cpu.cc revision 13610:5d5404ac6288
1/* 2 * Copyright (c) 2011-2012, 2014, 2016, 2017, 2019 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(RenameMode<TheISA::ISA>::init(params->isa[0])), 159 regFile(params->numPhysIntRegs, 160 params->numPhysFloatRegs, 161 params->numPhysVecRegs, 162 params->numPhysVecPredRegs, 163 params->numPhysCCRegs, 164 vecMode), 165 166 freeList(name() + ".freelist", ®File), 167 168 rob(this, params), 169 170 scoreboard(name() + ".scoreboard", 171 regFile.totalNumPhysRegs()), 172 173 isa(numThreads, NULL), 174 175 icachePort(&fetch, this), 176 dcachePort(&iew.ldstQueue, this), 177 178 timeBuffer(params->backComSize, params->forwardComSize), 179 fetchQueue(params->backComSize, params->forwardComSize), 180 decodeQueue(params->backComSize, params->forwardComSize), 181 renameQueue(params->backComSize, params->forwardComSize), 182 iewQueue(params->backComSize, params->forwardComSize), 183 activityRec(name(), NumStages, 184 params->backComSize + params->forwardComSize, 185 params->activity), 186 187 globalSeqNum(1), 188 system(params->system), 189 lastRunningCycle(curCycle()) 190{ 191 if (!params->switched_out) { 192 _status = Running; 193 } else { 194 _status = SwitchedOut; 195 } 196 197 if (params->checker) { 198 BaseCPU *temp_checker = params->checker; 199 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 200 checker->setIcachePort(&icachePort); 201 checker->setSystem(params->system); 202 } else { 203 checker = NULL; 204 } 205 206 if (!FullSystem) { 207 thread.resize(numThreads); 208 tids.resize(numThreads); 209 } 210 211 // The stages also need their CPU pointer setup. However this 212 // must be done at the upper level CPU because they have pointers 213 // to the upper level CPU, and not this FullO3CPU. 214 215 // Set up Pointers to the activeThreads list for each stage 216 fetch.setActiveThreads(&activeThreads); 217 decode.setActiveThreads(&activeThreads); 218 rename.setActiveThreads(&activeThreads); 219 iew.setActiveThreads(&activeThreads); 220 commit.setActiveThreads(&activeThreads); 221 222 // Give each of the stages the time buffer they will use. 223 fetch.setTimeBuffer(&timeBuffer); 224 decode.setTimeBuffer(&timeBuffer); 225 rename.setTimeBuffer(&timeBuffer); 226 iew.setTimeBuffer(&timeBuffer); 227 commit.setTimeBuffer(&timeBuffer); 228 229 // Also setup each of the stages' queues. 230 fetch.setFetchQueue(&fetchQueue); 231 decode.setFetchQueue(&fetchQueue); 232 commit.setFetchQueue(&fetchQueue); 233 decode.setDecodeQueue(&decodeQueue); 234 rename.setDecodeQueue(&decodeQueue); 235 rename.setRenameQueue(&renameQueue); 236 iew.setRenameQueue(&renameQueue); 237 iew.setIEWQueue(&iewQueue); 238 commit.setIEWQueue(&iewQueue); 239 commit.setRenameQueue(&renameQueue); 240 241 commit.setIEWStage(&iew); 242 rename.setIEWStage(&iew); 243 rename.setCommitStage(&commit); 244 245 ThreadID active_threads; 246 if (FullSystem) { 247 active_threads = 1; 248 } else { 249 active_threads = params->workload.size(); 250 251 if (active_threads > Impl::MaxThreads) { 252 panic("Workload Size too large. Increase the 'MaxThreads' " 253 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 254 "or edit your workload size."); 255 } 256 } 257 258 //Make Sure That this a Valid Architeture 259 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 260 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 261 assert(params->numPhysVecRegs >= numThreads * TheISA::NumVecRegs); 262 assert(params->numPhysVecPredRegs >= numThreads * TheISA::NumVecPredRegs); 263 assert(params->numPhysCCRegs >= numThreads * TheISA::NumCCRegs); 264 265 rename.setScoreboard(&scoreboard); 266 iew.setScoreboard(&scoreboard); 267 268 // Setup the rename map for whichever stages need it. 269 for (ThreadID tid = 0; tid < numThreads; tid++) { 270 isa[tid] = params->isa[tid]; 271 assert(RenameMode<TheISA::ISA>::equalsInit(isa[tid], isa[0])); 272 273 // Only Alpha has an FP zero register, so for other ISAs we 274 // use an invalid FP register index to avoid special treatment 275 // of any valid FP reg. 276 RegIndex invalidFPReg = TheISA::NumFloatRegs + 1; 277 RegIndex fpZeroReg = 278 (THE_ISA == ALPHA_ISA) ? TheISA::ZeroReg : invalidFPReg; 279 280 commitRenameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 281 &freeList, 282 vecMode); 283 284 renameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 285 &freeList, vecMode); 286 } 287 288 // Initialize rename map to assign physical registers to the 289 // architectural registers for active threads only. 290 for (ThreadID tid = 0; tid < active_threads; tid++) { 291 for (RegIndex ridx = 0; ridx < TheISA::NumIntRegs; ++ridx) { 292 // Note that we can't use the rename() method because we don't 293 // want special treatment for the zero register at this point 294 PhysRegIdPtr phys_reg = freeList.getIntReg(); 295 renameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 296 commitRenameMap[tid].setEntry(RegId(IntRegClass, ridx), phys_reg); 297 } 298 299 for (RegIndex ridx = 0; ridx < TheISA::NumFloatRegs; ++ridx) { 300 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 301 renameMap[tid].setEntry(RegId(FloatRegClass, ridx), phys_reg); 302 commitRenameMap[tid].setEntry( 303 RegId(FloatRegClass, ridx), phys_reg); 304 } 305 306 /* Here we need two 'interfaces' the 'whole register' and the 307 * 'register element'. At any point only one of them will be 308 * active. */ 309 if (vecMode == Enums::Full) { 310 /* Initialize the full-vector interface */ 311 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 312 RegId rid = RegId(VecRegClass, ridx); 313 PhysRegIdPtr phys_reg = freeList.getVecReg(); 314 renameMap[tid].setEntry(rid, phys_reg); 315 commitRenameMap[tid].setEntry(rid, phys_reg); 316 } 317 } else { 318 /* Initialize the vector-element interface */ 319 for (RegIndex ridx = 0; ridx < TheISA::NumVecRegs; ++ridx) { 320 for (ElemIndex ldx = 0; ldx < TheISA::NumVecElemPerVecReg; 321 ++ldx) { 322 RegId lrid = RegId(VecElemClass, ridx, ldx); 323 PhysRegIdPtr phys_elem = freeList.getVecElem(); 324 renameMap[tid].setEntry(lrid, phys_elem); 325 commitRenameMap[tid].setEntry(lrid, phys_elem); 326 } 327 } 328 } 329 330 for (RegIndex ridx = 0; ridx < TheISA::NumVecPredRegs; ++ridx) { 331 PhysRegIdPtr phys_reg = freeList.getVecPredReg(); 332 renameMap[tid].setEntry(RegId(VecPredRegClass, ridx), phys_reg); 333 commitRenameMap[tid].setEntry( 334 RegId(VecPredRegClass, ridx), phys_reg); 335 } 336 337 for (RegIndex ridx = 0; ridx < TheISA::NumCCRegs; ++ridx) { 338 PhysRegIdPtr phys_reg = freeList.getCCReg(); 339 renameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 340 commitRenameMap[tid].setEntry(RegId(CCRegClass, ridx), phys_reg); 341 } 342 } 343 344 rename.setRenameMap(renameMap); 345 commit.setRenameMap(commitRenameMap); 346 rename.setFreeList(&freeList); 347 348 // Setup the ROB for whichever stages need it. 349 commit.setROB(&rob); 350 351 lastActivatedCycle = 0; 352#if 0 353 // Give renameMap & rename stage access to the freeList; 354 for (ThreadID tid = 0; tid < numThreads; tid++) 355 globalSeqNum[tid] = 1; 356#endif 357 358 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 359 360 // Setup any thread state. 361 this->thread.resize(this->numThreads); 362 363 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 364 if (FullSystem) { 365 // SMT is not supported in FS mode yet. 366 assert(this->numThreads == 1); 367 this->thread[tid] = new Thread(this, 0, NULL); 368 } else { 369 if (tid < params->workload.size()) { 370 DPRINTF(O3CPU, "Workload[%i] process is %#x", 371 tid, this->thread[tid]); 372 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 373 (typename Impl::O3CPU *)(this), 374 tid, params->workload[tid]); 375 376 //usedTids[tid] = true; 377 //threadMap[tid] = tid; 378 } else { 379 //Allocate Empty thread so M5 can use later 380 //when scheduling threads to CPU 381 Process* dummy_proc = NULL; 382 383 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 384 (typename Impl::O3CPU *)(this), 385 tid, dummy_proc); 386 //usedTids[tid] = false; 387 } 388 } 389 390 ThreadContext *tc; 391 392 // Setup the TC that will serve as the interface to the threads/CPU. 393 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 394 395 tc = o3_tc; 396 397 // If we're using a checker, then the TC should be the 398 // CheckerThreadContext. 399 if (params->checker) { 400 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 401 o3_tc, this->checker); 402 } 403 404 o3_tc->cpu = (typename Impl::O3CPU *)(this); 405 assert(o3_tc->cpu); 406 o3_tc->thread = this->thread[tid]; 407 408 // Setup quiesce event. 409 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 410 411 // Give the thread the TC. 412 this->thread[tid]->tc = tc; 413 414 // Add the TC to the CPU's list of TC's. 415 this->threadContexts.push_back(tc); 416 } 417 418 // FullO3CPU always requires an interrupt controller. 419 if (!params->switched_out && interrupts.empty()) { 420 fatal("FullO3CPU %s has no interrupt controller.\n" 421 "Ensure createInterruptController() is called.\n", name()); 422 } 423 424 for (ThreadID tid = 0; tid < this->numThreads; tid++) 425 this->thread[tid]->setFuncExeInst(0); 426} 427 428template <class Impl> 429FullO3CPU<Impl>::~FullO3CPU() 430{ 431} 432 433template <class Impl> 434void 435FullO3CPU<Impl>::regProbePoints() 436{ 437 BaseCPU::regProbePoints(); 438 439 ppInstAccessComplete = new ProbePointArg<PacketPtr>(getProbeManager(), "InstAccessComplete"); 440 ppDataAccessComplete = new ProbePointArg<std::pair<DynInstPtr, PacketPtr> >(getProbeManager(), "DataAccessComplete"); 441 442 fetch.regProbePoints(); 443 rename.regProbePoints(); 444 iew.regProbePoints(); 445 commit.regProbePoints(); 446} 447 448template <class Impl> 449void 450FullO3CPU<Impl>::regStats() 451{ 452 BaseO3CPU::regStats(); 453 454 // Register any of the O3CPU's stats here. 455 timesIdled 456 .name(name() + ".timesIdled") 457 .desc("Number of times that the entire CPU went into an idle state and" 458 " unscheduled itself") 459 .prereq(timesIdled); 460 461 idleCycles 462 .name(name() + ".idleCycles") 463 .desc("Total number of cycles that the CPU has spent unscheduled due " 464 "to idling") 465 .prereq(idleCycles); 466 467 quiesceCycles 468 .name(name() + ".quiesceCycles") 469 .desc("Total number of cycles that CPU has spent quiesced or waiting " 470 "for an interrupt") 471 .prereq(quiesceCycles); 472 473 // Number of Instructions simulated 474 // -------------------------------- 475 // Should probably be in Base CPU but need templated 476 // MaxThreads so put in here instead 477 committedInsts 478 .init(numThreads) 479 .name(name() + ".committedInsts") 480 .desc("Number of Instructions Simulated") 481 .flags(Stats::total); 482 483 committedOps 484 .init(numThreads) 485 .name(name() + ".committedOps") 486 .desc("Number of Ops (including micro ops) Simulated") 487 .flags(Stats::total); 488 489 cpi 490 .name(name() + ".cpi") 491 .desc("CPI: Cycles Per Instruction") 492 .precision(6); 493 cpi = numCycles / committedInsts; 494 495 totalCpi 496 .name(name() + ".cpi_total") 497 .desc("CPI: Total CPI of All Threads") 498 .precision(6); 499 totalCpi = numCycles / sum(committedInsts); 500 501 ipc 502 .name(name() + ".ipc") 503 .desc("IPC: Instructions Per Cycle") 504 .precision(6); 505 ipc = committedInsts / numCycles; 506 507 totalIpc 508 .name(name() + ".ipc_total") 509 .desc("IPC: Total IPC of All Threads") 510 .precision(6); 511 totalIpc = sum(committedInsts) / numCycles; 512 513 this->fetch.regStats(); 514 this->decode.regStats(); 515 this->rename.regStats(); 516 this->iew.regStats(); 517 this->commit.regStats(); 518 this->rob.regStats(); 519 520 intRegfileReads 521 .name(name() + ".int_regfile_reads") 522 .desc("number of integer regfile reads") 523 .prereq(intRegfileReads); 524 525 intRegfileWrites 526 .name(name() + ".int_regfile_writes") 527 .desc("number of integer regfile writes") 528 .prereq(intRegfileWrites); 529 530 fpRegfileReads 531 .name(name() + ".fp_regfile_reads") 532 .desc("number of floating regfile reads") 533 .prereq(fpRegfileReads); 534 535 fpRegfileWrites 536 .name(name() + ".fp_regfile_writes") 537 .desc("number of floating regfile writes") 538 .prereq(fpRegfileWrites); 539 540 vecRegfileReads 541 .name(name() + ".vec_regfile_reads") 542 .desc("number of vector regfile reads") 543 .prereq(vecRegfileReads); 544 545 vecRegfileWrites 546 .name(name() + ".vec_regfile_writes") 547 .desc("number of vector regfile writes") 548 .prereq(vecRegfileWrites); 549 550 vecPredRegfileReads 551 .name(name() + ".pred_regfile_reads") 552 .desc("number of predicate regfile reads") 553 .prereq(vecPredRegfileReads); 554 555 vecPredRegfileWrites 556 .name(name() + ".pred_regfile_writes") 557 .desc("number of predicate regfile writes") 558 .prereq(vecPredRegfileWrites); 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 updateCycleCounters(BaseCPU::CPU_STATE_ON); 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 updateCycleCounters(BaseCPU::CPU_STATE_SLEEP); 820} 821 822template <class Impl> 823void 824FullO3CPU<Impl>::insertThread(ThreadID tid) 825{ 826 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 827 // Will change now that the PC and thread state is internal to the CPU 828 // and not in the ThreadContext. 829 ThreadContext *src_tc; 830 if (FullSystem) 831 src_tc = system->threadContexts[tid]; 832 else 833 src_tc = tcBase(tid); 834 835 //Bind Int Regs to Rename Map 836 837 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 838 reg_id.index()++) { 839 PhysRegIdPtr phys_reg = freeList.getIntReg(); 840 renameMap[tid].setEntry(reg_id, phys_reg); 841 scoreboard.setReg(phys_reg); 842 } 843 844 //Bind Float Regs to Rename Map 845 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 846 reg_id.index()++) { 847 PhysRegIdPtr phys_reg = freeList.getFloatReg(); 848 renameMap[tid].setEntry(reg_id, phys_reg); 849 scoreboard.setReg(phys_reg); 850 } 851 852 //Bind condition-code Regs to Rename Map 853 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 854 reg_id.index()++) { 855 PhysRegIdPtr phys_reg = freeList.getCCReg(); 856 renameMap[tid].setEntry(reg_id, phys_reg); 857 scoreboard.setReg(phys_reg); 858 } 859 860 //Copy Thread Data Into RegFile 861 //this->copyFromTC(tid); 862 863 //Set PC/NPC/NNPC 864 pcState(src_tc->pcState(), tid); 865 866 src_tc->setStatus(ThreadContext::Active); 867 868 activateContext(tid); 869 870 //Reset ROB/IQ/LSQ Entries 871 commit.rob->resetEntries(); 872} 873 874template <class Impl> 875void 876FullO3CPU<Impl>::removeThread(ThreadID tid) 877{ 878 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 879 880 // Copy Thread Data From RegFile 881 // If thread is suspended, it might be re-allocated 882 // this->copyToTC(tid); 883 884 885 // @todo: 2-27-2008: Fix how we free up rename mappings 886 // here to alleviate the case for double-freeing registers 887 // in SMT workloads. 888 889 // Unbind Int Regs from Rename Map 890 for (RegId reg_id(IntRegClass, 0); reg_id.index() < TheISA::NumIntRegs; 891 reg_id.index()++) { 892 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 893 scoreboard.unsetReg(phys_reg); 894 freeList.addReg(phys_reg); 895 } 896 897 // Unbind Float Regs from Rename Map 898 for (RegId reg_id(FloatRegClass, 0); reg_id.index() < TheISA::NumFloatRegs; 899 reg_id.index()++) { 900 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 901 scoreboard.unsetReg(phys_reg); 902 freeList.addReg(phys_reg); 903 } 904 905 // Unbind Float Regs from Rename Map 906 for (unsigned preg = 0; preg < TheISA::NumVecPredRegs; preg++) { 907 PhysRegIdPtr phys_reg = renameMap[tid].lookup( 908 RegId(VecPredRegClass, preg)); 909 scoreboard.unsetReg(phys_reg); 910 freeList.addReg(phys_reg); 911 } 912 913 // Unbind condition-code Regs from Rename Map 914 for (RegId reg_id(CCRegClass, 0); reg_id.index() < TheISA::NumCCRegs; 915 reg_id.index()++) { 916 PhysRegIdPtr phys_reg = renameMap[tid].lookup(reg_id); 917 scoreboard.unsetReg(phys_reg); 918 freeList.addReg(phys_reg); 919 } 920 921 // Squash Throughout Pipeline 922 DynInstPtr inst = commit.rob->readHeadInst(tid); 923 InstSeqNum squash_seq_num = inst->seqNum; 924 fetch.squash(0, squash_seq_num, inst, tid); 925 decode.squash(tid); 926 rename.squash(squash_seq_num, tid); 927 iew.squash(tid); 928 iew.ldstQueue.squash(squash_seq_num, tid); 929 commit.rob->squash(squash_seq_num, tid); 930 931 932 assert(iew.instQueue.getCount(tid) == 0); 933 assert(iew.ldstQueue.getCount(tid) == 0); 934 935 // Reset ROB/IQ/LSQ Entries 936 937 // Commented out for now. This should be possible to do by 938 // telling all the pipeline stages to drain first, and then 939 // checking until the drain completes. Once the pipeline is 940 // drained, call resetEntries(). - 10-09-06 ktlim 941/* 942 if (activeThreads.size() >= 1) { 943 commit.rob->resetEntries(); 944 iew.resetEntries(); 945 } 946*/ 947} 948 949template <class Impl> 950Fault 951FullO3CPU<Impl>::hwrei(ThreadID tid) 952{ 953#if THE_ISA == ALPHA_ISA 954 // Need to clear the lock flag upon returning from an interrupt. 955 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 956 957 this->thread[tid]->kernelStats->hwrei(); 958 959 // FIXME: XXX check for interrupts? XXX 960#endif 961 return NoFault; 962} 963 964template <class Impl> 965bool 966FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 967{ 968#if THE_ISA == ALPHA_ISA 969 if (this->thread[tid]->kernelStats) 970 this->thread[tid]->kernelStats->callpal(palFunc, 971 this->threadContexts[tid]); 972 973 switch (palFunc) { 974 case PAL::halt: 975 halt(); 976 if (--System::numSystemsRunning == 0) 977 exitSimLoop("all cpus halted"); 978 break; 979 980 case PAL::bpt: 981 case PAL::bugchk: 982 if (this->system->breakpoint()) 983 return false; 984 break; 985 } 986#endif 987 return true; 988} 989 990template <class Impl> 991void 992FullO3CPU<Impl>::switchRenameMode(ThreadID tid, UnifiedFreeList* freelist) 993{ 994 auto pc = this->pcState(tid); 995 996 // new_mode is the new vector renaming mode 997 auto new_mode = RenameMode<TheISA::ISA>::mode(pc); 998 999 // We update vecMode only if there has been a change 1000 if (new_mode != vecMode) { 1001 vecMode = new_mode; 1002 1003 renameMap[tid].switchMode(vecMode); 1004 commitRenameMap[tid].switchMode(vecMode); 1005 renameMap[tid].switchFreeList(freelist); 1006 } 1007} 1008 1009template <class Impl> 1010Fault 1011FullO3CPU<Impl>::getInterrupts() 1012{ 1013 // Check if there are any outstanding interrupts 1014 return this->interrupts[0]->getInterrupt(this->threadContexts[0]); 1015} 1016 1017template <class Impl> 1018void 1019FullO3CPU<Impl>::processInterrupts(const Fault &interrupt) 1020{ 1021 // Check for interrupts here. For now can copy the code that 1022 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1023 // is the one that handles the interrupts. 1024 // @todo: Possibly consolidate the interrupt checking code. 1025 // @todo: Allow other threads to handle interrupts. 1026 1027 assert(interrupt != NoFault); 1028 this->interrupts[0]->updateIntrInfo(this->threadContexts[0]); 1029 1030 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1031 this->trap(interrupt, 0, nullptr); 1032} 1033 1034template <class Impl> 1035void 1036FullO3CPU<Impl>::trap(const Fault &fault, ThreadID tid, 1037 const StaticInstPtr &inst) 1038{ 1039 // Pass the thread's TC into the invoke method. 1040 fault->invoke(this->threadContexts[tid], inst); 1041} 1042 1043template <class Impl> 1044void 1045FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid, Fault *fault) 1046{ 1047 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1048 1049 DPRINTF(Activity,"Activity: syscall() called.\n"); 1050 1051 // Temporarily increase this by one to account for the syscall 1052 // instruction. 1053 ++(this->thread[tid]->funcExeInst); 1054 1055 // Execute the actual syscall. 1056 this->thread[tid]->syscall(callnum, fault); 1057 1058 // Decrease funcExeInst by one as the normal commit will handle 1059 // incrementing it. 1060 --(this->thread[tid]->funcExeInst); 1061} 1062 1063template <class Impl> 1064void 1065FullO3CPU<Impl>::serializeThread(CheckpointOut &cp, ThreadID tid) const 1066{ 1067 thread[tid]->serialize(cp); 1068} 1069 1070template <class Impl> 1071void 1072FullO3CPU<Impl>::unserializeThread(CheckpointIn &cp, ThreadID tid) 1073{ 1074 thread[tid]->unserialize(cp); 1075} 1076 1077template <class Impl> 1078DrainState 1079FullO3CPU<Impl>::drain() 1080{ 1081 // Deschedule any power gating event (if any) 1082 deschedulePowerGatingEvent(); 1083 1084 // If the CPU isn't doing anything, then return immediately. 1085 if (switchedOut()) 1086 return DrainState::Drained; 1087 1088 DPRINTF(Drain, "Draining...\n"); 1089 1090 // We only need to signal a drain to the commit stage as this 1091 // initiates squashing controls the draining. Once the commit 1092 // stage commits an instruction where it is safe to stop, it'll 1093 // squash the rest of the instructions in the pipeline and force 1094 // the fetch stage to stall. The pipeline will be drained once all 1095 // in-flight instructions have retired. 1096 commit.drain(); 1097 1098 // Wake the CPU and record activity so everything can drain out if 1099 // the CPU was not able to immediately drain. 1100 if (!isDrained()) { 1101 // If a thread is suspended, wake it up so it can be drained 1102 for (auto t : threadContexts) { 1103 if (t->status() == ThreadContext::Suspended){ 1104 DPRINTF(Drain, "Currently suspended so activate %i \n", 1105 t->threadId()); 1106 t->activate(); 1107 // As the thread is now active, change the power state as well 1108 activateContext(t->threadId()); 1109 } 1110 } 1111 1112 wakeCPU(); 1113 activityRec.activity(); 1114 1115 DPRINTF(Drain, "CPU not drained\n"); 1116 1117 return DrainState::Draining; 1118 } else { 1119 DPRINTF(Drain, "CPU is already drained\n"); 1120 if (tickEvent.scheduled()) 1121 deschedule(tickEvent); 1122 1123 // Flush out any old data from the time buffers. In 1124 // particular, there might be some data in flight from the 1125 // fetch stage that isn't visible in any of the CPU buffers we 1126 // test in isDrained(). 1127 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1128 timeBuffer.advance(); 1129 fetchQueue.advance(); 1130 decodeQueue.advance(); 1131 renameQueue.advance(); 1132 iewQueue.advance(); 1133 } 1134 1135 drainSanityCheck(); 1136 return DrainState::Drained; 1137 } 1138} 1139 1140template <class Impl> 1141bool 1142FullO3CPU<Impl>::tryDrain() 1143{ 1144 if (drainState() != DrainState::Draining || !isDrained()) 1145 return false; 1146 1147 if (tickEvent.scheduled()) 1148 deschedule(tickEvent); 1149 1150 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1151 signalDrainDone(); 1152 1153 return true; 1154} 1155 1156template <class Impl> 1157void 1158FullO3CPU<Impl>::drainSanityCheck() const 1159{ 1160 assert(isDrained()); 1161 fetch.drainSanityCheck(); 1162 decode.drainSanityCheck(); 1163 rename.drainSanityCheck(); 1164 iew.drainSanityCheck(); 1165 commit.drainSanityCheck(); 1166} 1167 1168template <class Impl> 1169bool 1170FullO3CPU<Impl>::isDrained() const 1171{ 1172 bool drained(true); 1173 1174 if (!instList.empty() || !removeList.empty()) { 1175 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1176 drained = false; 1177 } 1178 1179 if (!fetch.isDrained()) { 1180 DPRINTF(Drain, "Fetch not drained.\n"); 1181 drained = false; 1182 } 1183 1184 if (!decode.isDrained()) { 1185 DPRINTF(Drain, "Decode not drained.\n"); 1186 drained = false; 1187 } 1188 1189 if (!rename.isDrained()) { 1190 DPRINTF(Drain, "Rename not drained.\n"); 1191 drained = false; 1192 } 1193 1194 if (!iew.isDrained()) { 1195 DPRINTF(Drain, "IEW not drained.\n"); 1196 drained = false; 1197 } 1198 1199 if (!commit.isDrained()) { 1200 DPRINTF(Drain, "Commit not drained.\n"); 1201 drained = false; 1202 } 1203 1204 return drained; 1205} 1206 1207template <class Impl> 1208void 1209FullO3CPU<Impl>::commitDrained(ThreadID tid) 1210{ 1211 fetch.drainStall(tid); 1212} 1213 1214template <class Impl> 1215void 1216FullO3CPU<Impl>::drainResume() 1217{ 1218 if (switchedOut()) 1219 return; 1220 1221 DPRINTF(Drain, "Resuming...\n"); 1222 verifyMemoryMode(); 1223 1224 fetch.drainResume(); 1225 commit.drainResume(); 1226 1227 _status = Idle; 1228 for (ThreadID i = 0; i < thread.size(); i++) { 1229 if (thread[i]->status() == ThreadContext::Active) { 1230 DPRINTF(Drain, "Activating thread: %i\n", i); 1231 activateThread(i); 1232 _status = Running; 1233 } 1234 } 1235 1236 assert(!tickEvent.scheduled()); 1237 if (_status == Running) 1238 schedule(tickEvent, nextCycle()); 1239 1240 // Reschedule any power gating event (if any) 1241 schedulePowerGatingEvent(); 1242} 1243 1244template <class Impl> 1245void 1246FullO3CPU<Impl>::switchOut() 1247{ 1248 DPRINTF(O3CPU, "Switching out\n"); 1249 BaseCPU::switchOut(); 1250 1251 activityRec.reset(); 1252 1253 _status = SwitchedOut; 1254 1255 if (checker) 1256 checker->switchOut(); 1257} 1258 1259template <class Impl> 1260void 1261FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1262{ 1263 BaseCPU::takeOverFrom(oldCPU); 1264 1265 fetch.takeOverFrom(); 1266 decode.takeOverFrom(); 1267 rename.takeOverFrom(); 1268 iew.takeOverFrom(); 1269 commit.takeOverFrom(); 1270 1271 assert(!tickEvent.scheduled()); 1272 1273 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1274 if (oldO3CPU) 1275 globalSeqNum = oldO3CPU->globalSeqNum; 1276 1277 lastRunningCycle = curCycle(); 1278 _status = Idle; 1279} 1280 1281template <class Impl> 1282void 1283FullO3CPU<Impl>::verifyMemoryMode() const 1284{ 1285 if (!system->isTimingMode()) { 1286 fatal("The O3 CPU requires the memory system to be in " 1287 "'timing' mode.\n"); 1288 } 1289} 1290 1291template <class Impl> 1292RegVal 1293FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) const 1294{ 1295 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1296} 1297 1298template <class Impl> 1299RegVal 1300FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1301{ 1302 miscRegfileReads++; 1303 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1304} 1305 1306template <class Impl> 1307void 1308FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, RegVal val, ThreadID tid) 1309{ 1310 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1311} 1312 1313template <class Impl> 1314void 1315FullO3CPU<Impl>::setMiscReg(int misc_reg, RegVal val, ThreadID tid) 1316{ 1317 miscRegfileWrites++; 1318 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1319} 1320 1321template <class Impl> 1322RegVal 1323FullO3CPU<Impl>::readIntReg(PhysRegIdPtr phys_reg) 1324{ 1325 intRegfileReads++; 1326 return regFile.readIntReg(phys_reg); 1327} 1328 1329template <class Impl> 1330RegVal 1331FullO3CPU<Impl>::readFloatRegBits(PhysRegIdPtr phys_reg) 1332{ 1333 fpRegfileReads++; 1334 return regFile.readFloatRegBits(phys_reg); 1335} 1336 1337template <class Impl> 1338auto 1339FullO3CPU<Impl>::readVecReg(PhysRegIdPtr phys_reg) const 1340 -> const VecRegContainer& 1341{ 1342 vecRegfileReads++; 1343 return regFile.readVecReg(phys_reg); 1344} 1345 1346template <class Impl> 1347auto 1348FullO3CPU<Impl>::getWritableVecReg(PhysRegIdPtr phys_reg) 1349 -> VecRegContainer& 1350{ 1351 vecRegfileWrites++; 1352 return regFile.getWritableVecReg(phys_reg); 1353} 1354 1355template <class Impl> 1356auto 1357FullO3CPU<Impl>::readVecElem(PhysRegIdPtr phys_reg) const -> const VecElem& 1358{ 1359 vecRegfileReads++; 1360 return regFile.readVecElem(phys_reg); 1361} 1362 1363template <class Impl> 1364auto 1365FullO3CPU<Impl>::readVecPredReg(PhysRegIdPtr phys_reg) const 1366 -> const VecPredRegContainer& 1367{ 1368 vecPredRegfileReads++; 1369 return regFile.readVecPredReg(phys_reg); 1370} 1371 1372template <class Impl> 1373auto 1374FullO3CPU<Impl>::getWritableVecPredReg(PhysRegIdPtr phys_reg) 1375 -> VecPredRegContainer& 1376{ 1377 vecPredRegfileWrites++; 1378 return regFile.getWritableVecPredReg(phys_reg); 1379} 1380 1381template <class Impl> 1382CCReg 1383FullO3CPU<Impl>::readCCReg(PhysRegIdPtr phys_reg) 1384{ 1385 ccRegfileReads++; 1386 return regFile.readCCReg(phys_reg); 1387} 1388 1389template <class Impl> 1390void 1391FullO3CPU<Impl>::setIntReg(PhysRegIdPtr phys_reg, RegVal val) 1392{ 1393 intRegfileWrites++; 1394 regFile.setIntReg(phys_reg, val); 1395} 1396 1397template <class Impl> 1398void 1399FullO3CPU<Impl>::setFloatRegBits(PhysRegIdPtr phys_reg, RegVal val) 1400{ 1401 fpRegfileWrites++; 1402 regFile.setFloatRegBits(phys_reg, val); 1403} 1404 1405template <class Impl> 1406void 1407FullO3CPU<Impl>::setVecReg(PhysRegIdPtr phys_reg, const VecRegContainer& val) 1408{ 1409 vecRegfileWrites++; 1410 regFile.setVecReg(phys_reg, val); 1411} 1412 1413template <class Impl> 1414void 1415FullO3CPU<Impl>::setVecElem(PhysRegIdPtr phys_reg, const VecElem& val) 1416{ 1417 vecRegfileWrites++; 1418 regFile.setVecElem(phys_reg, val); 1419} 1420 1421template <class Impl> 1422void 1423FullO3CPU<Impl>::setVecPredReg(PhysRegIdPtr phys_reg, 1424 const VecPredRegContainer& val) 1425{ 1426 vecPredRegfileWrites++; 1427 regFile.setVecPredReg(phys_reg, val); 1428} 1429 1430template <class Impl> 1431void 1432FullO3CPU<Impl>::setCCReg(PhysRegIdPtr phys_reg, CCReg val) 1433{ 1434 ccRegfileWrites++; 1435 regFile.setCCReg(phys_reg, val); 1436} 1437 1438template <class Impl> 1439RegVal 1440FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1441{ 1442 intRegfileReads++; 1443 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1444 RegId(IntRegClass, reg_idx)); 1445 1446 return regFile.readIntReg(phys_reg); 1447} 1448 1449template <class Impl> 1450RegVal 1451FullO3CPU<Impl>::readArchFloatRegBits(int reg_idx, ThreadID tid) 1452{ 1453 fpRegfileReads++; 1454 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1455 RegId(FloatRegClass, reg_idx)); 1456 1457 return regFile.readFloatRegBits(phys_reg); 1458} 1459 1460template <class Impl> 1461auto 1462FullO3CPU<Impl>::readArchVecReg(int reg_idx, ThreadID tid) const 1463 -> const VecRegContainer& 1464{ 1465 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1466 RegId(VecRegClass, reg_idx)); 1467 return readVecReg(phys_reg); 1468} 1469 1470template <class Impl> 1471auto 1472FullO3CPU<Impl>::getWritableArchVecReg(int reg_idx, ThreadID tid) 1473 -> VecRegContainer& 1474{ 1475 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1476 RegId(VecRegClass, reg_idx)); 1477 return getWritableVecReg(phys_reg); 1478} 1479 1480template <class Impl> 1481auto 1482FullO3CPU<Impl>::readArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1483 ThreadID tid) const -> const VecElem& 1484{ 1485 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1486 RegId(VecElemClass, reg_idx, ldx)); 1487 return readVecElem(phys_reg); 1488} 1489 1490template <class Impl> 1491auto 1492FullO3CPU<Impl>::readArchVecPredReg(int reg_idx, ThreadID tid) const 1493 -> const VecPredRegContainer& 1494{ 1495 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1496 RegId(VecPredRegClass, reg_idx)); 1497 return readVecPredReg(phys_reg); 1498} 1499 1500template <class Impl> 1501auto 1502FullO3CPU<Impl>::getWritableArchVecPredReg(int reg_idx, ThreadID tid) 1503 -> VecPredRegContainer& 1504{ 1505 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1506 RegId(VecPredRegClass, reg_idx)); 1507 return getWritableVecPredReg(phys_reg); 1508} 1509 1510template <class Impl> 1511CCReg 1512FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1513{ 1514 ccRegfileReads++; 1515 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1516 RegId(CCRegClass, reg_idx)); 1517 1518 return regFile.readCCReg(phys_reg); 1519} 1520 1521template <class Impl> 1522void 1523FullO3CPU<Impl>::setArchIntReg(int reg_idx, RegVal val, ThreadID tid) 1524{ 1525 intRegfileWrites++; 1526 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1527 RegId(IntRegClass, reg_idx)); 1528 1529 regFile.setIntReg(phys_reg, val); 1530} 1531 1532template <class Impl> 1533void 1534FullO3CPU<Impl>::setArchFloatRegBits(int reg_idx, RegVal val, ThreadID tid) 1535{ 1536 fpRegfileWrites++; 1537 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1538 RegId(FloatRegClass, reg_idx)); 1539 1540 regFile.setFloatRegBits(phys_reg, val); 1541} 1542 1543template <class Impl> 1544void 1545FullO3CPU<Impl>::setArchVecReg(int reg_idx, const VecRegContainer& val, 1546 ThreadID tid) 1547{ 1548 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1549 RegId(VecRegClass, reg_idx)); 1550 setVecReg(phys_reg, val); 1551} 1552 1553template <class Impl> 1554void 1555FullO3CPU<Impl>::setArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 1556 const VecElem& val, ThreadID tid) 1557{ 1558 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1559 RegId(VecElemClass, reg_idx, ldx)); 1560 setVecElem(phys_reg, val); 1561} 1562 1563template <class Impl> 1564void 1565FullO3CPU<Impl>::setArchVecPredReg(int reg_idx, const VecPredRegContainer& val, 1566 ThreadID tid) 1567{ 1568 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1569 RegId(VecPredRegClass, reg_idx)); 1570 setVecPredReg(phys_reg, val); 1571} 1572 1573template <class Impl> 1574void 1575FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1576{ 1577 ccRegfileWrites++; 1578 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 1579 RegId(CCRegClass, reg_idx)); 1580 1581 regFile.setCCReg(phys_reg, val); 1582} 1583 1584template <class Impl> 1585TheISA::PCState 1586FullO3CPU<Impl>::pcState(ThreadID tid) 1587{ 1588 return commit.pcState(tid); 1589} 1590 1591template <class Impl> 1592void 1593FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1594{ 1595 commit.pcState(val, tid); 1596} 1597 1598template <class Impl> 1599Addr 1600FullO3CPU<Impl>::instAddr(ThreadID tid) 1601{ 1602 return commit.instAddr(tid); 1603} 1604 1605template <class Impl> 1606Addr 1607FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1608{ 1609 return commit.nextInstAddr(tid); 1610} 1611 1612template <class Impl> 1613MicroPC 1614FullO3CPU<Impl>::microPC(ThreadID tid) 1615{ 1616 return commit.microPC(tid); 1617} 1618 1619template <class Impl> 1620void 1621FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1622{ 1623 this->thread[tid]->noSquashFromTC = true; 1624 this->commit.generateTCEvent(tid); 1625} 1626 1627template <class Impl> 1628typename FullO3CPU<Impl>::ListIt 1629FullO3CPU<Impl>::addInst(const DynInstPtr &inst) 1630{ 1631 instList.push_back(inst); 1632 1633 return --(instList.end()); 1634} 1635 1636template <class Impl> 1637void 1638FullO3CPU<Impl>::instDone(ThreadID tid, const DynInstPtr &inst) 1639{ 1640 // Keep an instruction count. 1641 if (!inst->isMicroop() || inst->isLastMicroop()) { 1642 thread[tid]->numInst++; 1643 thread[tid]->numInsts++; 1644 committedInsts[tid]++; 1645 system->totalNumInsts++; 1646 1647 // Check for instruction-count-based events. 1648 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1649 system->instEventQueue.serviceEvents(system->totalNumInsts); 1650 } 1651 thread[tid]->numOp++; 1652 thread[tid]->numOps++; 1653 committedOps[tid]++; 1654 1655 probeInstCommit(inst->staticInst); 1656} 1657 1658template <class Impl> 1659void 1660FullO3CPU<Impl>::removeFrontInst(const DynInstPtr &inst) 1661{ 1662 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1663 "[sn:%lli]\n", 1664 inst->threadNumber, inst->pcState(), inst->seqNum); 1665 1666 removeInstsThisCycle = true; 1667 1668 // Remove the front instruction. 1669 removeList.push(inst->getInstListIt()); 1670} 1671 1672template <class Impl> 1673void 1674FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1675{ 1676 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1677 " list.\n", tid); 1678 1679 ListIt end_it; 1680 1681 bool rob_empty = false; 1682 1683 if (instList.empty()) { 1684 return; 1685 } else if (rob.isEmpty(tid)) { 1686 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1687 end_it = instList.begin(); 1688 rob_empty = true; 1689 } else { 1690 end_it = (rob.readTailInst(tid))->getInstListIt(); 1691 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1692 } 1693 1694 removeInstsThisCycle = true; 1695 1696 ListIt inst_it = instList.end(); 1697 1698 inst_it--; 1699 1700 // Walk through the instruction list, removing any instructions 1701 // that were inserted after the given instruction iterator, end_it. 1702 while (inst_it != end_it) { 1703 assert(!instList.empty()); 1704 1705 squashInstIt(inst_it, tid); 1706 1707 inst_it--; 1708 } 1709 1710 // If the ROB was empty, then we actually need to remove the first 1711 // instruction as well. 1712 if (rob_empty) { 1713 squashInstIt(inst_it, tid); 1714 } 1715} 1716 1717template <class Impl> 1718void 1719FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1720{ 1721 assert(!instList.empty()); 1722 1723 removeInstsThisCycle = true; 1724 1725 ListIt inst_iter = instList.end(); 1726 1727 inst_iter--; 1728 1729 DPRINTF(O3CPU, "Deleting instructions from instruction " 1730 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1731 tid, seq_num, (*inst_iter)->seqNum); 1732 1733 while ((*inst_iter)->seqNum > seq_num) { 1734 1735 bool break_loop = (inst_iter == instList.begin()); 1736 1737 squashInstIt(inst_iter, tid); 1738 1739 inst_iter--; 1740 1741 if (break_loop) 1742 break; 1743 } 1744} 1745 1746template <class Impl> 1747inline void 1748FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1749{ 1750 if ((*instIt)->threadNumber == tid) { 1751 DPRINTF(O3CPU, "Squashing instruction, " 1752 "[tid:%i] [sn:%lli] PC %s\n", 1753 (*instIt)->threadNumber, 1754 (*instIt)->seqNum, 1755 (*instIt)->pcState()); 1756 1757 // Mark it as squashed. 1758 (*instIt)->setSquashed(); 1759 1760 // @todo: Formulate a consistent method for deleting 1761 // instructions from the instruction list 1762 // Remove the instruction from the list. 1763 removeList.push(instIt); 1764 } 1765} 1766 1767template <class Impl> 1768void 1769FullO3CPU<Impl>::cleanUpRemovedInsts() 1770{ 1771 while (!removeList.empty()) { 1772 DPRINTF(O3CPU, "Removing instruction, " 1773 "[tid:%i] [sn:%lli] PC %s\n", 1774 (*removeList.front())->threadNumber, 1775 (*removeList.front())->seqNum, 1776 (*removeList.front())->pcState()); 1777 1778 instList.erase(removeList.front()); 1779 1780 removeList.pop(); 1781 } 1782 1783 removeInstsThisCycle = false; 1784} 1785/* 1786template <class Impl> 1787void 1788FullO3CPU<Impl>::removeAllInsts() 1789{ 1790 instList.clear(); 1791} 1792*/ 1793template <class Impl> 1794void 1795FullO3CPU<Impl>::dumpInsts() 1796{ 1797 int num = 0; 1798 1799 ListIt inst_list_it = instList.begin(); 1800 1801 cprintf("Dumping Instruction List\n"); 1802 1803 while (inst_list_it != instList.end()) { 1804 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1805 "Squashed:%i\n\n", 1806 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1807 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1808 (*inst_list_it)->isSquashed()); 1809 inst_list_it++; 1810 ++num; 1811 } 1812} 1813/* 1814template <class Impl> 1815void 1816FullO3CPU<Impl>::wakeDependents(const DynInstPtr &inst) 1817{ 1818 iew.wakeDependents(inst); 1819} 1820*/ 1821template <class Impl> 1822void 1823FullO3CPU<Impl>::wakeCPU() 1824{ 1825 if (activityRec.active() || tickEvent.scheduled()) { 1826 DPRINTF(Activity, "CPU already running.\n"); 1827 return; 1828 } 1829 1830 DPRINTF(Activity, "Waking up CPU\n"); 1831 1832 Cycles cycles(curCycle() - lastRunningCycle); 1833 // @todo: This is an oddity that is only here to match the stats 1834 if (cycles > 1) { 1835 --cycles; 1836 idleCycles += cycles; 1837 numCycles += cycles; 1838 } 1839 1840 schedule(tickEvent, clockEdge()); 1841} 1842 1843template <class Impl> 1844void 1845FullO3CPU<Impl>::wakeup(ThreadID tid) 1846{ 1847 if (this->thread[tid]->status() != ThreadContext::Suspended) 1848 return; 1849 1850 this->wakeCPU(); 1851 1852 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1853 this->threadContexts[tid]->activate(); 1854} 1855 1856template <class Impl> 1857ThreadID 1858FullO3CPU<Impl>::getFreeTid() 1859{ 1860 for (ThreadID tid = 0; tid < numThreads; tid++) { 1861 if (!tids[tid]) { 1862 tids[tid] = true; 1863 return tid; 1864 } 1865 } 1866 1867 return InvalidThreadID; 1868} 1869 1870template <class Impl> 1871void 1872FullO3CPU<Impl>::updateThreadPriority() 1873{ 1874 if (activeThreads.size() > 1) { 1875 //DEFAULT TO ROUND ROBIN SCHEME 1876 //e.g. Move highest priority to end of thread list 1877 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1878 1879 unsigned high_thread = *list_begin; 1880 1881 activeThreads.erase(list_begin); 1882 1883 activeThreads.push_back(high_thread); 1884 } 1885} 1886 1887// Forward declaration of FullO3CPU. 1888template class FullO3CPU<O3CPUImpl>; 1889