cpu.cc revision 12276
1/* 2 * Copyright (c) 2011-2012, 2014, 2016, 2017 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>::FullO3CPU(DerivO3CPUParams *params) 141 : BaseO3CPU(params), 142 itb(params->itb), 143 dtb(params->dtb), 144 tickEvent([this]{ tick(); }, "FullO3CPU tick", 145 false, Event::CPU_Tick_Pri), 146#ifndef NDEBUG 147 instcount(0), 148#endif 149 removeInstsThisCycle(false), 150 fetch(this, params), 151 decode(this, params), 152 rename(this, params), 153 iew(this, params), 154 commit(this, params), 155 156 /* It is mandatory that all SMT threads use the same renaming mode as 157 * they are sharing registers and rename */ 158 vecMode(initRenameMode<TheISA::ISA>::mode(params->isa[0])), 159 regFile(params->numPhysIntRegs, 160 params->numPhysFloatRegs, 161 params->numPhysVecRegs, 162 params->numPhysCCRegs, 163 vecMode), 164 165 freeList(name() + ".freelist", ®File), 166 167 rob(this, params), 168 169 scoreboard(name() + ".scoreboard", 170 regFile.totalNumPhysRegs()), 171 172 isa(numThreads, NULL), 173 174 icachePort(&fetch, this), 175 dcachePort(&iew.ldstQueue, this), 176 177 timeBuffer(params->backComSize, params->forwardComSize), 178 fetchQueue(params->backComSize, params->forwardComSize), 179 decodeQueue(params->backComSize, params->forwardComSize), 180 renameQueue(params->backComSize, params->forwardComSize), 181 iewQueue(params->backComSize, params->forwardComSize), 182 activityRec(name(), NumStages, 183 params->backComSize + params->forwardComSize, 184 params->activity), 185 186 globalSeqNum(1), 187 system(params->system), 188 lastRunningCycle(curCycle()) 189{ 190 if (!params->switched_out) { 191 _status = Running; 192 } else { 193 _status = SwitchedOut; 194 } 195 196 if (params->checker) { 197 BaseCPU *temp_checker = params->checker; 198 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 199 checker->setIcachePort(&icachePort); 200 checker->setSystem(params->system); 201 } else { 202 checker = NULL; 203 } 204 205 if (!FullSystem) { 206 thread.resize(numThreads); 207 tids.resize(numThreads); 208 } 209 210 // The stages also need their CPU pointer setup. However this 211 // must be done at the upper level CPU because they have pointers 212 // to the upper level CPU, and not this FullO3CPU. 213 214 // Set up Pointers to the activeThreads list for each stage 215 fetch.setActiveThreads(&activeThreads); 216 decode.setActiveThreads(&activeThreads); 217 rename.setActiveThreads(&activeThreads); 218 iew.setActiveThreads(&activeThreads); 219 commit.setActiveThreads(&activeThreads); 220 221 // Give each of the stages the time buffer they will use. 222 fetch.setTimeBuffer(&timeBuffer); 223 decode.setTimeBuffer(&timeBuffer); 224 rename.setTimeBuffer(&timeBuffer); 225 iew.setTimeBuffer(&timeBuffer); 226 commit.setTimeBuffer(&timeBuffer); 227 228 // Also setup each of the stages' queues. 229 fetch.setFetchQueue(&fetchQueue); 230 decode.setFetchQueue(&fetchQueue); 231 commit.setFetchQueue(&fetchQueue); 232 decode.setDecodeQueue(&decodeQueue); 233 rename.setDecodeQueue(&decodeQueue); 234 rename.setRenameQueue(&renameQueue); 235 iew.setRenameQueue(&renameQueue); 236 iew.setIEWQueue(&iewQueue); 237 commit.setIEWQueue(&iewQueue); 238 commit.setRenameQueue(&renameQueue); 239 240 commit.setIEWStage(&iew); 241 rename.setIEWStage(&iew); 242 rename.setCommitStage(&commit); 243 244 ThreadID active_threads; 245 if (FullSystem) { 246 active_threads = 1; 247 } else { 248 active_threads = params->workload.size(); 249 250 if (active_threads > Impl::MaxThreads) { 251 panic("Workload Size too large. Increase the 'MaxThreads' " 252 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 253 "or edit your workload size."); 254 } 255 } 256 257 //Make Sure That this a Valid Architeture 258 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 259 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 260 assert(params->numPhysVecRegs >= numThreads * TheISA::NumVecRegs); 261 assert(params->numPhysCCRegs >= numThreads * TheISA::NumCCRegs); 262 263 rename.setScoreboard(&scoreboard); 264 iew.setScoreboard(&scoreboard); 265 266 // Setup the rename map for whichever stages need it. 267 for (ThreadID tid = 0; tid < numThreads; tid++) { 268 isa[tid] = params->isa[tid]; 269 assert(initRenameMode<TheISA::ISA>::equals(isa[tid], isa[0])); 270 271 // Only Alpha has an FP zero register, so for other ISAs we 272 // use an invalid FP register index to avoid special treatment 273 // of any valid FP reg. 274 RegIndex invalidFPReg = TheISA::NumFloatRegs + 1; 275 RegIndex fpZeroReg = 276 (THE_ISA == ALPHA_ISA) ? TheISA::ZeroReg : invalidFPReg; 277 278 commitRenameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 279 &freeList, 280 vecMode); 281 282 renameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 283 &freeList, vecMode); 284 } 285 286 // Initialize rename map to assign physical registers to the 287 // architectural registers for active threads only. 288 for (ThreadID tid = 0; tid < active_threads; tid++) { 289 for (RegIndex ridx = 0; ridx < TheISA::NumIntRegs; ++ridx) { 290 // Note that we can't use the rename() method because we don't 291 // want special treatment for the zero register at this point 292 PhysRegIdPtr phys_reg = freeList.getIntReg(); 293 renameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 294 commitRenameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 295 } 296 297 for (RegIndex ridx = 0; ridx < TheISA::NumFloatRegs; ++ridx) { 298 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 299 renameMap[tid].setEntry(RegId(FloatRegClass, ridx), phys_reg); 300 commitRenameMap[tid].setEntry( 301 RegId(FloatRegClass, ridx), phys_reg); 302 } 303 304 /* Here we need two 'interfaces' the 'whole register' and the 305 * 'register element'. At any point only one of them will be 306 * active. */ 307 if (vecMode == Enums::Full) { 308 /* Initialize the full-vector interface */ 309 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 310 RegId rid = RegId(VecRegClass, ridx); 311 PhysRegIdPtr phys_reg = freeList.getVecReg(); 312 renameMap[tid].setEntry(rid, phys_reg); 313 commitRenameMap[tid].setEntry(rid, phys_reg); 314 } 315 } else { 316 /* Initialize the vector-element interface */ 317 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 318 for (ElemIndex ldx = 0; ldx < TheISA::NumVecElemPerVecReg; 319 ++ldx) { 320 RegId lrid = RegId(VecElemClass, ridx, ldx); 321 PhysRegIdPtr phys_elem = freeList.getVecElem(); 322 renameMap[tid].setEntry(lrid, phys_elem); 323 commitRenameMap[tid].setEntry(lrid, phys_elem); 324 } 325 } 326 } 327 328 for (RegIndex ridx = 0; ridx < TheISA::NumCCRegs; ++ridx) { 329 PhysRegIdPtr phys_reg = freeList.getCCReg(); 330 renameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 331 commitRenameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 332 } 333 } 334 335 rename.setRenameMap(renameMap); 336 commit.setRenameMap(commitRenameMap); 337 rename.setFreeList(&freeList); 338 339 // Setup the ROB for whichever stages need it. 340 commit.setROB(&rob); 341 342 lastActivatedCycle = 0; 343#if 0 344 // Give renameMap & rename stage access to the freeList; 345 for (ThreadID tid = 0; tid < numThreads; tid++) 346 globalSeqNum[tid] = 1; 347#endif 348 349 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 350 351 // Setup any thread state. 352 this->thread.resize(this->numThreads); 353 354 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 355 if (FullSystem) { 356 // SMT is not supported in FS mode yet. 357 assert(this->numThreads == 1); 358 this->thread[tid] = new Thread(this, 0, NULL); 359 } else { 360 if (tid < params->workload.size()) { 361 DPRINTF(O3CPU, "Workload[%i] process is %#x", 362 tid, this->thread[tid]); 363 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 364 (typename Impl::O3CPU *)(this), 365 tid, params->workload[tid]); 366 367 //usedTids[tid] = true; 368 //threadMap[tid] = tid; 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[tid] = new typename FullO3CPU<Impl>::Thread( 375 (typename Impl::O3CPU *)(this), 376 tid, dummy_proc); 377 //usedTids[tid] = false; 378 } 379 } 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 (params->checker) { 391 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 392 o3_tc, this->checker); 393 } 394 395 o3_tc->cpu = (typename Impl::O3CPU *)(this); 396 assert(o3_tc->cpu); 397 o3_tc->thread = this->thread[tid]; 398 399 // Setup quiesce event. 400 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 401 402 // Give the thread the TC. 403 this->thread[tid]->tc = tc; 404 405 // Add the TC to the CPU's list of TC's. 406 this->threadContexts.push_back(tc); 407 } 408 409 // FullO3CPU always requires an interrupt controller. 410 if (!params->switched_out && interrupts.empty()) { 411 fatal("FullO3CPU %s has no interrupt controller.\n" 412 "Ensure createInterruptController() is called.\n", name()); 413 } 414 415 for (ThreadID tid = 0; tid < this->numThreads; tid++) 416 this->thread[tid]->setFuncExeInst(0); 417} 418 419template <class Impl> 420FullO3CPU<Impl>::~FullO3CPU() 421{ 422} 423 424template <class Impl> 425void 426FullO3CPU<Impl>::regProbePoints() 427{ 428 BaseCPU::regProbePoints(); 429 430 ppInstAccessComplete = new ProbePointArg<PacketPtr>(getProbeManager(), "InstAccessComplete"); 431 ppDataAccessComplete = new ProbePointArg<std::pair<DynInstPtr, PacketPtr> >(getProbeManager(), "DataAccessComplete"); 432 433 fetch.regProbePoints(); 434 rename.regProbePoints(); 435 iew.regProbePoints(); 436 commit.regProbePoints(); 437} 438 439template <class Impl> 440void 441FullO3CPU<Impl>::regStats() 442{ 443 BaseO3CPU::regStats(); 444 445 // Register any of the O3CPU's stats here. 446 timesIdled 447 .name(name() + ".timesIdled") 448 .desc("Number of times that the entire CPU went into an idle state and" 449 " unscheduled itself") 450 .prereq(timesIdled); 451 452 idleCycles 453 .name(name() + ".idleCycles") 454 .desc("Total number of cycles that the CPU has spent unscheduled due " 455 "to idling") 456 .prereq(idleCycles); 457 458 quiesceCycles 459 .name(name() + ".quiesceCycles") 460 .desc("Total number of cycles that CPU has spent quiesced or waiting " 461 "for an interrupt") 462 .prereq(quiesceCycles); 463 464 // Number of Instructions simulated 465 // -------------------------------- 466 // Should probably be in Base CPU but need templated 467 // MaxThreads so put in here instead 468 committedInsts 469 .init(numThreads) 470 .name(name() + ".committedInsts") 471 .desc("Number of Instructions Simulated") 472 .flags(Stats::total); 473 474 committedOps 475 .init(numThreads) 476 .name(name() + ".committedOps") 477 .desc("Number of Ops (including micro ops) Simulated") 478 .flags(Stats::total); 479 480 cpi 481 .name(name() + ".cpi") 482 .desc("CPI: Cycles Per Instruction") 483 .precision(6); 484 cpi = numCycles / committedInsts; 485 486 totalCpi 487 .name(name() + ".cpi_total") 488 .desc("CPI: Total CPI of All Threads") 489 .precision(6); 490 totalCpi = numCycles / sum(committedInsts); 491 492 ipc 493 .name(name() + ".ipc") 494 .desc("IPC: Instructions Per Cycle") 495 .precision(6); 496 ipc = committedInsts / numCycles; 497 498 totalIpc 499 .name(name() + ".ipc_total") 500 .desc("IPC: Total IPC of All Threads") 501 .precision(6); 502 totalIpc = sum(committedInsts) / numCycles; 503 504 this->fetch.regStats(); 505 this->decode.regStats(); 506 this->rename.regStats(); 507 this->iew.regStats(); 508 this->commit.regStats(); 509 this->rob.regStats(); 510 511 intRegfileReads 512 .name(name() + ".int_regfile_reads") 513 .desc("number of integer regfile reads") 514 .prereq(intRegfileReads); 515 516 intRegfileWrites 517 .name(name() + ".int_regfile_writes") 518 .desc("number of integer regfile writes") 519 .prereq(intRegfileWrites); 520 521 fpRegfileReads 522 .name(name() + ".fp_regfile_reads") 523 .desc("number of floating regfile reads") 524 .prereq(fpRegfileReads); 525 526 fpRegfileWrites 527 .name(name() + ".fp_regfile_writes") 528 .desc("number of floating regfile writes") 529 .prereq(fpRegfileWrites); 530 531 vecRegfileReads 532 .name(name() + ".vec_regfile_reads") 533 .desc("number of vector regfile reads") 534 .prereq(vecRegfileReads); 535 536 vecRegfileWrites 537 .name(name() + ".vec_regfile_writes") 538 .desc("number of vector regfile writes") 539 .prereq(vecRegfileWrites); 540 541 ccRegfileReads 542 .name(name() + ".cc_regfile_reads") 543 .desc("number of cc regfile reads") 544 .prereq(ccRegfileReads); 545 546 ccRegfileWrites 547 .name(name() + ".cc_regfile_writes") 548 .desc("number of cc regfile writes") 549 .prereq(ccRegfileWrites); 550 551 miscRegfileReads 552 .name(name() + ".misc_regfile_reads") 553 .desc("number of misc regfile reads") 554 .prereq(miscRegfileReads); 555 556 miscRegfileWrites 557 .name(name() + ".misc_regfile_writes") 558 .desc("number of misc regfile writes") 559 .prereq(miscRegfileWrites); 560} 561 562template <class Impl> 563void 564FullO3CPU<Impl>::tick() 565{ 566 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 567 assert(!switchedOut()); 568 assert(drainState() != DrainState::Drained); 569 570 ++numCycles; 571 ppCycles->notify(1); 572 573// activity = false; 574 575 //Tick each of the stages 576 fetch.tick(); 577 578 decode.tick(); 579 580 rename.tick(); 581 582 iew.tick(); 583 584 commit.tick(); 585 586 // Now advance the time buffers 587 timeBuffer.advance(); 588 589 fetchQueue.advance(); 590 decodeQueue.advance(); 591 renameQueue.advance(); 592 iewQueue.advance(); 593 594 activityRec.advance(); 595 596 if (removeInstsThisCycle) { 597 cleanUpRemovedInsts(); 598 } 599 600 if (!tickEvent.scheduled()) { 601 if (_status == SwitchedOut) { 602 DPRINTF(O3CPU, "Switched out!\n"); 603 // increment stat 604 lastRunningCycle = curCycle(); 605 } else if (!activityRec.active() || _status == Idle) { 606 DPRINTF(O3CPU, "Idle!\n"); 607 lastRunningCycle = curCycle(); 608 timesIdled++; 609 } else { 610 schedule(tickEvent, clockEdge(Cycles(1))); 611 DPRINTF(O3CPU, "Scheduling next tick!\n"); 612 } 613 } 614 615 if (!FullSystem) 616 updateThreadPriority(); 617 618 tryDrain(); 619} 620 621template <class Impl> 622void 623FullO3CPU<Impl>::init() 624{ 625 BaseCPU::init(); 626 627 for (ThreadID tid = 0; tid < numThreads; ++tid) { 628 // Set noSquashFromTC so that the CPU doesn't squash when initially 629 // setting up registers. 630 thread[tid]->noSquashFromTC = true; 631 // Initialise the ThreadContext's memory proxies 632 thread[tid]->initMemProxies(thread[tid]->getTC()); 633 } 634 635 if (FullSystem && !params()->switched_out) { 636 for (ThreadID tid = 0; tid < numThreads; tid++) { 637 ThreadContext *src_tc = threadContexts[tid]; 638 TheISA::initCPU(src_tc, src_tc->contextId()); 639 } 640 } 641 642 // Clear noSquashFromTC. 643 for (int tid = 0; tid < numThreads; ++tid) 644 thread[tid]->noSquashFromTC = false; 645 646 commit.setThreads(thread); 647} 648 649template <class Impl> 650void 651FullO3CPU<Impl>::startup() 652{ 653 BaseCPU::startup(); 654 for (int tid = 0; tid < numThreads; ++tid) 655 isa[tid]->startup(threadContexts[tid]); 656 657 fetch.startupStage(); 658 decode.startupStage(); 659 iew.startupStage(); 660 rename.startupStage(); 661 commit.startupStage(); 662} 663 664template <class Impl> 665void 666FullO3CPU<Impl>::activateThread(ThreadID tid) 667{ 668 list<ThreadID>::iterator isActive = 669 std::find(activeThreads.begin(), activeThreads.end(), tid); 670 671 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 672 assert(!switchedOut()); 673 674 if (isActive == activeThreads.end()) { 675 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 676 tid); 677 678 activeThreads.push_back(tid); 679 } 680} 681 682template <class Impl> 683void 684FullO3CPU<Impl>::deactivateThread(ThreadID tid) 685{ 686 //Remove From Active List, if Active 687 list<ThreadID>::iterator thread_it = 688 std::find(activeThreads.begin(), activeThreads.end(), tid); 689 690 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 691 assert(!switchedOut()); 692 693 if (thread_it != activeThreads.end()) { 694 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 695 tid); 696 activeThreads.erase(thread_it); 697 } 698 699 fetch.deactivateThread(tid); 700 commit.deactivateThread(tid); 701} 702 703template <class Impl> 704Counter 705FullO3CPU<Impl>::totalInsts() const 706{ 707 Counter total(0); 708 709 ThreadID size = thread.size(); 710 for (ThreadID i = 0; i < size; i++) 711 total += thread[i]->numInst; 712 713 return total; 714} 715 716template <class Impl> 717Counter 718FullO3CPU<Impl>::totalOps() const 719{ 720 Counter total(0); 721 722 ThreadID size = thread.size(); 723 for (ThreadID i = 0; i < size; i++) 724 total += thread[i]->numOp; 725 726 return total; 727} 728 729template <class Impl> 730void 731FullO3CPU<Impl>::activateContext(ThreadID tid) 732{ 733 assert(!switchedOut()); 734 735 // Needs to set each stage to running as well. 736 activateThread(tid); 737 738 // We don't want to wake the CPU if it is drained. In that case, 739 // we just want to flag the thread as active and schedule the tick 740 // event from drainResume() instead. 741 if (drainState() == DrainState::Drained) 742 return; 743 744 // If we are time 0 or if the last activation time is in the past, 745 // schedule the next tick and wake up the fetch unit 746 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 747 scheduleTickEvent(Cycles(0)); 748 749 // Be sure to signal that there's some activity so the CPU doesn't 750 // deschedule itself. 751 activityRec.activity(); 752 fetch.wakeFromQuiesce(); 753 754 Cycles cycles(curCycle() - lastRunningCycle); 755 // @todo: This is an oddity that is only here to match the stats 756 if (cycles != 0) 757 --cycles; 758 quiesceCycles += cycles; 759 760 lastActivatedCycle = curTick(); 761 762 _status = Running; 763 764 BaseCPU::activateContext(tid); 765 } 766} 767 768template <class Impl> 769void 770FullO3CPU<Impl>::suspendContext(ThreadID tid) 771{ 772 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 773 assert(!switchedOut()); 774 775 deactivateThread(tid); 776 777 // If this was the last thread then unschedule the tick event. 778 if (activeThreads.size() == 0) { 779 unscheduleTickEvent(); 780 lastRunningCycle = curCycle(); 781 _status = Idle; 782 } 783 784 DPRINTF(Quiesce, "Suspending Context\n"); 785 786 BaseCPU::suspendContext(tid); 787} 788 789template <class Impl> 790void 791FullO3CPU<Impl>::haltContext(ThreadID tid) 792{ 793 //For now, this is the same as deallocate 794 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 795 assert(!switchedOut()); 796 797 deactivateThread(tid); 798 removeThread(tid); 799} 800 801template <class Impl> 802void 803FullO3CPU<Impl>::insertThread(ThreadID tid) 804{ 805 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 806 // Will change now that the PC and thread state is internal to the CPU 807 // and not in the ThreadContext. 808 ThreadContext *src_tc; 809 if (FullSystem) 810 src_tc = system->threadContexts[tid]; 811 else 812 src_tc = tcBase(tid); 813 814 //Bind Int Regs to Rename Map 815 816 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 817 reg_id.index()++) { 818 PhysRegIdPtr phys_reg = freeList.getIntReg(); 819 renameMap[tid].setEntry(reg_id, phys_reg); 820 scoreboard.setReg(phys_reg); 821 } 822 823 //Bind Float Regs to Rename Map 824 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 825 reg_id.index()++) { 826 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 827 renameMap[tid].setEntry(reg_id, phys_reg); 828 scoreboard.setReg(phys_reg); 829 } 830 831 //Bind condition-code Regs to Rename Map 832 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 833 reg_id.index()++) { 834 PhysRegIdPtr phys_reg = freeList.getCCReg(); 835 renameMap[tid].setEntry(reg_id, phys_reg); 836 scoreboard.setReg(phys_reg); 837 } 838 839 //Copy Thread Data Into RegFile 840 //this->copyFromTC(tid); 841 842 //Set PC/NPC/NNPC 843 pcState(src_tc->pcState(), tid); 844 845 src_tc->setStatus(ThreadContext::Active); 846 847 activateContext(tid); 848 849 //Reset ROB/IQ/LSQ Entries 850 commit.rob->resetEntries(); 851 iew.resetEntries(); 852} 853 854template <class Impl> 855void 856FullO3CPU<Impl>::removeThread(ThreadID tid) 857{ 858 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 859 860 // Copy Thread Data From RegFile 861 // If thread is suspended, it might be re-allocated 862 // this->copyToTC(tid); 863 864 865 // @todo: 2-27-2008: Fix how we free up rename mappings 866 // here to alleviate the case for double-freeing registers 867 // in SMT workloads. 868 869 // Unbind Int Regs from Rename Map 870 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 871 reg_id.index()++) { 872 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 873 scoreboard.unsetReg(phys_reg); 874 freeList.addReg(phys_reg); 875 } 876 877 // Unbind Float Regs from Rename Map 878 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 879 reg_id.index()++) { 880 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 881 scoreboard.unsetReg(phys_reg); 882 freeList.addReg(phys_reg); 883 } 884 885 // Unbind condition-code Regs from Rename Map 886 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 887 reg_id.index()++) { 888 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 889 scoreboard.unsetReg(phys_reg); 890 freeList.addReg(phys_reg); 891 } 892 893 // Squash Throughout Pipeline 894 DynInstPtr inst = commit.rob->readHeadInst(tid); 895 InstSeqNum squash_seq_num = inst->seqNum; 896 fetch.squash(0, squash_seq_num, inst, tid); 897 decode.squash(tid); 898 rename.squash(squash_seq_num, tid); 899 iew.squash(tid); 900 iew.ldstQueue.squash(squash_seq_num, tid); 901 commit.rob->squash(squash_seq_num, tid); 902 903 904 assert(iew.instQueue.getCount(tid) == 0); 905 assert(iew.ldstQueue.getCount(tid) == 0); 906 907 // Reset ROB/IQ/LSQ Entries 908 909 // Commented out for now. This should be possible to do by 910 // telling all the pipeline stages to drain first, and then 911 // checking until the drain completes. Once the pipeline is 912 // drained, call resetEntries(). - 10-09-06 ktlim 913/* 914 if (activeThreads.size() >= 1) { 915 commit.rob->resetEntries(); 916 iew.resetEntries(); 917 } 918*/ 919} 920 921template <class Impl> 922Fault 923FullO3CPU<Impl>::hwrei(ThreadID tid) 924{ 925#if THE_ISA == ALPHA_ISA 926 // Need to clear the lock flag upon returning from an interrupt. 927 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 928 929 this->thread[tid]->kernelStats->hwrei(); 930 931 // FIXME: XXX check for interrupts? XXX 932#endif 933 return NoFault; 934} 935 936template <class Impl> 937bool 938FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 939{ 940#if THE_ISA == ALPHA_ISA 941 if (this->thread[tid]->kernelStats) 942 this->thread[tid]->kernelStats->callpal(palFunc, 943 this->threadContexts[tid]); 944 945 switch (palFunc) { 946 case PAL::halt: 947 halt(); 948 if (--System::numSystemsRunning == 0) 949 exitSimLoop("all cpus halted"); 950 break; 951 952 case PAL::bpt: 953 case PAL::bugchk: 954 if (this->system->breakpoint()) 955 return false; 956 break; 957 } 958#endif 959 return true; 960} 961 962template <class Impl> 963Fault 964FullO3CPU<Impl>::getInterrupts() 965{ 966 // Check if there are any outstanding interrupts 967 return this->interrupts[0]->getInterrupt(this->threadContexts[0]); 968} 969 970template <class Impl> 971void 972FullO3CPU<Impl>::processInterrupts(const Fault &interrupt) 973{ 974 // Check for interrupts here. For now can copy the code that 975 // exists within isa_fullsys_traits.hh. Also assume that thread 0 976 // is the one that handles the interrupts. 977 // @todo: Possibly consolidate the interrupt checking code. 978 // @todo: Allow other threads to handle interrupts. 979 980 assert(interrupt != NoFault); 981 this->interrupts[0]->updateIntrInfo(this->threadContexts[0]); 982 983 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 984 this->trap(interrupt, 0, nullptr); 985} 986 987template <class Impl> 988void 989FullO3CPU<Impl>::trap(const Fault &fault, ThreadID tid, 990 const StaticInstPtr &inst) 991{ 992 // Pass the thread's TC into the invoke method. 993 fault->invoke(this->threadContexts[tid], inst); 994} 995 996template <class Impl> 997void 998FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid, Fault *fault) 999{ 1000 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1001 1002 DPRINTF(Activity,"Activity: syscall() called.\n"); 1003 1004 // Temporarily increase this by one to account for the syscall 1005 // instruction. 1006 ++(this->thread[tid]->funcExeInst); 1007 1008 // Execute the actual syscall. 1009 this->thread[tid]->syscall(callnum, fault); 1010 1011 // Decrease funcExeInst by one as the normal commit will handle 1012 // incrementing it. 1013 --(this->thread[tid]->funcExeInst); 1014} 1015 1016template <class Impl> 1017void 1018FullO3CPU<Impl>::serializeThread(CheckpointOut &cp, ThreadID tid) const 1019{ 1020 thread[tid]->serialize(cp); 1021} 1022 1023template <class Impl> 1024void 1025FullO3CPU<Impl>::unserializeThread(CheckpointIn &cp, ThreadID tid) 1026{ 1027 thread[tid]->unserialize(cp); 1028} 1029 1030template <class Impl> 1031DrainState 1032FullO3CPU<Impl>::drain() 1033{ 1034 // Deschedule any power gating event (if any) 1035 deschedulePowerGatingEvent(); 1036 1037 // If the CPU isn't doing anything, then return immediately. 1038 if (switchedOut()) 1039 return DrainState::Drained; 1040 1041 DPRINTF(Drain, "Draining...\n"); 1042 1043 // We only need to signal a drain to the commit stage as this 1044 // initiates squashing controls the draining. Once the commit 1045 // stage commits an instruction where it is safe to stop, it'll 1046 // squash the rest of the instructions in the pipeline and force 1047 // the fetch stage to stall. The pipeline will be drained once all 1048 // in-flight instructions have retired. 1049 commit.drain(); 1050 1051 // Wake the CPU and record activity so everything can drain out if 1052 // the CPU was not able to immediately drain. 1053 if (!isDrained()) { 1054 // If a thread is suspended, wake it up so it can be drained 1055 for (auto t : threadContexts) { 1056 if (t->status() == ThreadContext::Suspended){ 1057 DPRINTF(Drain, "Currently suspended so activate %i \n", 1058 t->threadId()); 1059 t->activate(); 1060 // As the thread is now active, change the power state as well 1061 activateContext(t->threadId()); 1062 } 1063 } 1064 1065 wakeCPU(); 1066 activityRec.activity(); 1067 1068 DPRINTF(Drain, "CPU not drained\n"); 1069 1070 return DrainState::Draining; 1071 } else { 1072 DPRINTF(Drain, "CPU is already drained\n"); 1073 if (tickEvent.scheduled()) 1074 deschedule(tickEvent); 1075 1076 // Flush out any old data from the time buffers. In 1077 // particular, there might be some data in flight from the 1078 // fetch stage that isn't visible in any of the CPU buffers we 1079 // test in isDrained(). 1080 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1081 timeBuffer.advance(); 1082 fetchQueue.advance(); 1083 decodeQueue.advance(); 1084 renameQueue.advance(); 1085 iewQueue.advance(); 1086 } 1087 1088 drainSanityCheck(); 1089 return DrainState::Drained; 1090 } 1091} 1092 1093template <class Impl> 1094bool 1095FullO3CPU<Impl>::tryDrain() 1096{ 1097 if (drainState() != DrainState::Draining || !isDrained()) 1098 return false; 1099 1100 if (tickEvent.scheduled()) 1101 deschedule(tickEvent); 1102 1103 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1104 signalDrainDone(); 1105 1106 return true; 1107} 1108 1109template <class Impl> 1110void 1111FullO3CPU<Impl>::drainSanityCheck() const 1112{ 1113 assert(isDrained()); 1114 fetch.drainSanityCheck(); 1115 decode.drainSanityCheck(); 1116 rename.drainSanityCheck(); 1117 iew.drainSanityCheck(); 1118 commit.drainSanityCheck(); 1119} 1120 1121template <class Impl> 1122bool 1123FullO3CPU<Impl>::isDrained() const 1124{ 1125 bool drained(true); 1126 1127 if (!instList.empty() || !removeList.empty()) { 1128 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1129 drained = false; 1130 } 1131 1132 if (!fetch.isDrained()) { 1133 DPRINTF(Drain, "Fetch not drained.\n"); 1134 drained = false; 1135 } 1136 1137 if (!decode.isDrained()) { 1138 DPRINTF(Drain, "Decode not drained.\n"); 1139 drained = false; 1140 } 1141 1142 if (!rename.isDrained()) { 1143 DPRINTF(Drain, "Rename not drained.\n"); 1144 drained = false; 1145 } 1146 1147 if (!iew.isDrained()) { 1148 DPRINTF(Drain, "IEW not drained.\n"); 1149 drained = false; 1150 } 1151 1152 if (!commit.isDrained()) { 1153 DPRINTF(Drain, "Commit not drained.\n"); 1154 drained = false; 1155 } 1156 1157 return drained; 1158} 1159 1160template <class Impl> 1161void 1162FullO3CPU<Impl>::commitDrained(ThreadID tid) 1163{ 1164 fetch.drainStall(tid); 1165} 1166 1167template <class Impl> 1168void 1169FullO3CPU<Impl>::drainResume() 1170{ 1171 if (switchedOut()) 1172 return; 1173 1174 DPRINTF(Drain, "Resuming...\n"); 1175 verifyMemoryMode(); 1176 1177 fetch.drainResume(); 1178 commit.drainResume(); 1179 1180 _status = Idle; 1181 for (ThreadID i = 0; i < thread.size(); i++) { 1182 if (thread[i]->status() == ThreadContext::Active) { 1183 DPRINTF(Drain, "Activating thread: %i\n", i); 1184 activateThread(i); 1185 _status = Running; 1186 } 1187 } 1188 1189 assert(!tickEvent.scheduled()); 1190 if (_status == Running) 1191 schedule(tickEvent, nextCycle()); 1192 1193 // Reschedule any power gating event (if any) 1194 schedulePowerGatingEvent(); 1195} 1196 1197template <class Impl> 1198void 1199FullO3CPU<Impl>::switchOut() 1200{ 1201 DPRINTF(O3CPU, "Switching out\n"); 1202 BaseCPU::switchOut(); 1203 1204 activityRec.reset(); 1205 1206 _status = SwitchedOut; 1207 1208 if (checker) 1209 checker->switchOut(); 1210} 1211 1212template <class Impl> 1213void 1214FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1215{ 1216 BaseCPU::takeOverFrom(oldCPU); 1217 1218 fetch.takeOverFrom(); 1219 decode.takeOverFrom(); 1220 rename.takeOverFrom(); 1221 iew.takeOverFrom(); 1222 commit.takeOverFrom(); 1223 1224 assert(!tickEvent.scheduled()); 1225 1226 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1227 if (oldO3CPU) 1228 globalSeqNum = oldO3CPU->globalSeqNum; 1229 1230 lastRunningCycle = curCycle(); 1231 _status = Idle; 1232} 1233 1234template <class Impl> 1235void 1236FullO3CPU<Impl>::verifyMemoryMode() const 1237{ 1238 if (!system->isTimingMode()) { 1239 fatal("The O3 CPU requires the memory system to be in " 1240 "'timing' mode.\n"); 1241 } 1242} 1243 1244template <class Impl> 1245TheISA::MiscReg 1246FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) const 1247{ 1248 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1249} 1250 1251template <class Impl> 1252TheISA::MiscReg 1253FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1254{ 1255 miscRegfileReads++; 1256 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1257} 1258 1259template <class Impl> 1260void 1261FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1262 const TheISA::MiscReg &val, ThreadID tid) 1263{ 1264 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1265} 1266 1267template <class Impl> 1268void 1269FullO3CPU<Impl>::setMiscReg(int misc_reg, 1270 const TheISA::MiscReg &val, ThreadID tid) 1271{ 1272 miscRegfileWrites++; 1273 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1274} 1275 1276template <class Impl> 1277uint64_t 1278FullO3CPU<Impl>::readIntReg(PhysRegIdPtr phys_reg) 1279{ 1280 intRegfileReads++; 1281 return regFile.readIntReg(phys_reg); 1282} 1283 1284template <class Impl> 1285FloatReg 1286FullO3CPU<Impl>::readFloatReg(PhysRegIdPtr phys_reg) 1287{ 1288 fpRegfileReads++; 1289 return regFile.readFloatReg(phys_reg); 1290} 1291 1292template <class Impl> 1293FloatRegBits 1294FullO3CPU<Impl>::readFloatRegBits(PhysRegIdPtr phys_reg) 1295{ 1296 fpRegfileReads++; 1297 return regFile.readFloatRegBits(phys_reg); 1298} 1299 1300template <class Impl> 1301auto 1302FullO3CPU<Impl>::readVecReg(PhysRegIdPtr phys_reg) const 1303 -> const VecRegContainer& 1304{ 1305 vecRegfileReads++; 1306 return regFile.readVecReg(phys_reg); 1307} 1308 1309template <class Impl> 1310auto 1311FullO3CPU<Impl>::getWritableVecReg(PhysRegIdPtr phys_reg) 1312 -> VecRegContainer& 1313{ 1314 vecRegfileWrites++; 1315 return regFile.getWritableVecReg(phys_reg); 1316} 1317 1318template <class Impl> 1319auto 1320FullO3CPU<Impl>::readVecElem(PhysRegIdPtr phys_reg) const -> const VecElem& 1321{ 1322 vecRegfileReads++; 1323 return regFile.readVecElem(phys_reg); 1324} 1325 1326template <class Impl> 1327CCReg 1328FullO3CPU<Impl>::readCCReg(PhysRegIdPtr phys_reg) 1329{ 1330 ccRegfileReads++; 1331 return regFile.readCCReg(phys_reg); 1332} 1333 1334template <class Impl> 1335void 1336FullO3CPU<Impl>::setIntReg(PhysRegIdPtr phys_reg, uint64_t val) 1337{ 1338 intRegfileWrites++; 1339 regFile.setIntReg(phys_reg, val); 1340} 1341 1342template <class Impl> 1343void 1344FullO3CPU<Impl>::setFloatReg(PhysRegIdPtr phys_reg, FloatReg val) 1345{ 1346 fpRegfileWrites++; 1347 regFile.setFloatReg(phys_reg, val); 1348} 1349 1350template <class Impl> 1351void 1352FullO3CPU<Impl>::setFloatRegBits(PhysRegIdPtr phys_reg, FloatRegBits val) 1353{ 1354 fpRegfileWrites++; 1355 regFile.setFloatRegBits(phys_reg, val); 1356} 1357 1358template <class Impl> 1359void 1360FullO3CPU<Impl>::setVecReg(PhysRegIdPtr phys_reg, const VecRegContainer& val) 1361{ 1362 vecRegfileWrites++; 1363 regFile.setVecReg(phys_reg, val); 1364} 1365 1366template <class Impl> 1367void 1368FullO3CPU<Impl>::setVecElem(PhysRegIdPtr phys_reg, const VecElem& val) 1369{ 1370 vecRegfileWrites++; 1371 regFile.setVecElem(phys_reg, val); 1372} 1373 1374template <class Impl> 1375void 1376FullO3CPU<Impl>::setCCReg(PhysRegIdPtr phys_reg, CCReg val) 1377{ 1378 ccRegfileWrites++; 1379 regFile.setCCReg(phys_reg, val); 1380} 1381 1382template <class Impl> 1383uint64_t 1384FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1385{ 1386 intRegfileReads++; 1387 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1388 RegId(IntRegClass, reg_idx)); 1389 1390 return regFile.readIntReg(phys_reg); 1391} 1392 1393template <class Impl> 1394float 1395FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1396{ 1397 fpRegfileReads++; 1398 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1399 RegId(FloatRegClass, reg_idx)); 1400 1401 return regFile.readFloatReg(phys_reg); 1402} 1403 1404template <class Impl> 1405uint64_t 1406FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1407{ 1408 fpRegfileReads++; 1409 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1410 RegId(FloatRegClass, reg_idx)); 1411 1412 return regFile.readFloatRegBits(phys_reg); 1413} 1414 1415template <class Impl> 1416auto 1417FullO3CPU<Impl>::readArchVecReg(int reg_idx, ThreadID tid) const 1418 -> const VecRegContainer& 1419{ 1420 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1421 RegId(VecRegClass, reg_idx)); 1422 return readVecReg(phys_reg); 1423} 1424 1425template <class Impl> 1426auto 1427FullO3CPU<Impl>::getWritableArchVecReg(int reg_idx, ThreadID tid) 1428 -> VecRegContainer& 1429{ 1430 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1431 RegId(VecRegClass, reg_idx)); 1432 return getWritableVecReg(phys_reg); 1433} 1434 1435template <class Impl> 1436auto 1437FullO3CPU<Impl>::readArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1438 ThreadID tid) const -> const VecElem& 1439{ 1440 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1441 RegId(VecRegClass, reg_idx, ldx)); 1442 return readVecElem(phys_reg); 1443} 1444 1445template <class Impl> 1446CCReg 1447FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1448{ 1449 ccRegfileReads++; 1450 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1451 RegId(CCRegClass, reg_idx)); 1452 1453 return regFile.readCCReg(phys_reg); 1454} 1455 1456template <class Impl> 1457void 1458FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1459{ 1460 intRegfileWrites++; 1461 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1462 RegId(IntRegClass, reg_idx)); 1463 1464 regFile.setIntReg(phys_reg, val); 1465} 1466 1467template <class Impl> 1468void 1469FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1470{ 1471 fpRegfileWrites++; 1472 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1473 RegId(FloatRegClass, reg_idx)); 1474 1475 regFile.setFloatReg(phys_reg, val); 1476} 1477 1478template <class Impl> 1479void 1480FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1481{ 1482 fpRegfileWrites++; 1483 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1484 RegId(FloatRegClass, reg_idx)); 1485 1486 regFile.setFloatRegBits(phys_reg, val); 1487} 1488 1489template <class Impl> 1490void 1491FullO3CPU<Impl>::setArchVecReg(int reg_idx, const VecRegContainer& val, 1492 ThreadID tid) 1493{ 1494 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1495 RegId(VecRegClass, reg_idx)); 1496 setVecReg(phys_reg, val); 1497} 1498 1499template <class Impl> 1500void 1501FullO3CPU<Impl>::setArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1502 const VecElem& val, ThreadID tid) 1503{ 1504 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1505 RegId(VecRegClass, reg_idx, ldx)); 1506 setVecElem(phys_reg, val); 1507} 1508 1509template <class Impl> 1510void 1511FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1512{ 1513 ccRegfileWrites++; 1514 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1515 RegId(CCRegClass, reg_idx)); 1516 1517 regFile.setCCReg(phys_reg, val); 1518} 1519 1520template <class Impl> 1521TheISA::PCState 1522FullO3CPU<Impl>::pcState(ThreadID tid) 1523{ 1524 return commit.pcState(tid); 1525} 1526 1527template <class Impl> 1528void 1529FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1530{ 1531 commit.pcState(val, tid); 1532} 1533 1534template <class Impl> 1535Addr 1536FullO3CPU<Impl>::instAddr(ThreadID tid) 1537{ 1538 return commit.instAddr(tid); 1539} 1540 1541template <class Impl> 1542Addr 1543FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1544{ 1545 return commit.nextInstAddr(tid); 1546} 1547 1548template <class Impl> 1549MicroPC 1550FullO3CPU<Impl>::microPC(ThreadID tid) 1551{ 1552 return commit.microPC(tid); 1553} 1554 1555template <class Impl> 1556void 1557FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1558{ 1559 this->thread[tid]->noSquashFromTC = true; 1560 this->commit.generateTCEvent(tid); 1561} 1562 1563template <class Impl> 1564typename FullO3CPU<Impl>::ListIt 1565FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1566{ 1567 instList.push_back(inst); 1568 1569 return --(instList.end()); 1570} 1571 1572template <class Impl> 1573void 1574FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1575{ 1576 // Keep an instruction count. 1577 if (!inst->isMicroop() || inst->isLastMicroop()) { 1578 thread[tid]->numInst++; 1579 thread[tid]->numInsts++; 1580 committedInsts[tid]++; 1581 system->totalNumInsts++; 1582 1583 // Check for instruction-count-based events. 1584 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1585 system->instEventQueue.serviceEvents(system->totalNumInsts); 1586 } 1587 thread[tid]->numOp++; 1588 thread[tid]->numOps++; 1589 committedOps[tid]++; 1590 1591 probeInstCommit(inst->staticInst); 1592} 1593 1594template <class Impl> 1595void 1596FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1597{ 1598 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1599 "[sn:%lli]\n", 1600 inst->threadNumber, inst->pcState(), inst->seqNum); 1601 1602 removeInstsThisCycle = true; 1603 1604 // Remove the front instruction. 1605 removeList.push(inst->getInstListIt()); 1606} 1607 1608template <class Impl> 1609void 1610FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1611{ 1612 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1613 " list.\n", tid); 1614 1615 ListIt end_it; 1616 1617 bool rob_empty = false; 1618 1619 if (instList.empty()) { 1620 return; 1621 } else if (rob.isEmpty(tid)) { 1622 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1623 end_it = instList.begin(); 1624 rob_empty = true; 1625 } else { 1626 end_it = (rob.readTailInst(tid))->getInstListIt(); 1627 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1628 } 1629 1630 removeInstsThisCycle = true; 1631 1632 ListIt inst_it = instList.end(); 1633 1634 inst_it--; 1635 1636 // Walk through the instruction list, removing any instructions 1637 // that were inserted after the given instruction iterator, end_it. 1638 while (inst_it != end_it) { 1639 assert(!instList.empty()); 1640 1641 squashInstIt(inst_it, tid); 1642 1643 inst_it--; 1644 } 1645 1646 // If the ROB was empty, then we actually need to remove the first 1647 // instruction as well. 1648 if (rob_empty) { 1649 squashInstIt(inst_it, tid); 1650 } 1651} 1652 1653template <class Impl> 1654void 1655FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1656{ 1657 assert(!instList.empty()); 1658 1659 removeInstsThisCycle = true; 1660 1661 ListIt inst_iter = instList.end(); 1662 1663 inst_iter--; 1664 1665 DPRINTF(O3CPU, "Deleting instructions from instruction " 1666 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1667 tid, seq_num, (*inst_iter)->seqNum); 1668 1669 while ((*inst_iter)->seqNum > seq_num) { 1670 1671 bool break_loop = (inst_iter == instList.begin()); 1672 1673 squashInstIt(inst_iter, tid); 1674 1675 inst_iter--; 1676 1677 if (break_loop) 1678 break; 1679 } 1680} 1681 1682template <class Impl> 1683inline void 1684FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1685{ 1686 if ((*instIt)->threadNumber == tid) { 1687 DPRINTF(O3CPU, "Squashing instruction, " 1688 "[tid:%i] [sn:%lli] PC %s\n", 1689 (*instIt)->threadNumber, 1690 (*instIt)->seqNum, 1691 (*instIt)->pcState()); 1692 1693 // Mark it as squashed. 1694 (*instIt)->setSquashed(); 1695 1696 // @todo: Formulate a consistent method for deleting 1697 // instructions from the instruction list 1698 // Remove the instruction from the list. 1699 removeList.push(instIt); 1700 } 1701} 1702 1703template <class Impl> 1704void 1705FullO3CPU<Impl>::cleanUpRemovedInsts() 1706{ 1707 while (!removeList.empty()) { 1708 DPRINTF(O3CPU, "Removing instruction, " 1709 "[tid:%i] [sn:%lli] PC %s\n", 1710 (*removeList.front())->threadNumber, 1711 (*removeList.front())->seqNum, 1712 (*removeList.front())->pcState()); 1713 1714 instList.erase(removeList.front()); 1715 1716 removeList.pop(); 1717 } 1718 1719 removeInstsThisCycle = false; 1720} 1721/* 1722template <class Impl> 1723void 1724FullO3CPU<Impl>::removeAllInsts() 1725{ 1726 instList.clear(); 1727} 1728*/ 1729template <class Impl> 1730void 1731FullO3CPU<Impl>::dumpInsts() 1732{ 1733 int num = 0; 1734 1735 ListIt inst_list_it = instList.begin(); 1736 1737 cprintf("Dumping Instruction List\n"); 1738 1739 while (inst_list_it != instList.end()) { 1740 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1741 "Squashed:%i\n\n", 1742 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1743 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1744 (*inst_list_it)->isSquashed()); 1745 inst_list_it++; 1746 ++num; 1747 } 1748} 1749/* 1750template <class Impl> 1751void 1752FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1753{ 1754 iew.wakeDependents(inst); 1755} 1756*/ 1757template <class Impl> 1758void 1759FullO3CPU<Impl>::wakeCPU() 1760{ 1761 if (activityRec.active() || tickEvent.scheduled()) { 1762 DPRINTF(Activity, "CPU already running.\n"); 1763 return; 1764 } 1765 1766 DPRINTF(Activity, "Waking up CPU\n"); 1767 1768 Cycles cycles(curCycle() - lastRunningCycle); 1769 // @todo: This is an oddity that is only here to match the stats 1770 if (cycles > 1) { 1771 --cycles; 1772 idleCycles += cycles; 1773 numCycles += cycles; 1774 ppCycles->notify(cycles); 1775 } 1776 1777 schedule(tickEvent, clockEdge()); 1778} 1779 1780template <class Impl> 1781void 1782FullO3CPU<Impl>::wakeup(ThreadID tid) 1783{ 1784 if (this->thread[tid]->status() != ThreadContext::Suspended) 1785 return; 1786 1787 this->wakeCPU(); 1788 1789 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1790 this->threadContexts[tid]->activate(); 1791} 1792 1793template <class Impl> 1794ThreadID 1795FullO3CPU<Impl>::getFreeTid() 1796{ 1797 for (ThreadID tid = 0; tid < numThreads; tid++) { 1798 if (!tids[tid]) { 1799 tids[tid] = true; 1800 return tid; 1801 } 1802 } 1803 1804 return InvalidThreadID; 1805} 1806 1807template <class Impl> 1808void 1809FullO3CPU<Impl>::updateThreadPriority() 1810{ 1811 if (activeThreads.size() > 1) { 1812 //DEFAULT TO ROUND ROBIN SCHEME 1813 //e.g. Move highest priority to end of thread list 1814 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1815 1816 unsigned high_thread = *list_begin; 1817 1818 activeThreads.erase(list_begin); 1819 1820 activeThreads.push_back(high_thread); 1821 } 1822} 1823 1824// Forward declaration of FullO3CPU. 1825template class FullO3CPU<O3CPUImpl>; 1826