cpu.cc revision 2733
1/* 2 * Copyright (c) 2004-2006 The Regents of The University of Michigan 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are 7 * met: redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer; 9 * redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution; 12 * neither the name of the copyright holders nor the names of its 13 * contributors may be used to endorse or promote products derived from 14 * this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 * 28 * Authors: Kevin Lim 29 */ 30 31#include "config/full_system.hh" 32#include "config/use_checker.hh" 33 34#if FULL_SYSTEM 35#include "sim/system.hh" 36#else 37#include "sim/process.hh" 38#endif 39 40#include "cpu/activity.hh" 41#include "cpu/checker/cpu.hh" 42#include "cpu/simple_thread.hh" 43#include "cpu/thread_context.hh" 44#include "cpu/o3/alpha_dyn_inst.hh" 45#include "cpu/o3/alpha_impl.hh" 46#include "cpu/o3/cpu.hh" 47 48#include "sim/root.hh" 49#include "sim/stat_control.hh" 50 51using namespace std; 52using namespace TheISA; 53 54BaseO3CPU::BaseO3CPU(Params *params) 55 : BaseCPU(params), cpu_id(0) 56{ 57} 58 59void 60BaseO3CPU::regStats() 61{ 62 BaseCPU::regStats(); 63} 64 65template <class Impl> 66FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 67 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) 68{ 69} 70 71template <class Impl> 72void 73FullO3CPU<Impl>::TickEvent::process() 74{ 75 cpu->tick(); 76} 77 78template <class Impl> 79const char * 80FullO3CPU<Impl>::TickEvent::description() 81{ 82 return "FullO3CPU tick event"; 83} 84 85template <class Impl> 86FullO3CPU<Impl>::FullO3CPU(Params *params) 87 : BaseO3CPU(params), 88 tickEvent(this), 89 removeInstsThisCycle(false), 90 fetch(params), 91 decode(params), 92 rename(params), 93 iew(params), 94 commit(params), 95 96 regFile(params->numPhysIntRegs, params->numPhysFloatRegs), 97 98 freeList(params->numberOfThreads,//number of activeThreads 99 TheISA::NumIntRegs, params->numPhysIntRegs, 100 TheISA::NumFloatRegs, params->numPhysFloatRegs), 101 102 rob(params->numROBEntries, params->squashWidth, 103 params->smtROBPolicy, params->smtROBThreshold, 104 params->numberOfThreads), 105 106 scoreboard(params->numberOfThreads,//number of activeThreads 107 TheISA::NumIntRegs, params->numPhysIntRegs, 108 TheISA::NumFloatRegs, params->numPhysFloatRegs, 109 TheISA::NumMiscRegs * number_of_threads, 110 TheISA::ZeroReg), 111 112 // For now just have these time buffers be pretty big. 113 // @todo: Make these time buffer sizes parameters or derived 114 // from latencies 115 timeBuffer(5, 5), 116 fetchQueue(5, 5), 117 decodeQueue(5, 5), 118 renameQueue(5, 5), 119 iewQueue(5, 5), 120 activityRec(NumStages, 10, params->activity), 121 122 globalSeqNum(1), 123 124#if FULL_SYSTEM 125 system(params->system), 126 physmem(system->physmem), 127#endif // FULL_SYSTEM 128 mem(params->mem), 129 switchCount(0), 130 deferRegistration(params->deferRegistration), 131 numThreads(number_of_threads) 132{ 133 _status = Idle; 134 135 checker = NULL; 136 137#if USE_CHECKER 138 if (params->checker) { 139 BaseCPU *temp_checker = params->checker; 140 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 141 checker->setMemory(mem); 142#if FULL_SYSTEM 143 checker->setSystem(params->system); 144#endif 145 } 146#endif 147 148#if !FULL_SYSTEM 149 thread.resize(number_of_threads); 150 tids.resize(number_of_threads); 151#endif 152 153 // The stages also need their CPU pointer setup. However this 154 // must be done at the upper level CPU because they have pointers 155 // to the upper level CPU, and not this FullO3CPU. 156 157 // Set up Pointers to the activeThreads list for each stage 158 fetch.setActiveThreads(&activeThreads); 159 decode.setActiveThreads(&activeThreads); 160 rename.setActiveThreads(&activeThreads); 161 iew.setActiveThreads(&activeThreads); 162 commit.setActiveThreads(&activeThreads); 163 164 // Give each of the stages the time buffer they will use. 165 fetch.setTimeBuffer(&timeBuffer); 166 decode.setTimeBuffer(&timeBuffer); 167 rename.setTimeBuffer(&timeBuffer); 168 iew.setTimeBuffer(&timeBuffer); 169 commit.setTimeBuffer(&timeBuffer); 170 171 // Also setup each of the stages' queues. 172 fetch.setFetchQueue(&fetchQueue); 173 decode.setFetchQueue(&fetchQueue); 174 commit.setFetchQueue(&fetchQueue); 175 decode.setDecodeQueue(&decodeQueue); 176 rename.setDecodeQueue(&decodeQueue); 177 rename.setRenameQueue(&renameQueue); 178 iew.setRenameQueue(&renameQueue); 179 iew.setIEWQueue(&iewQueue); 180 commit.setIEWQueue(&iewQueue); 181 commit.setRenameQueue(&renameQueue); 182 183 commit.setFetchStage(&fetch); 184 commit.setIEWStage(&iew); 185 rename.setIEWStage(&iew); 186 rename.setCommitStage(&commit); 187 188#if !FULL_SYSTEM 189 int active_threads = params->workload.size(); 190#else 191 int active_threads = 1; 192#endif 193 194 //Make Sure That this a Valid Architeture 195 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 196 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 197 198 rename.setScoreboard(&scoreboard); 199 iew.setScoreboard(&scoreboard); 200 201 // Setup the rename map for whichever stages need it. 202 PhysRegIndex lreg_idx = 0; 203 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 204 205 for (int tid=0; tid < numThreads; tid++) { 206 bool bindRegs = (tid <= active_threads - 1); 207 208 commitRenameMap[tid].init(TheISA::NumIntRegs, 209 params->numPhysIntRegs, 210 lreg_idx, //Index for Logical. Regs 211 212 TheISA::NumFloatRegs, 213 params->numPhysFloatRegs, 214 freg_idx, //Index for Float Regs 215 216 TheISA::NumMiscRegs, 217 218 TheISA::ZeroReg, 219 TheISA::ZeroReg, 220 221 tid, 222 false); 223 224 renameMap[tid].init(TheISA::NumIntRegs, 225 params->numPhysIntRegs, 226 lreg_idx, //Index for Logical. Regs 227 228 TheISA::NumFloatRegs, 229 params->numPhysFloatRegs, 230 freg_idx, //Index for Float Regs 231 232 TheISA::NumMiscRegs, 233 234 TheISA::ZeroReg, 235 TheISA::ZeroReg, 236 237 tid, 238 bindRegs); 239 } 240 241 rename.setRenameMap(renameMap); 242 commit.setRenameMap(commitRenameMap); 243 244 // Give renameMap & rename stage access to the freeList; 245 for (int i=0; i < numThreads; i++) { 246 renameMap[i].setFreeList(&freeList); 247 } 248 rename.setFreeList(&freeList); 249 250 // Setup the ROB for whichever stages need it. 251 commit.setROB(&rob); 252 253 lastRunningCycle = curTick; 254 255 contextSwitch = false; 256} 257 258template <class Impl> 259FullO3CPU<Impl>::~FullO3CPU() 260{ 261} 262 263template <class Impl> 264void 265FullO3CPU<Impl>::fullCPURegStats() 266{ 267 BaseO3CPU::regStats(); 268 269 // Register any of the O3CPU's stats here. 270 timesIdled 271 .name(name() + ".timesIdled") 272 .desc("Number of times that the entire CPU went into an idle state and" 273 " unscheduled itself") 274 .prereq(timesIdled); 275 276 idleCycles 277 .name(name() + ".idleCycles") 278 .desc("Total number of cycles that the CPU has spent unscheduled due " 279 "to idling") 280 .prereq(idleCycles); 281 282 // Number of Instructions simulated 283 // -------------------------------- 284 // Should probably be in Base CPU but need templated 285 // MaxThreads so put in here instead 286 committedInsts 287 .init(numThreads) 288 .name(name() + ".committedInsts") 289 .desc("Number of Instructions Simulated"); 290 291 totalCommittedInsts 292 .name(name() + ".committedInsts_total") 293 .desc("Number of Instructions Simulated"); 294 295 cpi 296 .name(name() + ".cpi") 297 .desc("CPI: Cycles Per Instruction") 298 .precision(6); 299 cpi = simTicks / committedInsts; 300 301 totalCpi 302 .name(name() + ".cpi_total") 303 .desc("CPI: Total CPI of All Threads") 304 .precision(6); 305 totalCpi = simTicks / totalCommittedInsts; 306 307 ipc 308 .name(name() + ".ipc") 309 .desc("IPC: Instructions Per Cycle") 310 .precision(6); 311 ipc = committedInsts / simTicks; 312 313 totalIpc 314 .name(name() + ".ipc_total") 315 .desc("IPC: Total IPC of All Threads") 316 .precision(6); 317 totalIpc = totalCommittedInsts / simTicks; 318 319} 320 321template <class Impl> 322void 323FullO3CPU<Impl>::tick() 324{ 325 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 326 327 ++numCycles; 328 329// activity = false; 330 331 //Tick each of the stages 332 fetch.tick(); 333 334 decode.tick(); 335 336 rename.tick(); 337 338 iew.tick(); 339 340 commit.tick(); 341 342#if !FULL_SYSTEM 343 doContextSwitch(); 344#endif 345 346 // Now advance the time buffers 347 timeBuffer.advance(); 348 349 fetchQueue.advance(); 350 decodeQueue.advance(); 351 renameQueue.advance(); 352 iewQueue.advance(); 353 354 activityRec.advance(); 355 356 if (removeInstsThisCycle) { 357 cleanUpRemovedInsts(); 358 } 359 360 if (!tickEvent.scheduled()) { 361 if (_status == SwitchedOut) { 362 // increment stat 363 lastRunningCycle = curTick; 364 } else if (!activityRec.active()) { 365 lastRunningCycle = curTick; 366 timesIdled++; 367 } else { 368 tickEvent.schedule(curTick + cycles(1)); 369 } 370 } 371 372#if !FULL_SYSTEM 373 updateThreadPriority(); 374#endif 375 376} 377 378template <class Impl> 379void 380FullO3CPU<Impl>::init() 381{ 382 if (!deferRegistration) { 383 registerThreadContexts(); 384 } 385 386 // Set inSyscall so that the CPU doesn't squash when initially 387 // setting up registers. 388 for (int i = 0; i < number_of_threads; ++i) 389 thread[i]->inSyscall = true; 390 391 for (int tid=0; tid < number_of_threads; tid++) { 392#if FULL_SYSTEM 393 ThreadContext *src_tc = threadContexts[tid]; 394#else 395 ThreadContext *src_tc = thread[tid]->getTC(); 396#endif 397 // Threads start in the Suspended State 398 if (src_tc->status() != ThreadContext::Suspended) { 399 continue; 400 } 401 402#if FULL_SYSTEM 403 TheISA::initCPU(src_tc, src_tc->readCpuId()); 404#endif 405 } 406 407 // Clear inSyscall. 408 for (int i = 0; i < number_of_threads; ++i) 409 thread[i]->inSyscall = false; 410 411 // Initialize stages. 412 fetch.initStage(); 413 iew.initStage(); 414 rename.initStage(); 415 commit.initStage(); 416 417 commit.setThreads(thread); 418} 419 420template <class Impl> 421void 422FullO3CPU<Impl>::insertThread(unsigned tid) 423{ 424 DPRINTF(O3CPU,"[tid:%i] Initializing thread data"); 425 // Will change now that the PC and thread state is internal to the CPU 426 // and not in the ThreadContext. 427#if 0 428#if FULL_SYSTEM 429 ThreadContext *src_tc = system->threadContexts[tid]; 430#else 431 ThreadContext *src_tc = thread[tid]; 432#endif 433 434 //Bind Int Regs to Rename Map 435 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 436 PhysRegIndex phys_reg = freeList.getIntReg(); 437 438 renameMap[tid].setEntry(ireg,phys_reg); 439 scoreboard.setReg(phys_reg); 440 } 441 442 //Bind Float Regs to Rename Map 443 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 444 PhysRegIndex phys_reg = freeList.getFloatReg(); 445 446 renameMap[tid].setEntry(freg,phys_reg); 447 scoreboard.setReg(phys_reg); 448 } 449 450 //Copy Thread Data Into RegFile 451 this->copyFromTC(tid); 452 453 //Set PC/NPC 454 regFile.pc[tid] = src_tc->readPC(); 455 regFile.npc[tid] = src_tc->readNextPC(); 456 457 src_tc->setStatus(ThreadContext::Active); 458 459 activateContext(tid,1); 460 461 //Reset ROB/IQ/LSQ Entries 462 commit.rob->resetEntries(); 463 iew.resetEntries(); 464#endif 465} 466 467template <class Impl> 468void 469FullO3CPU<Impl>::removeThread(unsigned tid) 470{ 471 DPRINTF(O3CPU,"[tid:%i] Removing thread data"); 472#if 0 473 //Unbind Int Regs from Rename Map 474 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 475 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 476 477 scoreboard.unsetReg(phys_reg); 478 freeList.addReg(phys_reg); 479 } 480 481 //Unbind Float Regs from Rename Map 482 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 483 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 484 485 scoreboard.unsetReg(phys_reg); 486 freeList.addReg(phys_reg); 487 } 488 489 //Copy Thread Data From RegFile 490 /* Fix Me: 491 * Do we really need to do this if we are removing a thread 492 * in the sense that it's finished (exiting)? If the thread is just 493 * being suspended we might... 494 */ 495// this->copyToTC(tid); 496 497 //Squash Throughout Pipeline 498 fetch.squash(0,tid); 499 decode.squash(tid); 500 rename.squash(tid); 501 502 assert(iew.ldstQueue.getCount(tid) == 0); 503 504 //Reset ROB/IQ/LSQ Entries 505 if (activeThreads.size() >= 1) { 506 commit.rob->resetEntries(); 507 iew.resetEntries(); 508 } 509#endif 510} 511 512 513template <class Impl> 514void 515FullO3CPU<Impl>::activateWhenReady(int tid) 516{ 517 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 518 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 519 tid); 520 521 bool ready = true; 522 523 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 524 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 525 "Phys. Int. Regs.\n", 526 tid); 527 ready = false; 528 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 529 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 530 "Phys. Float. Regs.\n", 531 tid); 532 ready = false; 533 } else if (commit.rob->numFreeEntries() >= 534 commit.rob->entryAmount(activeThreads.size() + 1)) { 535 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 536 "ROB entries.\n", 537 tid); 538 ready = false; 539 } else if (iew.instQueue.numFreeEntries() >= 540 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 541 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 542 "IQ entries.\n", 543 tid); 544 ready = false; 545 } else if (iew.ldstQueue.numFreeEntries() >= 546 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 547 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 548 "LSQ entries.\n", 549 tid); 550 ready = false; 551 } 552 553 if (ready) { 554 insertThread(tid); 555 556 contextSwitch = false; 557 558 cpuWaitList.remove(tid); 559 } else { 560 suspendContext(tid); 561 562 //blocks fetch 563 contextSwitch = true; 564 565 //do waitlist 566 cpuWaitList.push_back(tid); 567 } 568} 569 570template <class Impl> 571void 572FullO3CPU<Impl>::activateContext(int tid, int delay) 573{ 574 // Needs to set each stage to running as well. 575 list<unsigned>::iterator isActive = find( 576 activeThreads.begin(), activeThreads.end(), tid); 577 578 if (isActive == activeThreads.end()) { 579 //May Need to Re-code this if the delay variable is the 580 //delay needed for thread to activate 581 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 582 tid); 583 584 activeThreads.push_back(tid); 585 } 586 587 assert(_status == Idle || _status == SwitchedOut); 588 589 scheduleTickEvent(delay); 590 591 // Be sure to signal that there's some activity so the CPU doesn't 592 // deschedule itself. 593 activityRec.activity(); 594 fetch.wakeFromQuiesce(); 595 596 _status = Running; 597} 598 599template <class Impl> 600void 601FullO3CPU<Impl>::suspendContext(int tid) 602{ 603 DPRINTF(O3CPU,"[tid: %i]: Suspended ...\n", tid); 604 unscheduleTickEvent(); 605 _status = Idle; 606/* 607 //Remove From Active List, if Active 608 list<unsigned>::iterator isActive = find( 609 activeThreads.begin(), activeThreads.end(), tid); 610 611 if (isActive != activeThreads.end()) { 612 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 613 tid); 614 activeThreads.erase(isActive); 615 } 616*/ 617} 618 619template <class Impl> 620void 621FullO3CPU<Impl>::deallocateContext(int tid) 622{ 623 DPRINTF(O3CPU,"[tid:%i]: Deallocating ...", tid); 624/* 625 //Remove From Active List, if Active 626 list<unsigned>::iterator isActive = find( 627 activeThreads.begin(), activeThreads.end(), tid); 628 629 if (isActive != activeThreads.end()) { 630 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 631 tid); 632 activeThreads.erase(isActive); 633 634 removeThread(tid); 635 } 636*/ 637} 638 639template <class Impl> 640void 641FullO3CPU<Impl>::haltContext(int tid) 642{ 643 DPRINTF(O3CPU,"[tid:%i]: Halted ...", tid); 644/* 645 //Remove From Active List, if Active 646 list<unsigned>::iterator isActive = find( 647 activeThreads.begin(), activeThreads.end(), tid); 648 649 if (isActive != activeThreads.end()) { 650 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 651 tid); 652 activeThreads.erase(isActive); 653 654 removeThread(tid); 655 } 656*/ 657} 658 659template <class Impl> 660void 661FullO3CPU<Impl>::switchOut(Sampler *_sampler) 662{ 663 sampler = _sampler; 664 switchCount = 0; 665 fetch.switchOut(); 666 decode.switchOut(); 667 rename.switchOut(); 668 iew.switchOut(); 669 commit.switchOut(); 670 671 // Wake the CPU and record activity so everything can drain out if 672 // the CPU is currently idle. 673 wakeCPU(); 674 activityRec.activity(); 675} 676 677template <class Impl> 678void 679FullO3CPU<Impl>::signalSwitched() 680{ 681 if (++switchCount == NumStages) { 682 fetch.doSwitchOut(); 683 rename.doSwitchOut(); 684 commit.doSwitchOut(); 685 instList.clear(); 686 while (!removeList.empty()) { 687 removeList.pop(); 688 } 689 690 if (checker) 691 checker->switchOut(sampler); 692 693 if (tickEvent.scheduled()) 694 tickEvent.squash(); 695 sampler->signalSwitched(); 696 _status = SwitchedOut; 697 } 698 assert(switchCount <= 5); 699} 700 701template <class Impl> 702void 703FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 704{ 705 // Flush out any old data from the time buffers. 706 for (int i = 0; i < 10; ++i) { 707 timeBuffer.advance(); 708 fetchQueue.advance(); 709 decodeQueue.advance(); 710 renameQueue.advance(); 711 iewQueue.advance(); 712 } 713 714 activityRec.reset(); 715 716 BaseCPU::takeOverFrom(oldCPU); 717 718 fetch.takeOverFrom(); 719 decode.takeOverFrom(); 720 rename.takeOverFrom(); 721 iew.takeOverFrom(); 722 commit.takeOverFrom(); 723 724 assert(!tickEvent.scheduled()); 725 726 // @todo: Figure out how to properly select the tid to put onto 727 // the active threads list. 728 int tid = 0; 729 730 list<unsigned>::iterator isActive = find( 731 activeThreads.begin(), activeThreads.end(), tid); 732 733 if (isActive == activeThreads.end()) { 734 //May Need to Re-code this if the delay variable is the delay 735 //needed for thread to activate 736 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 737 tid); 738 739 activeThreads.push_back(tid); 740 } 741 742 // Set all statuses to active, schedule the CPU's tick event. 743 // @todo: Fix up statuses so this is handled properly 744 for (int i = 0; i < threadContexts.size(); ++i) { 745 ThreadContext *tc = threadContexts[i]; 746 if (tc->status() == ThreadContext::Active && _status != Running) { 747 _status = Running; 748 tickEvent.schedule(curTick); 749 } 750 } 751 if (!tickEvent.scheduled()) 752 tickEvent.schedule(curTick); 753} 754 755template <class Impl> 756uint64_t 757FullO3CPU<Impl>::readIntReg(int reg_idx) 758{ 759 return regFile.readIntReg(reg_idx); 760} 761 762template <class Impl> 763FloatReg 764FullO3CPU<Impl>::readFloatReg(int reg_idx, int width) 765{ 766 return regFile.readFloatReg(reg_idx, width); 767} 768 769template <class Impl> 770FloatReg 771FullO3CPU<Impl>::readFloatReg(int reg_idx) 772{ 773 return regFile.readFloatReg(reg_idx); 774} 775 776template <class Impl> 777FloatRegBits 778FullO3CPU<Impl>::readFloatRegBits(int reg_idx, int width) 779{ 780 return regFile.readFloatRegBits(reg_idx, width); 781} 782 783template <class Impl> 784FloatRegBits 785FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 786{ 787 return regFile.readFloatRegBits(reg_idx); 788} 789 790template <class Impl> 791void 792FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 793{ 794 regFile.setIntReg(reg_idx, val); 795} 796 797template <class Impl> 798void 799FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val, int width) 800{ 801 regFile.setFloatReg(reg_idx, val, width); 802} 803 804template <class Impl> 805void 806FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 807{ 808 regFile.setFloatReg(reg_idx, val); 809} 810 811template <class Impl> 812void 813FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val, int width) 814{ 815 regFile.setFloatRegBits(reg_idx, val, width); 816} 817 818template <class Impl> 819void 820FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 821{ 822 regFile.setFloatRegBits(reg_idx, val); 823} 824 825template <class Impl> 826uint64_t 827FullO3CPU<Impl>::readArchIntReg(int reg_idx, unsigned tid) 828{ 829 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 830 831 return regFile.readIntReg(phys_reg); 832} 833 834template <class Impl> 835float 836FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid) 837{ 838 int idx = reg_idx + TheISA::FP_Base_DepTag; 839 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 840 841 return regFile.readFloatReg(phys_reg); 842} 843 844template <class Impl> 845double 846FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid) 847{ 848 int idx = reg_idx + TheISA::FP_Base_DepTag; 849 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 850 851 return regFile.readFloatReg(phys_reg, 64); 852} 853 854template <class Impl> 855uint64_t 856FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid) 857{ 858 int idx = reg_idx + TheISA::FP_Base_DepTag; 859 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 860 861 return regFile.readFloatRegBits(phys_reg); 862} 863 864template <class Impl> 865void 866FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, unsigned tid) 867{ 868 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 869 870 regFile.setIntReg(phys_reg, val); 871} 872 873template <class Impl> 874void 875FullO3CPU<Impl>::setArchFloatRegSingle(int reg_idx, float val, unsigned tid) 876{ 877 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 878 879 regFile.setFloatReg(phys_reg, val); 880} 881 882template <class Impl> 883void 884FullO3CPU<Impl>::setArchFloatRegDouble(int reg_idx, double val, unsigned tid) 885{ 886 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 887 888 regFile.setFloatReg(phys_reg, val, 64); 889} 890 891template <class Impl> 892void 893FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid) 894{ 895 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 896 897 regFile.setFloatRegBits(phys_reg, val); 898} 899 900template <class Impl> 901uint64_t 902FullO3CPU<Impl>::readPC(unsigned tid) 903{ 904 return commit.readPC(tid); 905} 906 907template <class Impl> 908void 909FullO3CPU<Impl>::setPC(Addr new_PC,unsigned tid) 910{ 911 commit.setPC(new_PC, tid); 912} 913 914template <class Impl> 915uint64_t 916FullO3CPU<Impl>::readNextPC(unsigned tid) 917{ 918 return commit.readNextPC(tid); 919} 920 921template <class Impl> 922void 923FullO3CPU<Impl>::setNextPC(uint64_t val,unsigned tid) 924{ 925 commit.setNextPC(val, tid); 926} 927 928template <class Impl> 929typename FullO3CPU<Impl>::ListIt 930FullO3CPU<Impl>::addInst(DynInstPtr &inst) 931{ 932 instList.push_back(inst); 933 934 return --(instList.end()); 935} 936 937template <class Impl> 938void 939FullO3CPU<Impl>::instDone(unsigned tid) 940{ 941 // Keep an instruction count. 942 thread[tid]->numInst++; 943 thread[tid]->numInsts++; 944 committedInsts[tid]++; 945 totalCommittedInsts++; 946 947 // Check for instruction-count-based events. 948 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 949} 950 951template <class Impl> 952void 953FullO3CPU<Impl>::addToRemoveList(DynInstPtr &inst) 954{ 955 removeInstsThisCycle = true; 956 957 removeList.push(inst->getInstListIt()); 958} 959 960template <class Impl> 961void 962FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 963{ 964 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %#x " 965 "[sn:%lli]\n", 966 inst->threadNumber, inst->readPC(), inst->seqNum); 967 968 removeInstsThisCycle = true; 969 970 // Remove the front instruction. 971 removeList.push(inst->getInstListIt()); 972} 973 974template <class Impl> 975void 976FullO3CPU<Impl>::removeInstsNotInROB(unsigned tid) 977{ 978 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 979 " list.\n", tid); 980 981 ListIt end_it; 982 983 bool rob_empty = false; 984 985 if (instList.empty()) { 986 return; 987 } else if (rob.isEmpty(/*tid*/)) { 988 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 989 end_it = instList.begin(); 990 rob_empty = true; 991 } else { 992 end_it = (rob.readTailInst(tid))->getInstListIt(); 993 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 994 } 995 996 removeInstsThisCycle = true; 997 998 ListIt inst_it = instList.end(); 999 1000 inst_it--; 1001 1002 // Walk through the instruction list, removing any instructions 1003 // that were inserted after the given instruction iterator, end_it. 1004 while (inst_it != end_it) { 1005 assert(!instList.empty()); 1006 1007 squashInstIt(inst_it, tid); 1008 1009 inst_it--; 1010 } 1011 1012 // If the ROB was empty, then we actually need to remove the first 1013 // instruction as well. 1014 if (rob_empty) { 1015 squashInstIt(inst_it, tid); 1016 } 1017} 1018 1019template <class Impl> 1020void 1021FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, 1022 unsigned tid) 1023{ 1024 assert(!instList.empty()); 1025 1026 removeInstsThisCycle = true; 1027 1028 ListIt inst_iter = instList.end(); 1029 1030 inst_iter--; 1031 1032 DPRINTF(O3CPU, "Deleting instructions from instruction " 1033 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1034 tid, seq_num, (*inst_iter)->seqNum); 1035 1036 while ((*inst_iter)->seqNum > seq_num) { 1037 1038 bool break_loop = (inst_iter == instList.begin()); 1039 1040 squashInstIt(inst_iter, tid); 1041 1042 inst_iter--; 1043 1044 if (break_loop) 1045 break; 1046 } 1047} 1048 1049template <class Impl> 1050inline void 1051FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, const unsigned &tid) 1052{ 1053 if ((*instIt)->threadNumber == tid) { 1054 DPRINTF(O3CPU, "Squashing instruction, " 1055 "[tid:%i] [sn:%lli] PC %#x\n", 1056 (*instIt)->threadNumber, 1057 (*instIt)->seqNum, 1058 (*instIt)->readPC()); 1059 1060 // Mark it as squashed. 1061 (*instIt)->setSquashed(); 1062 1063 // @todo: Formulate a consistent method for deleting 1064 // instructions from the instruction list 1065 // Remove the instruction from the list. 1066 removeList.push(instIt); 1067 } 1068} 1069 1070template <class Impl> 1071void 1072FullO3CPU<Impl>::cleanUpRemovedInsts() 1073{ 1074 while (!removeList.empty()) { 1075 DPRINTF(O3CPU, "Removing instruction, " 1076 "[tid:%i] [sn:%lli] PC %#x\n", 1077 (*removeList.front())->threadNumber, 1078 (*removeList.front())->seqNum, 1079 (*removeList.front())->readPC()); 1080 1081 instList.erase(removeList.front()); 1082 1083 removeList.pop(); 1084 } 1085 1086 removeInstsThisCycle = false; 1087} 1088/* 1089template <class Impl> 1090void 1091FullO3CPU<Impl>::removeAllInsts() 1092{ 1093 instList.clear(); 1094} 1095*/ 1096template <class Impl> 1097void 1098FullO3CPU<Impl>::dumpInsts() 1099{ 1100 int num = 0; 1101 1102 ListIt inst_list_it = instList.begin(); 1103 1104 cprintf("Dumping Instruction List\n"); 1105 1106 while (inst_list_it != instList.end()) { 1107 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1108 "Squashed:%i\n\n", 1109 num, (*inst_list_it)->readPC(), (*inst_list_it)->threadNumber, 1110 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1111 (*inst_list_it)->isSquashed()); 1112 inst_list_it++; 1113 ++num; 1114 } 1115} 1116/* 1117template <class Impl> 1118void 1119FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1120{ 1121 iew.wakeDependents(inst); 1122} 1123*/ 1124template <class Impl> 1125void 1126FullO3CPU<Impl>::wakeCPU() 1127{ 1128 if (activityRec.active() || tickEvent.scheduled()) { 1129 DPRINTF(Activity, "CPU already running.\n"); 1130 return; 1131 } 1132 1133 DPRINTF(Activity, "Waking up CPU\n"); 1134 1135 idleCycles += (curTick - 1) - lastRunningCycle; 1136 1137 tickEvent.schedule(curTick); 1138} 1139 1140template <class Impl> 1141int 1142FullO3CPU<Impl>::getFreeTid() 1143{ 1144 for (int i=0; i < numThreads; i++) { 1145 if (!tids[i]) { 1146 tids[i] = true; 1147 return i; 1148 } 1149 } 1150 1151 return -1; 1152} 1153 1154template <class Impl> 1155void 1156FullO3CPU<Impl>::doContextSwitch() 1157{ 1158 if (contextSwitch) { 1159 1160 //ADD CODE TO DEACTIVE THREAD HERE (???) 1161 1162 for (int tid=0; tid < cpuWaitList.size(); tid++) { 1163 activateWhenReady(tid); 1164 } 1165 1166 if (cpuWaitList.size() == 0) 1167 contextSwitch = true; 1168 } 1169} 1170 1171template <class Impl> 1172void 1173FullO3CPU<Impl>::updateThreadPriority() 1174{ 1175 if (activeThreads.size() > 1) 1176 { 1177 //DEFAULT TO ROUND ROBIN SCHEME 1178 //e.g. Move highest priority to end of thread list 1179 list<unsigned>::iterator list_begin = activeThreads.begin(); 1180 list<unsigned>::iterator list_end = activeThreads.end(); 1181 1182 unsigned high_thread = *list_begin; 1183 1184 activeThreads.erase(list_begin); 1185 1186 activeThreads.push_back(high_thread); 1187 } 1188} 1189 1190// Forward declaration of FullO3CPU. 1191template class FullO3CPU<AlphaSimpleImpl>; 1192