cpu.cc revision 12109:f29e9c5418aa
1/* 2 * Copyright (c) 2011-2012, 2014, 2016 ARM Limited 3 * Copyright (c) 2013 Advanced Micro Devices, Inc. 4 * All rights reserved 5 * 6 * The license below extends only to copyright in the software and shall 7 * not be construed as granting a license to any other intellectual 8 * property including but not limited to intellectual property relating 9 * to a hardware implementation of the functionality of the software 10 * licensed hereunder. You may use the software subject to the license 11 * terms below provided that you ensure that this notice is replicated 12 * unmodified and in its entirety in all distributions of the software, 13 * modified or unmodified, in source code or in binary form. 14 * 15 * Copyright (c) 2004-2006 The Regents of The University of Michigan 16 * Copyright (c) 2011 Regents of the University of California 17 * All rights reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions are 21 * met: redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer; 23 * redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution; 26 * neither the name of the copyright holders nor the names of its 27 * contributors may be used to endorse or promote products derived from 28 * this software without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Authors: Kevin Lim 43 * Korey Sewell 44 * Rick Strong 45 */ 46 47#include "cpu/o3/cpu.hh" 48 49#include "arch/generic/traits.hh" 50#include "arch/kernel_stats.hh" 51#include "config/the_isa.hh" 52#include "cpu/activity.hh" 53#include "cpu/checker/cpu.hh" 54#include "cpu/checker/thread_context.hh" 55#include "cpu/o3/isa_specific.hh" 56#include "cpu/o3/thread_context.hh" 57#include "cpu/quiesce_event.hh" 58#include "cpu/simple_thread.hh" 59#include "cpu/thread_context.hh" 60#include "debug/Activity.hh" 61#include "debug/Drain.hh" 62#include "debug/O3CPU.hh" 63#include "debug/Quiesce.hh" 64#include "enums/MemoryMode.hh" 65#include "sim/core.hh" 66#include "sim/full_system.hh" 67#include "sim/process.hh" 68#include "sim/stat_control.hh" 69#include "sim/system.hh" 70 71#if THE_ISA == ALPHA_ISA 72#include "arch/alpha/osfpal.hh" 73#include "debug/Activity.hh" 74 75#endif 76 77struct BaseCPUParams; 78 79using namespace TheISA; 80using namespace std; 81 82BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 83 : BaseCPU(params) 84{ 85} 86 87void 88BaseO3CPU::regStats() 89{ 90 BaseCPU::regStats(); 91} 92 93template<class Impl> 94bool 95FullO3CPU<Impl>::IcachePort::recvTimingResp(PacketPtr pkt) 96{ 97 DPRINTF(O3CPU, "Fetch unit received timing\n"); 98 // We shouldn't ever get a cacheable block in Modified state 99 assert(pkt->req->isUncacheable() || 100 !(pkt->cacheResponding() && !pkt->hasSharers())); 101 fetch->processCacheCompletion(pkt); 102 103 return true; 104} 105 106template<class Impl> 107void 108FullO3CPU<Impl>::IcachePort::recvReqRetry() 109{ 110 fetch->recvReqRetry(); 111} 112 113template <class Impl> 114bool 115FullO3CPU<Impl>::DcachePort::recvTimingResp(PacketPtr pkt) 116{ 117 return lsq->recvTimingResp(pkt); 118} 119 120template <class Impl> 121void 122FullO3CPU<Impl>::DcachePort::recvTimingSnoopReq(PacketPtr pkt) 123{ 124 for (ThreadID tid = 0; tid < cpu->numThreads; tid++) { 125 if (cpu->getCpuAddrMonitor(tid)->doMonitor(pkt)) { 126 cpu->wakeup(tid); 127 } 128 } 129 lsq->recvTimingSnoopReq(pkt); 130} 131 132template <class Impl> 133void 134FullO3CPU<Impl>::DcachePort::recvReqRetry() 135{ 136 lsq->recvReqRetry(); 137} 138 139template <class Impl> 140FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 141 : Event(CPU_Tick_Pri), cpu(c) 142{ 143} 144 145template <class Impl> 146void 147FullO3CPU<Impl>::TickEvent::process() 148{ 149 cpu->tick(); 150} 151 152template <class Impl> 153const char * 154FullO3CPU<Impl>::TickEvent::description() const 155{ 156 return "FullO3CPU tick"; 157} 158 159template <class Impl> 160FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 161 : BaseO3CPU(params), 162 itb(params->itb), 163 dtb(params->dtb), 164 tickEvent(this), 165#ifndef NDEBUG 166 instcount(0), 167#endif 168 removeInstsThisCycle(false), 169 fetch(this, params), 170 decode(this, params), 171 rename(this, params), 172 iew(this, params), 173 commit(this, params), 174 175 /* It is mandatory that all SMT threads use the same renaming mode as 176 * they are sharing registers and rename */ 177 vecMode(initRenameMode<TheISA::ISA>::mode(params->isa[0])), 178 regFile(params->numPhysIntRegs, 179 params->numPhysFloatRegs, 180 params->numPhysVecRegs, 181 params->numPhysCCRegs, 182 vecMode), 183 184 freeList(name() + ".freelist", ®File), 185 186 rob(this, params), 187 188 scoreboard(name() + ".scoreboard", 189 regFile.totalNumPhysRegs()), 190 191 isa(numThreads, NULL), 192 193 icachePort(&fetch, this), 194 dcachePort(&iew.ldstQueue, this), 195 196 timeBuffer(params->backComSize, params->forwardComSize), 197 fetchQueue(params->backComSize, params->forwardComSize), 198 decodeQueue(params->backComSize, params->forwardComSize), 199 renameQueue(params->backComSize, params->forwardComSize), 200 iewQueue(params->backComSize, params->forwardComSize), 201 activityRec(name(), NumStages, 202 params->backComSize + params->forwardComSize, 203 params->activity), 204 205 globalSeqNum(1), 206 system(params->system), 207 lastRunningCycle(curCycle()) 208{ 209 if (!params->switched_out) { 210 _status = Running; 211 } else { 212 _status = SwitchedOut; 213 } 214 215 if (params->checker) { 216 BaseCPU *temp_checker = params->checker; 217 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 218 checker->setIcachePort(&icachePort); 219 checker->setSystem(params->system); 220 } else { 221 checker = NULL; 222 } 223 224 if (!FullSystem) { 225 thread.resize(numThreads); 226 tids.resize(numThreads); 227 } 228 229 // The stages also need their CPU pointer setup. However this 230 // must be done at the upper level CPU because they have pointers 231 // to the upper level CPU, and not this FullO3CPU. 232 233 // Set up Pointers to the activeThreads list for each stage 234 fetch.setActiveThreads(&activeThreads); 235 decode.setActiveThreads(&activeThreads); 236 rename.setActiveThreads(&activeThreads); 237 iew.setActiveThreads(&activeThreads); 238 commit.setActiveThreads(&activeThreads); 239 240 // Give each of the stages the time buffer they will use. 241 fetch.setTimeBuffer(&timeBuffer); 242 decode.setTimeBuffer(&timeBuffer); 243 rename.setTimeBuffer(&timeBuffer); 244 iew.setTimeBuffer(&timeBuffer); 245 commit.setTimeBuffer(&timeBuffer); 246 247 // Also setup each of the stages' queues. 248 fetch.setFetchQueue(&fetchQueue); 249 decode.setFetchQueue(&fetchQueue); 250 commit.setFetchQueue(&fetchQueue); 251 decode.setDecodeQueue(&decodeQueue); 252 rename.setDecodeQueue(&decodeQueue); 253 rename.setRenameQueue(&renameQueue); 254 iew.setRenameQueue(&renameQueue); 255 iew.setIEWQueue(&iewQueue); 256 commit.setIEWQueue(&iewQueue); 257 commit.setRenameQueue(&renameQueue); 258 259 commit.setIEWStage(&iew); 260 rename.setIEWStage(&iew); 261 rename.setCommitStage(&commit); 262 263 ThreadID active_threads; 264 if (FullSystem) { 265 active_threads = 1; 266 } else { 267 active_threads = params->workload.size(); 268 269 if (active_threads > Impl::MaxThreads) { 270 panic("Workload Size too large. Increase the 'MaxThreads' " 271 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 272 "or edit your workload size."); 273 } 274 } 275 276 //Make Sure That this a Valid Architeture 277 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 278 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 279 assert(params->numPhysVecRegs >= numThreads * TheISA::NumVecRegs); 280 assert(params->numPhysCCRegs >= numThreads * TheISA::NumCCRegs); 281 282 rename.setScoreboard(&scoreboard); 283 iew.setScoreboard(&scoreboard); 284 285 // Setup the rename map for whichever stages need it. 286 for (ThreadID tid = 0; tid < numThreads; tid++) { 287 isa[tid] = params->isa[tid]; 288 assert(initRenameMode<TheISA::ISA>::equals(isa[tid], isa[0])); 289 290 // Only Alpha has an FP zero register, so for other ISAs we 291 // use an invalid FP register index to avoid special treatment 292 // of any valid FP reg. 293 RegIndex invalidFPReg = TheISA::NumFloatRegs + 1; 294 RegIndex fpZeroReg = 295 (THE_ISA == ALPHA_ISA) ? TheISA::ZeroReg : invalidFPReg; 296 297 commitRenameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 298 &freeList, 299 vecMode); 300 301 renameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 302 &freeList, vecMode); 303 } 304 305 // Initialize rename map to assign physical registers to the 306 // architectural registers for active threads only. 307 for (ThreadID tid = 0; tid < active_threads; tid++) { 308 for (RegIndex ridx = 0; ridx < TheISA::NumIntRegs; ++ridx) { 309 // Note that we can't use the rename() method because we don't 310 // want special treatment for the zero register at this point 311 PhysRegIdPtr phys_reg = freeList.getIntReg(); 312 renameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 313 commitRenameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 314 } 315 316 for (RegIndex ridx = 0; ridx < TheISA::NumFloatRegs; ++ridx) { 317 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 318 renameMap[tid].setEntry(RegId(FloatRegClass, ridx), phys_reg); 319 commitRenameMap[tid].setEntry( 320 RegId(FloatRegClass, ridx), phys_reg); 321 } 322 323 /* Here we need two 'interfaces' the 'whole register' and the 324 * 'register element'. At any point only one of them will be 325 * active. */ 326 if (vecMode == Enums::Full) { 327 /* Initialize the full-vector interface */ 328 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 329 RegId rid = RegId(VecRegClass, ridx); 330 PhysRegIdPtr phys_reg = freeList.getVecReg(); 331 renameMap[tid].setEntry(rid, phys_reg); 332 commitRenameMap[tid].setEntry(rid, phys_reg); 333 } 334 } else { 335 /* Initialize the vector-element interface */ 336 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 337 for (ElemIndex ldx = 0; ldx < TheISA::NumVecElemPerVecReg; 338 ++ldx) { 339 RegId lrid = RegId(VecElemClass, ridx, ldx); 340 PhysRegIdPtr phys_elem = freeList.getVecElem(); 341 renameMap[tid].setEntry(lrid, phys_elem); 342 commitRenameMap[tid].setEntry(lrid, phys_elem); 343 } 344 } 345 } 346 347 for (RegIndex ridx = 0; ridx < TheISA::NumCCRegs; ++ridx) { 348 PhysRegIdPtr phys_reg = freeList.getCCReg(); 349 renameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 350 commitRenameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 351 } 352 } 353 354 rename.setRenameMap(renameMap); 355 commit.setRenameMap(commitRenameMap); 356 rename.setFreeList(&freeList); 357 358 // Setup the ROB for whichever stages need it. 359 commit.setROB(&rob); 360 361 lastActivatedCycle = 0; 362#if 0 363 // Give renameMap & rename stage access to the freeList; 364 for (ThreadID tid = 0; tid < numThreads; tid++) 365 globalSeqNum[tid] = 1; 366#endif 367 368 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 369 370 // Setup any thread state. 371 this->thread.resize(this->numThreads); 372 373 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 374 if (FullSystem) { 375 // SMT is not supported in FS mode yet. 376 assert(this->numThreads == 1); 377 this->thread[tid] = new Thread(this, 0, NULL); 378 } else { 379 if (tid < params->workload.size()) { 380 DPRINTF(O3CPU, "Workload[%i] process is %#x", 381 tid, this->thread[tid]); 382 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 383 (typename Impl::O3CPU *)(this), 384 tid, params->workload[tid]); 385 386 //usedTids[tid] = true; 387 //threadMap[tid] = tid; 388 } else { 389 //Allocate Empty thread so M5 can use later 390 //when scheduling threads to CPU 391 Process* dummy_proc = NULL; 392 393 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 394 (typename Impl::O3CPU *)(this), 395 tid, dummy_proc); 396 //usedTids[tid] = false; 397 } 398 } 399 400 ThreadContext *tc; 401 402 // Setup the TC that will serve as the interface to the threads/CPU. 403 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 404 405 tc = o3_tc; 406 407 // If we're using a checker, then the TC should be the 408 // CheckerThreadContext. 409 if (params->checker) { 410 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 411 o3_tc, this->checker); 412 } 413 414 o3_tc->cpu = (typename Impl::O3CPU *)(this); 415 assert(o3_tc->cpu); 416 o3_tc->thread = this->thread[tid]; 417 418 // Setup quiesce event. 419 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 420 421 // Give the thread the TC. 422 this->thread[tid]->tc = tc; 423 424 // Add the TC to the CPU's list of TC's. 425 this->threadContexts.push_back(tc); 426 } 427 428 // FullO3CPU always requires an interrupt controller. 429 if (!params->switched_out && interrupts.empty()) { 430 fatal("FullO3CPU %s has no interrupt controller.\n" 431 "Ensure createInterruptController() is called.\n", name()); 432 } 433 434 for (ThreadID tid = 0; tid < this->numThreads; tid++) 435 this->thread[tid]->setFuncExeInst(0); 436} 437 438template <class Impl> 439FullO3CPU<Impl>::~FullO3CPU() 440{ 441} 442 443template <class Impl> 444void 445FullO3CPU<Impl>::regProbePoints() 446{ 447 BaseCPU::regProbePoints(); 448 449 ppInstAccessComplete = new ProbePointArg<PacketPtr>(getProbeManager(), "InstAccessComplete"); 450 ppDataAccessComplete = new ProbePointArg<std::pair<DynInstPtr, PacketPtr> >(getProbeManager(), "DataAccessComplete"); 451 452 fetch.regProbePoints(); 453 rename.regProbePoints(); 454 iew.regProbePoints(); 455 commit.regProbePoints(); 456} 457 458template <class Impl> 459void 460FullO3CPU<Impl>::regStats() 461{ 462 BaseO3CPU::regStats(); 463 464 // Register any of the O3CPU's stats here. 465 timesIdled 466 .name(name() + ".timesIdled") 467 .desc("Number of times that the entire CPU went into an idle state and" 468 " unscheduled itself") 469 .prereq(timesIdled); 470 471 idleCycles 472 .name(name() + ".idleCycles") 473 .desc("Total number of cycles that the CPU has spent unscheduled due " 474 "to idling") 475 .prereq(idleCycles); 476 477 quiesceCycles 478 .name(name() + ".quiesceCycles") 479 .desc("Total number of cycles that CPU has spent quiesced or waiting " 480 "for an interrupt") 481 .prereq(quiesceCycles); 482 483 // Number of Instructions simulated 484 // -------------------------------- 485 // Should probably be in Base CPU but need templated 486 // MaxThreads so put in here instead 487 committedInsts 488 .init(numThreads) 489 .name(name() + ".committedInsts") 490 .desc("Number of Instructions Simulated") 491 .flags(Stats::total); 492 493 committedOps 494 .init(numThreads) 495 .name(name() + ".committedOps") 496 .desc("Number of Ops (including micro ops) Simulated") 497 .flags(Stats::total); 498 499 cpi 500 .name(name() + ".cpi") 501 .desc("CPI: Cycles Per Instruction") 502 .precision(6); 503 cpi = numCycles / committedInsts; 504 505 totalCpi 506 .name(name() + ".cpi_total") 507 .desc("CPI: Total CPI of All Threads") 508 .precision(6); 509 totalCpi = numCycles / sum(committedInsts); 510 511 ipc 512 .name(name() + ".ipc") 513 .desc("IPC: Instructions Per Cycle") 514 .precision(6); 515 ipc = committedInsts / numCycles; 516 517 totalIpc 518 .name(name() + ".ipc_total") 519 .desc("IPC: Total IPC of All Threads") 520 .precision(6); 521 totalIpc = sum(committedInsts) / numCycles; 522 523 this->fetch.regStats(); 524 this->decode.regStats(); 525 this->rename.regStats(); 526 this->iew.regStats(); 527 this->commit.regStats(); 528 this->rob.regStats(); 529 530 intRegfileReads 531 .name(name() + ".int_regfile_reads") 532 .desc("number of integer regfile reads") 533 .prereq(intRegfileReads); 534 535 intRegfileWrites 536 .name(name() + ".int_regfile_writes") 537 .desc("number of integer regfile writes") 538 .prereq(intRegfileWrites); 539 540 fpRegfileReads 541 .name(name() + ".fp_regfile_reads") 542 .desc("number of floating regfile reads") 543 .prereq(fpRegfileReads); 544 545 fpRegfileWrites 546 .name(name() + ".fp_regfile_writes") 547 .desc("number of floating regfile writes") 548 .prereq(fpRegfileWrites); 549 550 vecRegfileReads 551 .name(name() + ".vec_regfile_reads") 552 .desc("number of vector regfile reads") 553 .prereq(vecRegfileReads); 554 555 vecRegfileWrites 556 .name(name() + ".vec_regfile_writes") 557 .desc("number of vector regfile writes") 558 .prereq(vecRegfileWrites); 559 560 ccRegfileReads 561 .name(name() + ".cc_regfile_reads") 562 .desc("number of cc regfile reads") 563 .prereq(ccRegfileReads); 564 565 ccRegfileWrites 566 .name(name() + ".cc_regfile_writes") 567 .desc("number of cc regfile writes") 568 .prereq(ccRegfileWrites); 569 570 miscRegfileReads 571 .name(name() + ".misc_regfile_reads") 572 .desc("number of misc regfile reads") 573 .prereq(miscRegfileReads); 574 575 miscRegfileWrites 576 .name(name() + ".misc_regfile_writes") 577 .desc("number of misc regfile writes") 578 .prereq(miscRegfileWrites); 579} 580 581template <class Impl> 582void 583FullO3CPU<Impl>::tick() 584{ 585 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 586 assert(!switchedOut()); 587 assert(drainState() != DrainState::Drained); 588 589 ++numCycles; 590 ppCycles->notify(1); 591 592// activity = false; 593 594 //Tick each of the stages 595 fetch.tick(); 596 597 decode.tick(); 598 599 rename.tick(); 600 601 iew.tick(); 602 603 commit.tick(); 604 605 // Now advance the time buffers 606 timeBuffer.advance(); 607 608 fetchQueue.advance(); 609 decodeQueue.advance(); 610 renameQueue.advance(); 611 iewQueue.advance(); 612 613 activityRec.advance(); 614 615 if (removeInstsThisCycle) { 616 cleanUpRemovedInsts(); 617 } 618 619 if (!tickEvent.scheduled()) { 620 if (_status == SwitchedOut) { 621 DPRINTF(O3CPU, "Switched out!\n"); 622 // increment stat 623 lastRunningCycle = curCycle(); 624 } else if (!activityRec.active() || _status == Idle) { 625 DPRINTF(O3CPU, "Idle!\n"); 626 lastRunningCycle = curCycle(); 627 timesIdled++; 628 } else { 629 schedule(tickEvent, clockEdge(Cycles(1))); 630 DPRINTF(O3CPU, "Scheduling next tick!\n"); 631 } 632 } 633 634 if (!FullSystem) 635 updateThreadPriority(); 636 637 tryDrain(); 638} 639 640template <class Impl> 641void 642FullO3CPU<Impl>::init() 643{ 644 BaseCPU::init(); 645 646 for (ThreadID tid = 0; tid < numThreads; ++tid) { 647 // Set noSquashFromTC so that the CPU doesn't squash when initially 648 // setting up registers. 649 thread[tid]->noSquashFromTC = true; 650 // Initialise the ThreadContext's memory proxies 651 thread[tid]->initMemProxies(thread[tid]->getTC()); 652 } 653 654 if (FullSystem && !params()->switched_out) { 655 for (ThreadID tid = 0; tid < numThreads; tid++) { 656 ThreadContext *src_tc = threadContexts[tid]; 657 TheISA::initCPU(src_tc, src_tc->contextId()); 658 } 659 } 660 661 // Clear noSquashFromTC. 662 for (int tid = 0; tid < numThreads; ++tid) 663 thread[tid]->noSquashFromTC = false; 664 665 commit.setThreads(thread); 666} 667 668template <class Impl> 669void 670FullO3CPU<Impl>::startup() 671{ 672 BaseCPU::startup(); 673 for (int tid = 0; tid < numThreads; ++tid) 674 isa[tid]->startup(threadContexts[tid]); 675 676 fetch.startupStage(); 677 decode.startupStage(); 678 iew.startupStage(); 679 rename.startupStage(); 680 commit.startupStage(); 681} 682 683template <class Impl> 684void 685FullO3CPU<Impl>::activateThread(ThreadID tid) 686{ 687 list<ThreadID>::iterator isActive = 688 std::find(activeThreads.begin(), activeThreads.end(), tid); 689 690 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 691 assert(!switchedOut()); 692 693 if (isActive == activeThreads.end()) { 694 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 695 tid); 696 697 activeThreads.push_back(tid); 698 } 699} 700 701template <class Impl> 702void 703FullO3CPU<Impl>::deactivateThread(ThreadID tid) 704{ 705 //Remove From Active List, if Active 706 list<ThreadID>::iterator thread_it = 707 std::find(activeThreads.begin(), activeThreads.end(), tid); 708 709 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 710 assert(!switchedOut()); 711 712 if (thread_it != activeThreads.end()) { 713 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 714 tid); 715 activeThreads.erase(thread_it); 716 } 717 718 fetch.deactivateThread(tid); 719 commit.deactivateThread(tid); 720} 721 722template <class Impl> 723Counter 724FullO3CPU<Impl>::totalInsts() const 725{ 726 Counter total(0); 727 728 ThreadID size = thread.size(); 729 for (ThreadID i = 0; i < size; i++) 730 total += thread[i]->numInst; 731 732 return total; 733} 734 735template <class Impl> 736Counter 737FullO3CPU<Impl>::totalOps() const 738{ 739 Counter total(0); 740 741 ThreadID size = thread.size(); 742 for (ThreadID i = 0; i < size; i++) 743 total += thread[i]->numOp; 744 745 return total; 746} 747 748template <class Impl> 749void 750FullO3CPU<Impl>::activateContext(ThreadID tid) 751{ 752 assert(!switchedOut()); 753 754 // Needs to set each stage to running as well. 755 activateThread(tid); 756 757 // We don't want to wake the CPU if it is drained. In that case, 758 // we just want to flag the thread as active and schedule the tick 759 // event from drainResume() instead. 760 if (drainState() == DrainState::Drained) 761 return; 762 763 // If we are time 0 or if the last activation time is in the past, 764 // schedule the next tick and wake up the fetch unit 765 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 766 scheduleTickEvent(Cycles(0)); 767 768 // Be sure to signal that there's some activity so the CPU doesn't 769 // deschedule itself. 770 activityRec.activity(); 771 fetch.wakeFromQuiesce(); 772 773 Cycles cycles(curCycle() - lastRunningCycle); 774 // @todo: This is an oddity that is only here to match the stats 775 if (cycles != 0) 776 --cycles; 777 quiesceCycles += cycles; 778 779 lastActivatedCycle = curTick(); 780 781 _status = Running; 782 783 BaseCPU::activateContext(tid); 784 } 785} 786 787template <class Impl> 788void 789FullO3CPU<Impl>::suspendContext(ThreadID tid) 790{ 791 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 792 assert(!switchedOut()); 793 794 deactivateThread(tid); 795 796 // If this was the last thread then unschedule the tick event. 797 if (activeThreads.size() == 0) { 798 unscheduleTickEvent(); 799 lastRunningCycle = curCycle(); 800 _status = Idle; 801 } 802 803 DPRINTF(Quiesce, "Suspending Context\n"); 804 805 BaseCPU::suspendContext(tid); 806} 807 808template <class Impl> 809void 810FullO3CPU<Impl>::haltContext(ThreadID tid) 811{ 812 //For now, this is the same as deallocate 813 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 814 assert(!switchedOut()); 815 816 deactivateThread(tid); 817 removeThread(tid); 818} 819 820template <class Impl> 821void 822FullO3CPU<Impl>::insertThread(ThreadID tid) 823{ 824 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 825 // Will change now that the PC and thread state is internal to the CPU 826 // and not in the ThreadContext. 827 ThreadContext *src_tc; 828 if (FullSystem) 829 src_tc = system->threadContexts[tid]; 830 else 831 src_tc = tcBase(tid); 832 833 //Bind Int Regs to Rename Map 834 835 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 836 reg_id.index()++) { 837 PhysRegIdPtr phys_reg = freeList.getIntReg(); 838 renameMap[tid].setEntry(reg_id, phys_reg); 839 scoreboard.setReg(phys_reg); 840 } 841 842 //Bind Float Regs to Rename Map 843 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 844 reg_id.index()++) { 845 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 846 renameMap[tid].setEntry(reg_id, phys_reg); 847 scoreboard.setReg(phys_reg); 848 } 849 850 //Bind condition-code Regs to Rename Map 851 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 852 reg_id.index()++) { 853 PhysRegIdPtr phys_reg = freeList.getCCReg(); 854 renameMap[tid].setEntry(reg_id, phys_reg); 855 scoreboard.setReg(phys_reg); 856 } 857 858 //Copy Thread Data Into RegFile 859 //this->copyFromTC(tid); 860 861 //Set PC/NPC/NNPC 862 pcState(src_tc->pcState(), tid); 863 864 src_tc->setStatus(ThreadContext::Active); 865 866 activateContext(tid); 867 868 //Reset ROB/IQ/LSQ Entries 869 commit.rob->resetEntries(); 870 iew.resetEntries(); 871} 872 873template <class Impl> 874void 875FullO3CPU<Impl>::removeThread(ThreadID tid) 876{ 877 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 878 879 // Copy Thread Data From RegFile 880 // If thread is suspended, it might be re-allocated 881 // this->copyToTC(tid); 882 883 884 // @todo: 2-27-2008: Fix how we free up rename mappings 885 // here to alleviate the case for double-freeing registers 886 // in SMT workloads. 887 888 // Unbind Int Regs from Rename Map 889 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 890 reg_id.index()++) { 891 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 892 scoreboard.unsetReg(phys_reg); 893 freeList.addReg(phys_reg); 894 } 895 896 // Unbind Float Regs from Rename Map 897 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 898 reg_id.index()++) { 899 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 900 scoreboard.unsetReg(phys_reg); 901 freeList.addReg(phys_reg); 902 } 903 904 // Unbind condition-code Regs from Rename Map 905 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 906 reg_id.index()++) { 907 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 908 scoreboard.unsetReg(phys_reg); 909 freeList.addReg(phys_reg); 910 } 911 912 // Squash Throughout Pipeline 913 DynInstPtr inst = commit.rob->readHeadInst(tid); 914 InstSeqNum squash_seq_num = inst->seqNum; 915 fetch.squash(0, squash_seq_num, inst, tid); 916 decode.squash(tid); 917 rename.squash(squash_seq_num, tid); 918 iew.squash(tid); 919 iew.ldstQueue.squash(squash_seq_num, tid); 920 commit.rob->squash(squash_seq_num, tid); 921 922 923 assert(iew.instQueue.getCount(tid) == 0); 924 assert(iew.ldstQueue.getCount(tid) == 0); 925 926 // Reset ROB/IQ/LSQ Entries 927 928 // Commented out for now. This should be possible to do by 929 // telling all the pipeline stages to drain first, and then 930 // checking until the drain completes. Once the pipeline is 931 // drained, call resetEntries(). - 10-09-06 ktlim 932/* 933 if (activeThreads.size() >= 1) { 934 commit.rob->resetEntries(); 935 iew.resetEntries(); 936 } 937*/ 938} 939 940template <class Impl> 941Fault 942FullO3CPU<Impl>::hwrei(ThreadID tid) 943{ 944#if THE_ISA == ALPHA_ISA 945 // Need to clear the lock flag upon returning from an interrupt. 946 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 947 948 this->thread[tid]->kernelStats->hwrei(); 949 950 // FIXME: XXX check for interrupts? XXX 951#endif 952 return NoFault; 953} 954 955template <class Impl> 956bool 957FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 958{ 959#if THE_ISA == ALPHA_ISA 960 if (this->thread[tid]->kernelStats) 961 this->thread[tid]->kernelStats->callpal(palFunc, 962 this->threadContexts[tid]); 963 964 switch (palFunc) { 965 case PAL::halt: 966 halt(); 967 if (--System::numSystemsRunning == 0) 968 exitSimLoop("all cpus halted"); 969 break; 970 971 case PAL::bpt: 972 case PAL::bugchk: 973 if (this->system->breakpoint()) 974 return false; 975 break; 976 } 977#endif 978 return true; 979} 980 981template <class Impl> 982Fault 983FullO3CPU<Impl>::getInterrupts() 984{ 985 // Check if there are any outstanding interrupts 986 return this->interrupts[0]->getInterrupt(this->threadContexts[0]); 987} 988 989template <class Impl> 990void 991FullO3CPU<Impl>::processInterrupts(const Fault &interrupt) 992{ 993 // Check for interrupts here. For now can copy the code that 994 // exists within isa_fullsys_traits.hh. Also assume that thread 0 995 // is the one that handles the interrupts. 996 // @todo: Possibly consolidate the interrupt checking code. 997 // @todo: Allow other threads to handle interrupts. 998 999 assert(interrupt != NoFault); 1000 this->interrupts[0]->updateIntrInfo(this->threadContexts[0]); 1001 1002 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1003 this->trap(interrupt, 0, nullptr); 1004} 1005 1006template <class Impl> 1007void 1008FullO3CPU<Impl>::trap(const Fault &fault, ThreadID tid, 1009 const StaticInstPtr &inst) 1010{ 1011 // Pass the thread's TC into the invoke method. 1012 fault->invoke(this->threadContexts[tid], inst); 1013} 1014 1015template <class Impl> 1016void 1017FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid, Fault *fault) 1018{ 1019 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1020 1021 DPRINTF(Activity,"Activity: syscall() called.\n"); 1022 1023 // Temporarily increase this by one to account for the syscall 1024 // instruction. 1025 ++(this->thread[tid]->funcExeInst); 1026 1027 // Execute the actual syscall. 1028 this->thread[tid]->syscall(callnum, fault); 1029 1030 // Decrease funcExeInst by one as the normal commit will handle 1031 // incrementing it. 1032 --(this->thread[tid]->funcExeInst); 1033} 1034 1035template <class Impl> 1036void 1037FullO3CPU<Impl>::serializeThread(CheckpointOut &cp, ThreadID tid) const 1038{ 1039 thread[tid]->serialize(cp); 1040} 1041 1042template <class Impl> 1043void 1044FullO3CPU<Impl>::unserializeThread(CheckpointIn &cp, ThreadID tid) 1045{ 1046 thread[tid]->unserialize(cp); 1047} 1048 1049template <class Impl> 1050DrainState 1051FullO3CPU<Impl>::drain() 1052{ 1053 // If the CPU isn't doing anything, then return immediately. 1054 if (switchedOut()) 1055 return DrainState::Drained; 1056 1057 DPRINTF(Drain, "Draining...\n"); 1058 1059 // We only need to signal a drain to the commit stage as this 1060 // initiates squashing controls the draining. Once the commit 1061 // stage commits an instruction where it is safe to stop, it'll 1062 // squash the rest of the instructions in the pipeline and force 1063 // the fetch stage to stall. The pipeline will be drained once all 1064 // in-flight instructions have retired. 1065 commit.drain(); 1066 1067 // Wake the CPU and record activity so everything can drain out if 1068 // the CPU was not able to immediately drain. 1069 if (!isDrained()) { 1070 wakeCPU(); 1071 activityRec.activity(); 1072 1073 DPRINTF(Drain, "CPU not drained\n"); 1074 1075 return DrainState::Draining; 1076 } else { 1077 DPRINTF(Drain, "CPU is already drained\n"); 1078 if (tickEvent.scheduled()) 1079 deschedule(tickEvent); 1080 1081 // Flush out any old data from the time buffers. In 1082 // particular, there might be some data in flight from the 1083 // fetch stage that isn't visible in any of the CPU buffers we 1084 // test in isDrained(). 1085 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1086 timeBuffer.advance(); 1087 fetchQueue.advance(); 1088 decodeQueue.advance(); 1089 renameQueue.advance(); 1090 iewQueue.advance(); 1091 } 1092 1093 drainSanityCheck(); 1094 return DrainState::Drained; 1095 } 1096} 1097 1098template <class Impl> 1099bool 1100FullO3CPU<Impl>::tryDrain() 1101{ 1102 if (drainState() != DrainState::Draining || !isDrained()) 1103 return false; 1104 1105 if (tickEvent.scheduled()) 1106 deschedule(tickEvent); 1107 1108 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1109 signalDrainDone(); 1110 1111 return true; 1112} 1113 1114template <class Impl> 1115void 1116FullO3CPU<Impl>::drainSanityCheck() const 1117{ 1118 assert(isDrained()); 1119 fetch.drainSanityCheck(); 1120 decode.drainSanityCheck(); 1121 rename.drainSanityCheck(); 1122 iew.drainSanityCheck(); 1123 commit.drainSanityCheck(); 1124} 1125 1126template <class Impl> 1127bool 1128FullO3CPU<Impl>::isDrained() const 1129{ 1130 bool drained(true); 1131 1132 if (!instList.empty() || !removeList.empty()) { 1133 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1134 drained = false; 1135 } 1136 1137 if (!fetch.isDrained()) { 1138 DPRINTF(Drain, "Fetch not drained.\n"); 1139 drained = false; 1140 } 1141 1142 if (!decode.isDrained()) { 1143 DPRINTF(Drain, "Decode not drained.\n"); 1144 drained = false; 1145 } 1146 1147 if (!rename.isDrained()) { 1148 DPRINTF(Drain, "Rename not drained.\n"); 1149 drained = false; 1150 } 1151 1152 if (!iew.isDrained()) { 1153 DPRINTF(Drain, "IEW not drained.\n"); 1154 drained = false; 1155 } 1156 1157 if (!commit.isDrained()) { 1158 DPRINTF(Drain, "Commit not drained.\n"); 1159 drained = false; 1160 } 1161 1162 return drained; 1163} 1164 1165template <class Impl> 1166void 1167FullO3CPU<Impl>::commitDrained(ThreadID tid) 1168{ 1169 fetch.drainStall(tid); 1170} 1171 1172template <class Impl> 1173void 1174FullO3CPU<Impl>::drainResume() 1175{ 1176 if (switchedOut()) 1177 return; 1178 1179 DPRINTF(Drain, "Resuming...\n"); 1180 verifyMemoryMode(); 1181 1182 fetch.drainResume(); 1183 commit.drainResume(); 1184 1185 _status = Idle; 1186 for (ThreadID i = 0; i < thread.size(); i++) { 1187 if (thread[i]->status() == ThreadContext::Active) { 1188 DPRINTF(Drain, "Activating thread: %i\n", i); 1189 activateThread(i); 1190 _status = Running; 1191 } 1192 } 1193 1194 assert(!tickEvent.scheduled()); 1195 if (_status == Running) 1196 schedule(tickEvent, nextCycle()); 1197} 1198 1199template <class Impl> 1200void 1201FullO3CPU<Impl>::switchOut() 1202{ 1203 DPRINTF(O3CPU, "Switching out\n"); 1204 BaseCPU::switchOut(); 1205 1206 activityRec.reset(); 1207 1208 _status = SwitchedOut; 1209 1210 if (checker) 1211 checker->switchOut(); 1212} 1213 1214template <class Impl> 1215void 1216FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1217{ 1218 BaseCPU::takeOverFrom(oldCPU); 1219 1220 fetch.takeOverFrom(); 1221 decode.takeOverFrom(); 1222 rename.takeOverFrom(); 1223 iew.takeOverFrom(); 1224 commit.takeOverFrom(); 1225 1226 assert(!tickEvent.scheduled()); 1227 1228 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1229 if (oldO3CPU) 1230 globalSeqNum = oldO3CPU->globalSeqNum; 1231 1232 lastRunningCycle = curCycle(); 1233 _status = Idle; 1234} 1235 1236template <class Impl> 1237void 1238FullO3CPU<Impl>::verifyMemoryMode() const 1239{ 1240 if (!system->isTimingMode()) { 1241 fatal("The O3 CPU requires the memory system to be in " 1242 "'timing' mode.\n"); 1243 } 1244} 1245 1246template <class Impl> 1247TheISA::MiscReg 1248FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) const 1249{ 1250 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1251} 1252 1253template <class Impl> 1254TheISA::MiscReg 1255FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1256{ 1257 miscRegfileReads++; 1258 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1259} 1260 1261template <class Impl> 1262void 1263FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1264 const TheISA::MiscReg &val, ThreadID tid) 1265{ 1266 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1267} 1268 1269template <class Impl> 1270void 1271FullO3CPU<Impl>::setMiscReg(int misc_reg, 1272 const TheISA::MiscReg &val, ThreadID tid) 1273{ 1274 miscRegfileWrites++; 1275 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1276} 1277 1278template <class Impl> 1279uint64_t 1280FullO3CPU<Impl>::readIntReg(PhysRegIdPtr phys_reg) 1281{ 1282 intRegfileReads++; 1283 return regFile.readIntReg(phys_reg); 1284} 1285 1286template <class Impl> 1287FloatReg 1288FullO3CPU<Impl>::readFloatReg(PhysRegIdPtr phys_reg) 1289{ 1290 fpRegfileReads++; 1291 return regFile.readFloatReg(phys_reg); 1292} 1293 1294template <class Impl> 1295FloatRegBits 1296FullO3CPU<Impl>::readFloatRegBits(PhysRegIdPtr phys_reg) 1297{ 1298 fpRegfileReads++; 1299 return regFile.readFloatRegBits(phys_reg); 1300} 1301 1302template <class Impl> 1303auto 1304FullO3CPU<Impl>::readVecReg(PhysRegIdPtr phys_reg) const 1305 -> const VecRegContainer& 1306{ 1307 vecRegfileReads++; 1308 return regFile.readVecReg(phys_reg); 1309} 1310 1311template <class Impl> 1312auto 1313FullO3CPU<Impl>::getWritableVecReg(PhysRegIdPtr phys_reg) 1314 -> VecRegContainer& 1315{ 1316 vecRegfileWrites++; 1317 return regFile.getWritableVecReg(phys_reg); 1318} 1319 1320template <class Impl> 1321auto 1322FullO3CPU<Impl>::readVecElem(PhysRegIdPtr phys_reg) const -> const VecElem& 1323{ 1324 vecRegfileReads++; 1325 return regFile.readVecElem(phys_reg); 1326} 1327 1328template <class Impl> 1329CCReg 1330FullO3CPU<Impl>::readCCReg(PhysRegIdPtr phys_reg) 1331{ 1332 ccRegfileReads++; 1333 return regFile.readCCReg(phys_reg); 1334} 1335 1336template <class Impl> 1337void 1338FullO3CPU<Impl>::setIntReg(PhysRegIdPtr phys_reg, uint64_t val) 1339{ 1340 intRegfileWrites++; 1341 regFile.setIntReg(phys_reg, val); 1342} 1343 1344template <class Impl> 1345void 1346FullO3CPU<Impl>::setFloatReg(PhysRegIdPtr phys_reg, FloatReg val) 1347{ 1348 fpRegfileWrites++; 1349 regFile.setFloatReg(phys_reg, val); 1350} 1351 1352template <class Impl> 1353void 1354FullO3CPU<Impl>::setFloatRegBits(PhysRegIdPtr phys_reg, FloatRegBits val) 1355{ 1356 fpRegfileWrites++; 1357 regFile.setFloatRegBits(phys_reg, val); 1358} 1359 1360template <class Impl> 1361void 1362FullO3CPU<Impl>::setVecReg(PhysRegIdPtr phys_reg, const VecRegContainer& val) 1363{ 1364 vecRegfileWrites++; 1365 regFile.setVecReg(phys_reg, val); 1366} 1367 1368template <class Impl> 1369void 1370FullO3CPU<Impl>::setVecElem(PhysRegIdPtr phys_reg, const VecElem& val) 1371{ 1372 vecRegfileWrites++; 1373 regFile.setVecElem(phys_reg, val); 1374} 1375 1376template <class Impl> 1377void 1378FullO3CPU<Impl>::setCCReg(PhysRegIdPtr phys_reg, CCReg val) 1379{ 1380 ccRegfileWrites++; 1381 regFile.setCCReg(phys_reg, val); 1382} 1383 1384template <class Impl> 1385uint64_t 1386FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1387{ 1388 intRegfileReads++; 1389 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1390 RegId(IntRegClass, reg_idx)); 1391 1392 return regFile.readIntReg(phys_reg); 1393} 1394 1395template <class Impl> 1396float 1397FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1398{ 1399 fpRegfileReads++; 1400 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1401 RegId(FloatRegClass, reg_idx)); 1402 1403 return regFile.readFloatReg(phys_reg); 1404} 1405 1406template <class Impl> 1407uint64_t 1408FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1409{ 1410 fpRegfileReads++; 1411 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1412 RegId(FloatRegClass, reg_idx)); 1413 1414 return regFile.readFloatRegBits(phys_reg); 1415} 1416 1417template <class Impl> 1418auto 1419FullO3CPU<Impl>::readArchVecReg(int reg_idx, ThreadID tid) const 1420 -> const VecRegContainer& 1421{ 1422 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1423 RegId(VecRegClass, reg_idx)); 1424 return readVecReg(phys_reg); 1425} 1426 1427template <class Impl> 1428auto 1429FullO3CPU<Impl>::getWritableArchVecReg(int reg_idx, ThreadID tid) 1430 -> VecRegContainer& 1431{ 1432 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1433 RegId(VecRegClass, reg_idx)); 1434 return getWritableVecReg(phys_reg); 1435} 1436 1437template <class Impl> 1438auto 1439FullO3CPU<Impl>::readArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1440 ThreadID tid) const -> const VecElem& 1441{ 1442 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1443 RegId(VecRegClass, reg_idx, ldx)); 1444 return readVecElem(phys_reg); 1445} 1446 1447template <class Impl> 1448CCReg 1449FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1450{ 1451 ccRegfileReads++; 1452 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1453 RegId(CCRegClass, reg_idx)); 1454 1455 return regFile.readCCReg(phys_reg); 1456} 1457 1458template <class Impl> 1459void 1460FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1461{ 1462 intRegfileWrites++; 1463 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1464 RegId(IntRegClass, reg_idx)); 1465 1466 regFile.setIntReg(phys_reg, val); 1467} 1468 1469template <class Impl> 1470void 1471FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1472{ 1473 fpRegfileWrites++; 1474 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1475 RegId(FloatRegClass, reg_idx)); 1476 1477 regFile.setFloatReg(phys_reg, val); 1478} 1479 1480template <class Impl> 1481void 1482FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1483{ 1484 fpRegfileWrites++; 1485 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1486 RegId(FloatRegClass, reg_idx)); 1487 1488 regFile.setFloatRegBits(phys_reg, val); 1489} 1490 1491template <class Impl> 1492void 1493FullO3CPU<Impl>::setArchVecReg(int reg_idx, const VecRegContainer& val, 1494 ThreadID tid) 1495{ 1496 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1497 RegId(VecRegClass, reg_idx)); 1498 setVecReg(phys_reg, val); 1499} 1500 1501template <class Impl> 1502void 1503FullO3CPU<Impl>::setArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1504 const VecElem& val, ThreadID tid) 1505{ 1506 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1507 RegId(VecRegClass, reg_idx, ldx)); 1508 setVecElem(phys_reg, val); 1509} 1510 1511template <class Impl> 1512void 1513FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1514{ 1515 ccRegfileWrites++; 1516 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1517 RegId(CCRegClass, reg_idx)); 1518 1519 regFile.setCCReg(phys_reg, val); 1520} 1521 1522template <class Impl> 1523TheISA::PCState 1524FullO3CPU<Impl>::pcState(ThreadID tid) 1525{ 1526 return commit.pcState(tid); 1527} 1528 1529template <class Impl> 1530void 1531FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1532{ 1533 commit.pcState(val, tid); 1534} 1535 1536template <class Impl> 1537Addr 1538FullO3CPU<Impl>::instAddr(ThreadID tid) 1539{ 1540 return commit.instAddr(tid); 1541} 1542 1543template <class Impl> 1544Addr 1545FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1546{ 1547 return commit.nextInstAddr(tid); 1548} 1549 1550template <class Impl> 1551MicroPC 1552FullO3CPU<Impl>::microPC(ThreadID tid) 1553{ 1554 return commit.microPC(tid); 1555} 1556 1557template <class Impl> 1558void 1559FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1560{ 1561 this->thread[tid]->noSquashFromTC = true; 1562 this->commit.generateTCEvent(tid); 1563} 1564 1565template <class Impl> 1566typename FullO3CPU<Impl>::ListIt 1567FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1568{ 1569 instList.push_back(inst); 1570 1571 return --(instList.end()); 1572} 1573 1574template <class Impl> 1575void 1576FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1577{ 1578 // Keep an instruction count. 1579 if (!inst->isMicroop() || inst->isLastMicroop()) { 1580 thread[tid]->numInst++; 1581 thread[tid]->numInsts++; 1582 committedInsts[tid]++; 1583 system->totalNumInsts++; 1584 1585 // Check for instruction-count-based events. 1586 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1587 system->instEventQueue.serviceEvents(system->totalNumInsts); 1588 } 1589 thread[tid]->numOp++; 1590 thread[tid]->numOps++; 1591 committedOps[tid]++; 1592 1593 probeInstCommit(inst->staticInst); 1594} 1595 1596template <class Impl> 1597void 1598FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1599{ 1600 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1601 "[sn:%lli]\n", 1602 inst->threadNumber, inst->pcState(), inst->seqNum); 1603 1604 removeInstsThisCycle = true; 1605 1606 // Remove the front instruction. 1607 removeList.push(inst->getInstListIt()); 1608} 1609 1610template <class Impl> 1611void 1612FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1613{ 1614 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1615 " list.\n", tid); 1616 1617 ListIt end_it; 1618 1619 bool rob_empty = false; 1620 1621 if (instList.empty()) { 1622 return; 1623 } else if (rob.isEmpty(tid)) { 1624 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1625 end_it = instList.begin(); 1626 rob_empty = true; 1627 } else { 1628 end_it = (rob.readTailInst(tid))->getInstListIt(); 1629 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1630 } 1631 1632 removeInstsThisCycle = true; 1633 1634 ListIt inst_it = instList.end(); 1635 1636 inst_it--; 1637 1638 // Walk through the instruction list, removing any instructions 1639 // that were inserted after the given instruction iterator, end_it. 1640 while (inst_it != end_it) { 1641 assert(!instList.empty()); 1642 1643 squashInstIt(inst_it, tid); 1644 1645 inst_it--; 1646 } 1647 1648 // If the ROB was empty, then we actually need to remove the first 1649 // instruction as well. 1650 if (rob_empty) { 1651 squashInstIt(inst_it, tid); 1652 } 1653} 1654 1655template <class Impl> 1656void 1657FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1658{ 1659 assert(!instList.empty()); 1660 1661 removeInstsThisCycle = true; 1662 1663 ListIt inst_iter = instList.end(); 1664 1665 inst_iter--; 1666 1667 DPRINTF(O3CPU, "Deleting instructions from instruction " 1668 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1669 tid, seq_num, (*inst_iter)->seqNum); 1670 1671 while ((*inst_iter)->seqNum > seq_num) { 1672 1673 bool break_loop = (inst_iter == instList.begin()); 1674 1675 squashInstIt(inst_iter, tid); 1676 1677 inst_iter--; 1678 1679 if (break_loop) 1680 break; 1681 } 1682} 1683 1684template <class Impl> 1685inline void 1686FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1687{ 1688 if ((*instIt)->threadNumber == tid) { 1689 DPRINTF(O3CPU, "Squashing instruction, " 1690 "[tid:%i] [sn:%lli] PC %s\n", 1691 (*instIt)->threadNumber, 1692 (*instIt)->seqNum, 1693 (*instIt)->pcState()); 1694 1695 // Mark it as squashed. 1696 (*instIt)->setSquashed(); 1697 1698 // @todo: Formulate a consistent method for deleting 1699 // instructions from the instruction list 1700 // Remove the instruction from the list. 1701 removeList.push(instIt); 1702 } 1703} 1704 1705template <class Impl> 1706void 1707FullO3CPU<Impl>::cleanUpRemovedInsts() 1708{ 1709 while (!removeList.empty()) { 1710 DPRINTF(O3CPU, "Removing instruction, " 1711 "[tid:%i] [sn:%lli] PC %s\n", 1712 (*removeList.front())->threadNumber, 1713 (*removeList.front())->seqNum, 1714 (*removeList.front())->pcState()); 1715 1716 instList.erase(removeList.front()); 1717 1718 removeList.pop(); 1719 } 1720 1721 removeInstsThisCycle = false; 1722} 1723/* 1724template <class Impl> 1725void 1726FullO3CPU<Impl>::removeAllInsts() 1727{ 1728 instList.clear(); 1729} 1730*/ 1731template <class Impl> 1732void 1733FullO3CPU<Impl>::dumpInsts() 1734{ 1735 int num = 0; 1736 1737 ListIt inst_list_it = instList.begin(); 1738 1739 cprintf("Dumping Instruction List\n"); 1740 1741 while (inst_list_it != instList.end()) { 1742 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1743 "Squashed:%i\n\n", 1744 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1745 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1746 (*inst_list_it)->isSquashed()); 1747 inst_list_it++; 1748 ++num; 1749 } 1750} 1751/* 1752template <class Impl> 1753void 1754FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1755{ 1756 iew.wakeDependents(inst); 1757} 1758*/ 1759template <class Impl> 1760void 1761FullO3CPU<Impl>::wakeCPU() 1762{ 1763 if (activityRec.active() || tickEvent.scheduled()) { 1764 DPRINTF(Activity, "CPU already running.\n"); 1765 return; 1766 } 1767 1768 DPRINTF(Activity, "Waking up CPU\n"); 1769 1770 Cycles cycles(curCycle() - lastRunningCycle); 1771 // @todo: This is an oddity that is only here to match the stats 1772 if (cycles > 1) { 1773 --cycles; 1774 idleCycles += cycles; 1775 numCycles += cycles; 1776 ppCycles->notify(cycles); 1777 } 1778 1779 schedule(tickEvent, clockEdge()); 1780} 1781 1782template <class Impl> 1783void 1784FullO3CPU<Impl>::wakeup(ThreadID tid) 1785{ 1786 if (this->thread[tid]->status() != ThreadContext::Suspended) 1787 return; 1788 1789 this->wakeCPU(); 1790 1791 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1792 this->threadContexts[tid]->activate(); 1793} 1794 1795template <class Impl> 1796ThreadID 1797FullO3CPU<Impl>::getFreeTid() 1798{ 1799 for (ThreadID tid = 0; tid < numThreads; tid++) { 1800 if (!tids[tid]) { 1801 tids[tid] = true; 1802 return tid; 1803 } 1804 } 1805 1806 return InvalidThreadID; 1807} 1808 1809template <class Impl> 1810void 1811FullO3CPU<Impl>::updateThreadPriority() 1812{ 1813 if (activeThreads.size() > 1) { 1814 //DEFAULT TO ROUND ROBIN SCHEME 1815 //e.g. Move highest priority to end of thread list 1816 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1817 1818 unsigned high_thread = *list_begin; 1819 1820 activeThreads.erase(list_begin); 1821 1822 activeThreads.push_back(high_thread); 1823 } 1824} 1825 1826// Forward declaration of FullO3CPU. 1827template class FullO3CPU<O3CPUImpl>; 1828