 964 // Ports) 965 for (int i = 0; i < thread.size(); ++i) 966 thread[i]->connectMemPorts(thread[i]->getTC()); 967} 968#endif 969 970template <class Impl> 971void 972FullO3CPU<Impl>::trap(Fault fault, unsigned tid) 973{ 974 // Pass the thread's TC into the invoke method. 975 fault->invoke(this->threadContexts[tid]); 976} 977 978#if !FULL_SYSTEM 979 980template <class Impl> 981void 982FullO3CPU<Impl>::syscall(int64_t callnum, int tid) 983{ 984 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 985 986 DPRINTF(Activity,"Activity: syscall() called.\n"); 987 988 // Temporarily increase this by one to account for the syscall 989 // instruction. 990 ++(this->thread[tid]->funcExeInst); 991 992 // Execute the actual syscall. 993 this->thread[tid]->syscall(callnum); 994 995 // Decrease funcExeInst by one as the normal commit will handle 996 // incrementing it. 997 --(this->thread[tid]->funcExeInst); 998} 999 1000template <class Impl> 1001void 1002FullO3CPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) 1003{ 1004 TheISA::setSyscallReturn(return_value, this->tcBase(tid)); 1005} 1006 1007#endif 1008 1009template <class Impl> 1010void 1011FullO3CPU<Impl>::serialize(std::ostream &os) 1012{ 1013 SimObject::State so_state = SimObject::getState(); 1014 SERIALIZE_ENUM(so_state); 1015 BaseCPU::serialize(os); 1016 nameOut(os, csprintf("%s.tickEvent", name())); 1017 tickEvent.serialize(os); 1018 1019 // Use SimpleThread's ability to checkpoint to make it easier to 1020 // write out the registers. Also make this static so it doesn't 1021 // get instantiated multiple times (causes a panic in statistics). 1022 static SimpleThread temp; 1023 1024 for (int i = 0; i < thread.size(); i++) { 1025 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1026 temp.copyTC(thread[i]->getTC()); 1027 temp.serialize(os); 1028 } 1029} 1030 1031template <class Impl> 1032void 1033FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1034{ 1035 SimObject::State so_state; 1036 UNSERIALIZE_ENUM(so_state); 1037 BaseCPU::unserialize(cp, section); 1038 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1039 1040 // Use SimpleThread's ability to checkpoint to make it easier to 1041 // read in the registers. Also make this static so it doesn't 1042 // get instantiated multiple times (causes a panic in statistics). 1043 static SimpleThread temp; 1044 1045 for (int i = 0; i < thread.size(); i++) { 1046 temp.copyTC(thread[i]->getTC()); 1047 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1048 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1049 } 1050} 1051 1052template <class Impl> 1053unsigned int 1054FullO3CPU<Impl>::drain(Event *drain_event) 1055{ 1056 DPRINTF(O3CPU, "Switching out\n"); 1057 1058 // If the CPU isn't doing anything, then return immediately. 1059 if (_status == Idle || _status == SwitchedOut) { 1060 return 0; 1061 } 1062 1063 drainCount = 0; 1064 fetch.drain(); 1065 decode.drain(); 1066 rename.drain(); 1067 iew.drain(); 1068 commit.drain(); 1069 1070 // Wake the CPU and record activity so everything can drain out if 1071 // the CPU was not able to immediately drain. 1072 if (getState() != SimObject::Drained) { 1073 // A bit of a hack...set the drainEvent after all the drain() 1074 // calls have been made, that way if all of the stages drain 1075 // immediately, the signalDrained() function knows not to call 1076 // process on the drain event. 1077 drainEvent = drain_event; 1078 1079 wakeCPU(); 1080 activityRec.activity(); 1081 1082 return 1; 1083 } else { 1084 return 0; 1085 } 1086} 1087 1088template <class Impl> 1089void 1090FullO3CPU<Impl>::resume() 1091{ 1092 fetch.resume(); 1093 decode.resume(); 1094 rename.resume(); 1095 iew.resume(); 1096 commit.resume(); 1097 1098 changeState(SimObject::Running); 1099 1100 if (_status == SwitchedOut || _status == Idle) 1101 return; 1102 1103#if FULL_SYSTEM 1104 assert(system->getMemoryMode() == Enums::timing); 1105#endif 1106 1107 if (!tickEvent.scheduled()) 1108 schedule(tickEvent, nextCycle()); 1109 _status = Running; 1110} 1111 1112template <class Impl> 1113void 1114FullO3CPU<Impl>::signalDrained() 1115{ 1116 if (++drainCount == NumStages) { 1117 if (tickEvent.scheduled()) 1118 tickEvent.squash(); 1119 1120 changeState(SimObject::Drained); 1121 1122 BaseCPU::switchOut(); 1123 1124 if (drainEvent) { 1125 drainEvent->process(); 1126 drainEvent = NULL; 1127 } 1128 } 1129 assert(drainCount <= 5); 1130} 1131 1132template <class Impl> 1133void 1134FullO3CPU<Impl>::switchOut() 1135{ 1136 fetch.switchOut(); 1137 rename.switchOut(); 1138 iew.switchOut(); 1139 commit.switchOut(); 1140 instList.clear(); 1141 while (!removeList.empty()) { 1142 removeList.pop(); 1143 } 1144 1145 _status = SwitchedOut; 1146#if USE_CHECKER 1147 if (checker) 1148 checker->switchOut(); 1149#endif 1150 if (tickEvent.scheduled()) 1151 tickEvent.squash(); 1152} 1153 1154template <class Impl> 1155void 1156FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1157{ 1158 // Flush out any old data from the time buffers. 1159 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1160 timeBuffer.advance(); 1161 fetchQueue.advance(); 1162 decodeQueue.advance(); 1163 renameQueue.advance(); 1164 iewQueue.advance(); 1165 } 1166 1167 activityRec.reset(); 1168 1169 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort()); 1170 1171 fetch.takeOverFrom(); 1172 decode.takeOverFrom(); 1173 rename.takeOverFrom(); 1174 iew.takeOverFrom(); 1175 commit.takeOverFrom(); 1176 1177 assert(!tickEvent.scheduled()); 1178 1179 // @todo: Figure out how to properly select the tid to put onto 1180 // the active threads list. 1181 int tid = 0; 1182 1183 std::list<unsigned>::iterator isActive = 1184 std::find(activeThreads.begin(), activeThreads.end(), tid); 1185 1186 if (isActive == activeThreads.end()) { 1187 //May Need to Re-code this if the delay variable is the delay 1188 //needed for thread to activate 1189 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1190 tid); 1191 1192 activeThreads.push_back(tid); 1193 } 1194 1195 // Set all statuses to active, schedule the CPU's tick event. 1196 // @todo: Fix up statuses so this is handled properly 1197 for (int i = 0; i < threadContexts.size(); ++i) { 1198 ThreadContext *tc = threadContexts[i]; 1199 if (tc->status() == ThreadContext::Active && _status != Running) { 1200 _status = Running; 1201 schedule(tickEvent, nextCycle()); 1202 } 1203 } 1204 if (!tickEvent.scheduled()) 1205 schedule(tickEvent, nextCycle()); 1206} 1207 1208template <class Impl> 1209TheISA::MiscReg 1210FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, unsigned tid) 1211{ 1212 return this->regFile.readMiscRegNoEffect(misc_reg, tid); 1213} 1214 1215template <class Impl> 1216TheISA::MiscReg 1217FullO3CPU<Impl>::readMiscReg(int misc_reg, unsigned tid) 1218{ 1219 return this->regFile.readMiscReg(misc_reg, tid); 1220} 1221 1222template <class Impl> 1223void 1224FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1225 const TheISA::MiscReg &val, unsigned tid) 1226{ 1227 this->regFile.setMiscRegNoEffect(misc_reg, val, tid); 1228} 1229 1230template <class Impl> 1231void 1232FullO3CPU<Impl>::setMiscReg(int misc_reg, 1233 const TheISA::MiscReg &val, unsigned tid) 1234{ 1235 this->regFile.setMiscReg(misc_reg, val, tid); 1236} 1237 1238template <class Impl> 1239uint64_t 1240FullO3CPU<Impl>::readIntReg(int reg_idx) 1241{ 1242 return regFile.readIntReg(reg_idx); 1243} 1244 1245template <class Impl> 1246FloatReg 1247FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 1248{ 1249 return regFile.readFloatReg(reg_idx, width); 1250} 1251 1252template <class Impl> 1253FloatReg 1254FullO3CPU<Impl>::readFloatReg(int reg_idx) 1255{ 1256 return regFile.readFloatReg(reg_idx); 1257} 1258 1259template <class Impl> 1260FloatRegBits 1261FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 1262{ 1263 return regFile.readFloatRegBits(reg_idx, width); 1264} 1265 1266template <class Impl> 1267FloatRegBits 1268FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1269{ 1270 return regFile.readFloatRegBits(reg_idx); 1271} 1272 1273template <class Impl> 1274void 1275FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1276{ 1277 regFile.setIntReg(reg_idx, val); 1278} 1279 1280template <class Impl> 1281void 1282FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 1283{ 1284 regFile.setFloatReg(reg_idx, val, width); 1285} 1286 1287template <class Impl> 1288void 1289FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1290{ 1291 regFile.setFloatReg(reg_idx, val); 1292} 1293 1294template <class Impl> 1295void 1296FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 1297{ 1298 regFile.setFloatRegBits(reg_idx, val, width); 1299} 1300 1301template <class Impl> 1302void 1303FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1304{ 1305 regFile.setFloatRegBits(reg_idx, val); 1306} 1307 1308template <class Impl> 1309uint64_t 1310FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 1311{ 1312 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1313 1314 return regFile.readIntReg(phys_reg); 1315} 1316 1317template <class Impl> 1318float 1319FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 1320{ 1321 int idx = reg_idx + TheISA::FP_Base_DepTag; 1322 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1323 1324 return regFile.readFloatReg(phys_reg); 1325} 1326 1327template <class Impl> 1328double 1329FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 1330{ 1331 int idx = reg_idx + TheISA::FP_Base_DepTag; 1332 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1333 1334 return regFile.readFloatReg(phys_reg, 64); 1335} 1336 1337template <class Impl> 1338uint64_t 1339FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 1340{ 1341 int idx = reg_idx + TheISA::FP_Base_DepTag; 1342 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1343 1344 return regFile.readFloatRegBits(phys_reg); 1345} 1346 1347template <class Impl> 1348void 1349FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 1350{ 1351 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1352 1353 regFile.setIntReg(phys_reg, val); 1354} 1355 1356template <class Impl> 1357void 1358FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 1359{ 1360 int idx = reg_idx + TheISA::FP_Base_DepTag; 1361 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1362 1363 regFile.setFloatReg(phys_reg, val); 1364} 1365 1366template <class Impl> 1367void 1368FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 1369{ 1370 int idx = reg_idx + TheISA::FP_Base_DepTag; 1371 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1372 1373 regFile.setFloatReg(phys_reg, val, 64); 1374} 1375 1376template <class Impl> 1377void 1378FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 1379{ 1380 int idx = reg_idx + TheISA::FP_Base_DepTag; 1381 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1382 1383 regFile.setFloatRegBits(phys_reg, val); 1384} 1385 1386template <class Impl> 1387uint64_t 1388FullO3CPU<Impl>::readPC(unsigned tid) 1389{ 1390 return commit.readPC(tid); 1391} 1392 1393template <class Impl> 1394void 1395FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 1396{ 1397 commit.setPC(new_PC, tid); 1398} 1399 1400template <class Impl> 1401uint64_t 1402FullO3CPU<Impl>::readMicroPC(unsigned tid) 1403{ 1404 return commit.readMicroPC(tid); 1405} 1406 1407template <class Impl> 1408void 1409FullO3CPU<Impl>::setMicroPC(Addr new_PC,unsigned tid) 1410{ 1411 commit.setMicroPC(new_PC, tid); 1412} 1413 1414template <class Impl> 1415uint64_t 1416FullO3CPU<Impl>::readNextPC(unsigned tid) 1417{ 1418 return commit.readNextPC(tid); 1419} 1420 1421template <class Impl> 1422void 1423FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 1424{ 1425 commit.setNextPC(val, tid); 1426} 1427 1428template <class Impl> 1429uint64_t 1430FullO3CPU<Impl>::readNextNPC(unsigned tid) 1431{ 1432 return commit.readNextNPC(tid); 1433} 1434 1435template <class Impl> 1436void 1437FullO3CPU<Impl>::setNextNPC(uint64_t val,unsigned tid) 1438{ 1439 commit.setNextNPC(val, tid); 1440} 1441 1442template <class Impl> 1443uint64_t 1444FullO3CPU<Impl>::readNextMicroPC(unsigned tid) 1445{ 1446 return commit.readNextMicroPC(tid); 1447} 1448 1449template <class Impl> 1450void 1451FullO3CPU<Impl>::setNextMicroPC(Addr new_PC,unsigned tid) 1452{ 1453 commit.setNextMicroPC(new_PC, tid); 1454} 1455 1456template <class Impl> 1457void 1458FullO3CPU<Impl>::squashFromTC(unsigned tid) 1459{ 1460 this->thread[tid]->inSyscall = true; 1461 this->commit.generateTCEvent(tid); 1462} 1463 1464template <class Impl> 1465typename FullO3CPU<Impl>::ListIt 1466FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1467{ 1468 instList.push_back(inst); 1469 1470 return --(instList.end()); 1471} 1472 1473template <class Impl> 1474void 1475FullO3CPU<Impl>::instDone(unsigned tid) 1476{ 1477 // Keep an instruction count. 1478 thread[tid]->numInst++; 1479 thread[tid]->numInsts++; 1480 committedInsts[tid]++; 1481 totalCommittedInsts++; 1482 1483 // Check for instruction-count-based events. 1484 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1485} 1486 1487template <class Impl> 1488void 1489FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 1490{ 1491 removeInstsThisCycle = true; 1492 1493 removeList.push(inst->getInstListIt()); 1494} 1495 1496template <class Impl> 1497void 1498FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1499{ 1500 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 1501 "[sn:%lli]\n", 1502 inst->threadNumber, inst->readPC(), inst->seqNum); 1503 1504 removeInstsThisCycle = true; 1505 1506 // Remove the front instruction. 1507 removeList.push(inst->getInstListIt()); 1508} 1509 1510template <class Impl> 1511void 1512FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 1513{ 1514 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1515 " list.\n", tid); 1516 1517 ListIt end_it; 1518 1519 bool rob_empty = false; 1520 1521 if (instList.empty()) { 1522 return; 1523 } else if (rob.isEmpty(/*tid*/)) { 1524 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1525 end_it = instList.begin(); 1526 rob_empty = true; 1527 } else { 1528 end_it = (rob.readTailInst(tid))->getInstListIt(); 1529 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1530 } 1531 1532 removeInstsThisCycle = true; 1533 1534 ListIt inst_it = instList.end(); 1535 1536 inst_it--; 1537 1538 // Walk through the instruction list, removing any instructions 1539 // that were inserted after the given instruction iterator, end_it. 1540 while (inst_it != end_it) { 1541 assert(!instList.empty()); 1542 1543 squashInstIt(inst_it, tid); 1544 1545 inst_it--; 1546 } 1547 1548 // If the ROB was empty, then we actually need to remove the first 1549 // instruction as well. 1550 if (rob_empty) { 1551 squashInstIt(inst_it, tid); 1552 } 1553} 1554 1555template <class Impl> 1556void 1557FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1558 unsigned tid) 1559{ 1560 assert(!instList.empty()); 1561 1562 removeInstsThisCycle = true; 1563 1564 ListIt inst_iter = instList.end(); 1565 1566 inst_iter--; 1567 1568 DPRINTF(O3CPU, "Deleting instructions from instruction " 1569 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1570 tid, seq_num, (*inst_iter)->seqNum); 1571 1572 while ((*inst_iter)->seqNum > seq_num) { 1573 1574 bool break_loop = (inst_iter == instList.begin()); 1575 1576 squashInstIt(inst_iter, tid); 1577 1578 inst_iter--; 1579 1580 if (break_loop) 1581 break; 1582 } 1583} 1584 1585template <class Impl> 1586inline void 1587FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1588{ 1589 if ((*instIt)->threadNumber == tid) { 1590 DPRINTF(O3CPU, "Squashing instruction, " 1591 "[tid:%i] [sn:%lli] PC %#x\n", 1592 (*instIt)->threadNumber, 1593 (*instIt)->seqNum, 1594 (*instIt)->readPC()); 1595 1596 // Mark it as squashed. 1597 (*instIt)->setSquashed(); 1598 1599 // @todo: Formulate a consistent method for deleting 1600 // instructions from the instruction list 1601 // Remove the instruction from the list. 1602 removeList.push(instIt); 1603 } 1604} 1605 1606template <class Impl> 1607void 1608FullO3CPU<Impl>::cleanUpRemovedInsts() 1609{ 1610 while (!removeList.empty()) { 1611 DPRINTF(O3CPU, "Removing instruction, " 1612 "[tid:%i] [sn:%lli] PC %#x\n", 1613 (*removeList.front())->threadNumber, 1614 (*removeList.front())->seqNum, 1615 (*removeList.front())->readPC()); 1616 1617 instList.erase(removeList.front()); 1618 1619 removeList.pop(); 1620 } 1621 1622 removeInstsThisCycle = false; 1623} 1624/* 1625template <class Impl> 1626void 1627FullO3CPU<Impl>::removeAllInsts() 1628{ 1629 instList.clear(); 1630} 1631*/ 1632template <class Impl> 1633void 1634FullO3CPU<Impl>::dumpInsts() 1635{ 1636 int num = 0; 1637 1638 ListIt inst_list_it = instList.begin(); 1639 1640 cprintf("Dumping Instruction List\n"); 1641 1642 while (inst_list_it != instList.end()) { 1643 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1644 "Squashed:%i\n\n", 1645 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1646 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1647 (*inst_list_it)->isSquashed()); 1648 inst_list_it++; 1649 ++num; 1650 } 1651} 1652/* 1653template <class Impl> 1654void 1655FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1656{ 1657 iew.wakeDependents(inst); 1658} 1659*/ 1660template <class Impl> 1661void 1662FullO3CPU<Impl>::wakeCPU() 1663{ 1664 if (activityRec.active() || tickEvent.scheduled()) { 1665 DPRINTF(Activity, "CPU already running.\n"); 1666 return; 1667 } 1668 1669 DPRINTF(Activity, "Waking up CPU\n"); 1670 1671 idleCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1672 numCycles += tickToCycles((curTick - 1) - lastRunningCycle); 1673 1674 schedule(tickEvent, nextCycle()); 1675} 1676 1677template <class Impl> 1678int 1679FullO3CPU<Impl>::getFreeTid() 1680{ 1681 for (int i=0; i < numThreads; i++) { 1682 if (!tids[i]) { 1683 tids[i] = true; 1684 return i; 1685 } 1686 } 1687 1688 return -1; 1689} 1690 1691template <class Impl> 1692void 1693FullO3CPU<Impl>::doContextSwitch() 1694{ 1695 if (contextSwitch) { 1696 1697 //ADD CODE TO DEACTIVE THREAD HERE (???) 1698 1699 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1700 activateWhenReady(tid); 1701 } 1702 1703 if (cpuWaitList.size() == 0) 1704 contextSwitch = true; 1705 } 1706} 1707 1708template <class Impl> 1709void 1710FullO3CPU<Impl>::updateThreadPriority() 1711{ 1712 if (activeThreads.size() > 1) 1713 { 1714 //DEFAULT TO ROUND ROBIN SCHEME 1715 //e.g. Move highest priority to end of thread list 1716 std::list<unsigned>::iterator list_begin = activeThreads.begin(); 1717 std::list<unsigned>::iterator list_end = activeThreads.end(); 1718 1719 unsigned high_thread = *list_begin; 1720 1721 activeThreads.erase(list_begin); 1722 1723 activeThreads.push_back(high_thread); 1724 } 1725} 1726 1727// Forward declaration of FullO3CPU. 1728template class FullO3CPU<O3CPUImpl>; 1729