fetch_impl.hh revision 8460:3893d9d2c6c2
1/* 2 * Copyright (c) 2010 ARM Limited 3 * All rights reserved. 4 * 5 * The license below extends only to copyright in the software and shall 6 * not be construed as granting a license to any other intellectual 7 * property including but not limited to intellectual property relating 8 * to a hardware implementation of the functionality of the software 9 * licensed hereunder. You may use the software subject to the license 10 * terms below provided that you ensure that this notice is replicated 11 * unmodified and in its entirety in all distributions of the software, 12 * modified or unmodified, in source code or in binary form. 13 * 14 * Copyright (c) 2004-2006 The Regents of The University of Michigan 15 * All rights reserved. 16 * 17 * Redistribution and use in source and binary forms, with or without 18 * modification, are permitted provided that the following conditions are 19 * met: redistributions of source code must retain the above copyright 20 * notice, this list of conditions and the following disclaimer; 21 * redistributions in binary form must reproduce the above copyright 22 * notice, this list of conditions and the following disclaimer in the 23 * documentation and/or other materials provided with the distribution; 24 * neither the name of the copyright holders nor the names of its 25 * contributors may be used to endorse or promote products derived from 26 * this software without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 39 * 40 * Authors: Kevin Lim 41 * Korey Sewell 42 */ 43 44#include <algorithm> 45#include <cstring> 46 47#include "arch/isa_traits.hh" 48#include "arch/utility.hh" 49#include "base/types.hh" 50#include "config/the_isa.hh" 51#include "config/use_checker.hh" 52#include "cpu/checker/cpu.hh" 53#include "cpu/o3/fetch.hh" 54#include "cpu/exetrace.hh" 55#include "debug/Activity.hh" 56#include "debug/Fetch.hh" 57#include "mem/packet.hh" 58#include "mem/request.hh" 59#include "params/DerivO3CPU.hh" 60#include "sim/byteswap.hh" 61#include "sim/core.hh" 62 63#if FULL_SYSTEM 64#include "arch/tlb.hh" 65#include "arch/vtophys.hh" 66#include "sim/system.hh" 67#endif // FULL_SYSTEM 68 69using namespace std; 70 71template<class Impl> 72void 73DefaultFetch<Impl>::IcachePort::setPeer(Port *port) 74{ 75 Port::setPeer(port); 76 77 fetch->setIcache(); 78} 79 80template<class Impl> 81Tick 82DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 83{ 84 panic("DefaultFetch doesn't expect recvAtomic callback!"); 85 return curTick(); 86} 87 88template<class Impl> 89void 90DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 91{ 92 DPRINTF(Fetch, "DefaultFetch doesn't update its state from a " 93 "functional call.\n"); 94} 95 96template<class Impl> 97void 98DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 99{ 100 if (status == RangeChange) { 101 if (!snoopRangeSent) { 102 snoopRangeSent = true; 103 sendStatusChange(Port::RangeChange); 104 } 105 return; 106 } 107 108 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 109} 110 111template<class Impl> 112bool 113DefaultFetch<Impl>::IcachePort::recvTiming(PacketPtr pkt) 114{ 115 DPRINTF(Fetch, "Received timing\n"); 116 if (pkt->isResponse()) { 117 // We shouldn't ever get a block in ownership state 118 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 119 120 fetch->processCacheCompletion(pkt); 121 } 122 //else Snooped a coherence request, just return 123 return true; 124} 125 126template<class Impl> 127void 128DefaultFetch<Impl>::IcachePort::recvRetry() 129{ 130 fetch->recvRetry(); 131} 132 133template<class Impl> 134DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params) 135 : cpu(_cpu), 136 branchPred(params), 137 predecoder(NULL), 138 decodeToFetchDelay(params->decodeToFetchDelay), 139 renameToFetchDelay(params->renameToFetchDelay), 140 iewToFetchDelay(params->iewToFetchDelay), 141 commitToFetchDelay(params->commitToFetchDelay), 142 fetchWidth(params->fetchWidth), 143 cacheBlocked(false), 144 retryPkt(NULL), 145 retryTid(InvalidThreadID), 146 numThreads(params->numThreads), 147 numFetchingThreads(params->smtNumFetchingThreads), 148 interruptPending(false), 149 drainPending(false), 150 switchedOut(false) 151{ 152 if (numThreads > Impl::MaxThreads) 153 fatal("numThreads (%d) is larger than compiled limit (%d),\n" 154 "\tincrease MaxThreads in src/cpu/o3/impl.hh\n", 155 numThreads, static_cast<int>(Impl::MaxThreads)); 156 157 // Set fetch stage's status to inactive. 158 _status = Inactive; 159 160 std::string policy = params->smtFetchPolicy; 161 162 // Convert string to lowercase 163 std::transform(policy.begin(), policy.end(), policy.begin(), 164 (int(*)(int)) tolower); 165 166 // Figure out fetch policy 167 if (policy == "singlethread") { 168 fetchPolicy = SingleThread; 169 if (numThreads > 1) 170 panic("Invalid Fetch Policy for a SMT workload."); 171 } else if (policy == "roundrobin") { 172 fetchPolicy = RoundRobin; 173 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 174 } else if (policy == "branch") { 175 fetchPolicy = Branch; 176 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 177 } else if (policy == "iqcount") { 178 fetchPolicy = IQ; 179 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 180 } else if (policy == "lsqcount") { 181 fetchPolicy = LSQ; 182 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 183 } else { 184 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 185 " RoundRobin,LSQcount,IQcount}\n"); 186 } 187 188 // Get the size of an instruction. 189 instSize = sizeof(TheISA::MachInst); 190 191 // Name is finally available, so create the port. 192 icachePort = new IcachePort(this); 193 194 icachePort->snoopRangeSent = false; 195 196#if USE_CHECKER 197 if (cpu->checker) { 198 cpu->checker->setIcachePort(icachePort); 199 } 200#endif 201} 202 203template <class Impl> 204std::string 205DefaultFetch<Impl>::name() const 206{ 207 return cpu->name() + ".fetch"; 208} 209 210template <class Impl> 211void 212DefaultFetch<Impl>::regStats() 213{ 214 icacheStallCycles 215 .name(name() + ".icacheStallCycles") 216 .desc("Number of cycles fetch is stalled on an Icache miss") 217 .prereq(icacheStallCycles); 218 219 fetchedInsts 220 .name(name() + ".Insts") 221 .desc("Number of instructions fetch has processed") 222 .prereq(fetchedInsts); 223 224 fetchedBranches 225 .name(name() + ".Branches") 226 .desc("Number of branches that fetch encountered") 227 .prereq(fetchedBranches); 228 229 predictedBranches 230 .name(name() + ".predictedBranches") 231 .desc("Number of branches that fetch has predicted taken") 232 .prereq(predictedBranches); 233 234 fetchCycles 235 .name(name() + ".Cycles") 236 .desc("Number of cycles fetch has run and was not squashing or" 237 " blocked") 238 .prereq(fetchCycles); 239 240 fetchSquashCycles 241 .name(name() + ".SquashCycles") 242 .desc("Number of cycles fetch has spent squashing") 243 .prereq(fetchSquashCycles); 244 245 fetchTlbCycles 246 .name(name() + ".TlbCycles") 247 .desc("Number of cycles fetch has spent waiting for tlb") 248 .prereq(fetchTlbCycles); 249 250 fetchIdleCycles 251 .name(name() + ".IdleCycles") 252 .desc("Number of cycles fetch was idle") 253 .prereq(fetchIdleCycles); 254 255 fetchBlockedCycles 256 .name(name() + ".BlockedCycles") 257 .desc("Number of cycles fetch has spent blocked") 258 .prereq(fetchBlockedCycles); 259 260 fetchedCacheLines 261 .name(name() + ".CacheLines") 262 .desc("Number of cache lines fetched") 263 .prereq(fetchedCacheLines); 264 265 fetchMiscStallCycles 266 .name(name() + ".MiscStallCycles") 267 .desc("Number of cycles fetch has spent waiting on interrupts, or " 268 "bad addresses, or out of MSHRs") 269 .prereq(fetchMiscStallCycles); 270 271 fetchIcacheSquashes 272 .name(name() + ".IcacheSquashes") 273 .desc("Number of outstanding Icache misses that were squashed") 274 .prereq(fetchIcacheSquashes); 275 276 fetchTlbSquashes 277 .name(name() + ".ItlbSquashes") 278 .desc("Number of outstanding ITLB misses that were squashed") 279 .prereq(fetchTlbSquashes); 280 281 fetchNisnDist 282 .init(/* base value */ 0, 283 /* last value */ fetchWidth, 284 /* bucket size */ 1) 285 .name(name() + ".rateDist") 286 .desc("Number of instructions fetched each cycle (Total)") 287 .flags(Stats::pdf); 288 289 idleRate 290 .name(name() + ".idleRate") 291 .desc("Percent of cycles fetch was idle") 292 .prereq(idleRate); 293 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 294 295 branchRate 296 .name(name() + ".branchRate") 297 .desc("Number of branch fetches per cycle") 298 .flags(Stats::total); 299 branchRate = fetchedBranches / cpu->numCycles; 300 301 fetchRate 302 .name(name() + ".rate") 303 .desc("Number of inst fetches per cycle") 304 .flags(Stats::total); 305 fetchRate = fetchedInsts / cpu->numCycles; 306 307 branchPred.regStats(); 308} 309 310template<class Impl> 311void 312DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 313{ 314 timeBuffer = time_buffer; 315 316 // Create wires to get information from proper places in time buffer. 317 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 318 fromRename = timeBuffer->getWire(-renameToFetchDelay); 319 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 320 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 321} 322 323template<class Impl> 324void 325DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr) 326{ 327 activeThreads = at_ptr; 328} 329 330template<class Impl> 331void 332DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 333{ 334 fetchQueue = fq_ptr; 335 336 // Create wire to write information to proper place in fetch queue. 337 toDecode = fetchQueue->getWire(0); 338} 339 340template<class Impl> 341void 342DefaultFetch<Impl>::initStage() 343{ 344 // Setup PC and nextPC with initial state. 345 for (ThreadID tid = 0; tid < numThreads; tid++) { 346 pc[tid] = cpu->pcState(tid); 347 fetchOffset[tid] = 0; 348 macroop[tid] = NULL; 349 delayedCommit[tid] = false; 350 } 351 352 for (ThreadID tid = 0; tid < numThreads; tid++) { 353 354 fetchStatus[tid] = Running; 355 356 priorityList.push_back(tid); 357 358 memReq[tid] = NULL; 359 360 stalls[tid].decode = false; 361 stalls[tid].rename = false; 362 stalls[tid].iew = false; 363 stalls[tid].commit = false; 364 } 365 366 // Schedule fetch to get the correct PC from the CPU 367 // scheduleFetchStartupEvent(1); 368 369 // Fetch needs to start fetching instructions at the very beginning, 370 // so it must start up in active state. 371 switchToActive(); 372} 373 374template<class Impl> 375void 376DefaultFetch<Impl>::setIcache() 377{ 378 // Size of cache block. 379 cacheBlkSize = icachePort->peerBlockSize(); 380 381 // Create mask to get rid of offset bits. 382 cacheBlkMask = (cacheBlkSize - 1); 383 384 for (ThreadID tid = 0; tid < numThreads; tid++) { 385 // Create space to store a cache line. 386 cacheData[tid] = new uint8_t[cacheBlkSize]; 387 cacheDataPC[tid] = 0; 388 cacheDataValid[tid] = false; 389 } 390} 391 392template<class Impl> 393void 394DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 395{ 396 ThreadID tid = pkt->req->threadId(); 397 398 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n", tid); 399 400 assert(!pkt->wasNacked()); 401 402 // Only change the status if it's still waiting on the icache access 403 // to return. 404 if (fetchStatus[tid] != IcacheWaitResponse || 405 pkt->req != memReq[tid] || 406 isSwitchedOut()) { 407 ++fetchIcacheSquashes; 408 delete pkt->req; 409 delete pkt; 410 return; 411 } 412 413 memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize); 414 cacheDataValid[tid] = true; 415 416 if (!drainPending) { 417 // Wake up the CPU (if it went to sleep and was waiting on 418 // this completion event). 419 cpu->wakeCPU(); 420 421 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 422 tid); 423 424 switchToActive(); 425 } 426 427 // Only switch to IcacheAccessComplete if we're not stalled as well. 428 if (checkStall(tid)) { 429 fetchStatus[tid] = Blocked; 430 } else { 431 fetchStatus[tid] = IcacheAccessComplete; 432 } 433 434 // Reset the mem req to NULL. 435 delete pkt->req; 436 delete pkt; 437 memReq[tid] = NULL; 438} 439 440template <class Impl> 441bool 442DefaultFetch<Impl>::drain() 443{ 444 // Fetch is ready to drain at any time. 445 cpu->signalDrained(); 446 drainPending = true; 447 return true; 448} 449 450template <class Impl> 451void 452DefaultFetch<Impl>::resume() 453{ 454 drainPending = false; 455} 456 457template <class Impl> 458void 459DefaultFetch<Impl>::switchOut() 460{ 461 switchedOut = true; 462 // Branch predictor needs to have its state cleared. 463 branchPred.switchOut(); 464} 465 466template <class Impl> 467void 468DefaultFetch<Impl>::takeOverFrom() 469{ 470 // Reset all state 471 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 472 stalls[i].decode = 0; 473 stalls[i].rename = 0; 474 stalls[i].iew = 0; 475 stalls[i].commit = 0; 476 pc[i] = cpu->pcState(i); 477 fetchStatus[i] = Running; 478 } 479 numInst = 0; 480 wroteToTimeBuffer = false; 481 _status = Inactive; 482 switchedOut = false; 483 interruptPending = false; 484 branchPred.takeOverFrom(); 485} 486 487template <class Impl> 488void 489DefaultFetch<Impl>::wakeFromQuiesce() 490{ 491 DPRINTF(Fetch, "Waking up from quiesce\n"); 492 // Hopefully this is safe 493 // @todo: Allow other threads to wake from quiesce. 494 fetchStatus[0] = Running; 495} 496 497template <class Impl> 498inline void 499DefaultFetch<Impl>::switchToActive() 500{ 501 if (_status == Inactive) { 502 DPRINTF(Activity, "Activating stage.\n"); 503 504 cpu->activateStage(O3CPU::FetchIdx); 505 506 _status = Active; 507 } 508} 509 510template <class Impl> 511inline void 512DefaultFetch<Impl>::switchToInactive() 513{ 514 if (_status == Active) { 515 DPRINTF(Activity, "Deactivating stage.\n"); 516 517 cpu->deactivateStage(O3CPU::FetchIdx); 518 519 _status = Inactive; 520 } 521} 522 523template <class Impl> 524bool 525DefaultFetch<Impl>::lookupAndUpdateNextPC( 526 DynInstPtr &inst, TheISA::PCState &nextPC) 527{ 528 // Do branch prediction check here. 529 // A bit of a misnomer...next_PC is actually the current PC until 530 // this function updates it. 531 bool predict_taken; 532 533 if (!inst->isControl()) { 534 TheISA::advancePC(nextPC, inst->staticInst); 535 inst->setPredTarg(nextPC); 536 inst->setPredTaken(false); 537 return false; 538 } 539 540 ThreadID tid = inst->threadNumber; 541 predict_taken = branchPred.predict(inst, nextPC, tid); 542 543 if (predict_taken) { 544 DPRINTF(Fetch, "[tid:%i]: [sn:%i]: Branch predicted to be taken to %s.\n", 545 tid, inst->seqNum, nextPC); 546 } else { 547 DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n", 548 tid, inst->seqNum); 549 } 550 551 DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n", 552 tid, inst->seqNum, nextPC); 553 inst->setPredTarg(nextPC); 554 inst->setPredTaken(predict_taken); 555 556 ++fetchedBranches; 557 558 if (predict_taken) { 559 ++predictedBranches; 560 } 561 562 return predict_taken; 563} 564 565template <class Impl> 566bool 567DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc) 568{ 569 Fault fault = NoFault; 570 571 // @todo: not sure if these should block translation. 572 //AlphaDep 573 if (cacheBlocked) { 574 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 575 tid); 576 return false; 577 } else if (isSwitchedOut()) { 578 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 579 tid); 580 return false; 581 } else if (checkInterrupt(pc)) { 582 // Hold off fetch from getting new instructions when: 583 // Cache is blocked, or 584 // while an interrupt is pending and we're not in PAL mode, or 585 // fetch is switched out. 586 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 587 tid); 588 return false; 589 } 590 591 // Align the fetch address so it's at the start of a cache block. 592 Addr block_PC = icacheBlockAlignPC(vaddr); 593 594 DPRINTF(Fetch, "[tid:%i] Fetching cache line %#x for addr %#x\n", 595 tid, block_PC, vaddr); 596 597 // Setup the memReq to do a read of the first instruction's address. 598 // Set the appropriate read size and flags as well. 599 // Build request here. 600 RequestPtr mem_req = 601 new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH, 602 pc, cpu->thread[tid]->contextId(), tid); 603 604 memReq[tid] = mem_req; 605 606 // Initiate translation of the icache block 607 fetchStatus[tid] = ItlbWait; 608 FetchTranslation *trans = new FetchTranslation(this); 609 cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(), 610 trans, BaseTLB::Execute); 611 return true; 612} 613 614template <class Impl> 615void 616DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req) 617{ 618 ThreadID tid = mem_req->threadId(); 619 Addr block_PC = mem_req->getVaddr(); 620 621 // Wake up CPU if it was idle 622 cpu->wakeCPU(); 623 624 if (fetchStatus[tid] != ItlbWait || mem_req != memReq[tid] || 625 mem_req->getVaddr() != memReq[tid]->getVaddr() || isSwitchedOut()) { 626 DPRINTF(Fetch, "[tid:%i] Ignoring itlb completed after squash\n", 627 tid); 628 ++fetchTlbSquashes; 629 delete mem_req; 630 return; 631 } 632 633 634 // If translation was successful, attempt to read the icache block. 635 if (fault == NoFault) { 636 // Check that we're not going off into random memory 637 // If we have, just wait around for commit to squash something and put 638 // us on the right track 639 if (!cpu->system->isMemory(mem_req->getPaddr())) { 640 warn("Address %#x is outside of physical memory, stopping fetch\n", 641 mem_req->getPaddr()); 642 fetchStatus[tid] = NoGoodAddr; 643 delete mem_req; 644 memReq[tid] = NULL; 645 return; 646 } 647 648 // Build packet here. 649 PacketPtr data_pkt = new Packet(mem_req, 650 MemCmd::ReadReq, Packet::Broadcast); 651 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 652 653 cacheDataPC[tid] = block_PC; 654 cacheDataValid[tid] = false; 655 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 656 657 fetchedCacheLines++; 658 659 // Access the cache. 660 if (!icachePort->sendTiming(data_pkt)) { 661 assert(retryPkt == NULL); 662 assert(retryTid == InvalidThreadID); 663 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 664 665 fetchStatus[tid] = IcacheWaitRetry; 666 retryPkt = data_pkt; 667 retryTid = tid; 668 cacheBlocked = true; 669 } else { 670 DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid); 671 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 672 "response.\n", tid); 673 674 lastIcacheStall[tid] = curTick(); 675 fetchStatus[tid] = IcacheWaitResponse; 676 } 677 } else { 678 DPRINTF(Fetch, "[tid:%i] Got back req with addr %#x but expected %#x\n", 679 mem_req->getVaddr(), memReq[tid]->getVaddr()); 680 // Translation faulted, icache request won't be sent. 681 delete mem_req; 682 memReq[tid] = NULL; 683 684 // Send the fault to commit. This thread will not do anything 685 // until commit handles the fault. The only other way it can 686 // wake up is if a squash comes along and changes the PC. 687 TheISA::PCState fetchPC = pc[tid]; 688 689 DPRINTF(Fetch, "[tid:%i]: Translation faulted, building noop.\n", tid); 690 // We will use a nop in ordier to carry the fault. 691 DynInstPtr instruction = buildInst(tid, 692 StaticInstPtr(TheISA::NoopMachInst, fetchPC.instAddr()), 693 NULL, fetchPC, fetchPC, false); 694 695 instruction->setPredTarg(fetchPC); 696 instruction->fault = fault; 697 wroteToTimeBuffer = true; 698 699 DPRINTF(Activity, "Activity this cycle.\n"); 700 cpu->activityThisCycle(); 701 702 fetchStatus[tid] = TrapPending; 703 704 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid); 705 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n", 706 tid, fault->name(), pc[tid]); 707 } 708 _status = updateFetchStatus(); 709} 710 711template <class Impl> 712inline void 713DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC, ThreadID tid) 714{ 715 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n", 716 tid, newPC); 717 718 pc[tid] = newPC; 719 fetchOffset[tid] = 0; 720 macroop[tid] = NULL; 721 predecoder.reset(); 722 723 // Clear the icache miss if it's outstanding. 724 if (fetchStatus[tid] == IcacheWaitResponse) { 725 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 726 tid); 727 memReq[tid] = NULL; 728 } else if (fetchStatus[tid] == ItlbWait) { 729 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding ITLB miss.\n", 730 tid); 731 memReq[tid] = NULL; 732 } 733 734 // Get rid of the retrying packet if it was from this thread. 735 if (retryTid == tid) { 736 assert(cacheBlocked); 737 if (retryPkt) { 738 delete retryPkt->req; 739 delete retryPkt; 740 } 741 retryPkt = NULL; 742 retryTid = InvalidThreadID; 743 } 744 745 fetchStatus[tid] = Squashing; 746 747 ++fetchSquashCycles; 748} 749 750template<class Impl> 751void 752DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC, 753 const InstSeqNum &seq_num, ThreadID tid) 754{ 755 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid); 756 757 doSquash(newPC, tid); 758 759 // Tell the CPU to remove any instructions that are in flight between 760 // fetch and decode. 761 cpu->removeInstsUntil(seq_num, tid); 762} 763 764template<class Impl> 765bool 766DefaultFetch<Impl>::checkStall(ThreadID tid) const 767{ 768 bool ret_val = false; 769 770 if (cpu->contextSwitch) { 771 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 772 ret_val = true; 773 } else if (stalls[tid].decode) { 774 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 775 ret_val = true; 776 } else if (stalls[tid].rename) { 777 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 778 ret_val = true; 779 } else if (stalls[tid].iew) { 780 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 781 ret_val = true; 782 } else if (stalls[tid].commit) { 783 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 784 ret_val = true; 785 } 786 787 return ret_val; 788} 789 790template<class Impl> 791typename DefaultFetch<Impl>::FetchStatus 792DefaultFetch<Impl>::updateFetchStatus() 793{ 794 //Check Running 795 list<ThreadID>::iterator threads = activeThreads->begin(); 796 list<ThreadID>::iterator end = activeThreads->end(); 797 798 while (threads != end) { 799 ThreadID tid = *threads++; 800 801 if (fetchStatus[tid] == Running || 802 fetchStatus[tid] == Squashing || 803 fetchStatus[tid] == IcacheAccessComplete) { 804 805 if (_status == Inactive) { 806 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 807 808 if (fetchStatus[tid] == IcacheAccessComplete) { 809 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 810 "completion\n",tid); 811 } 812 813 cpu->activateStage(O3CPU::FetchIdx); 814 } 815 816 return Active; 817 } 818 } 819 820 // Stage is switching from active to inactive, notify CPU of it. 821 if (_status == Active) { 822 DPRINTF(Activity, "Deactivating stage.\n"); 823 824 cpu->deactivateStage(O3CPU::FetchIdx); 825 } 826 827 return Inactive; 828} 829 830template <class Impl> 831void 832DefaultFetch<Impl>::squash(const TheISA::PCState &newPC, 833 const InstSeqNum &seq_num, DynInstPtr &squashInst, 834 ThreadID tid) 835{ 836 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid); 837 838 doSquash(newPC, tid); 839 840 // Tell the CPU to remove any instructions that are not in the ROB. 841 cpu->removeInstsNotInROB(tid); 842} 843 844template <class Impl> 845void 846DefaultFetch<Impl>::tick() 847{ 848 list<ThreadID>::iterator threads = activeThreads->begin(); 849 list<ThreadID>::iterator end = activeThreads->end(); 850 bool status_change = false; 851 852 wroteToTimeBuffer = false; 853 854 while (threads != end) { 855 ThreadID tid = *threads++; 856 857 // Check the signals for each thread to determine the proper status 858 // for each thread. 859 bool updated_status = checkSignalsAndUpdate(tid); 860 status_change = status_change || updated_status; 861 } 862 863 DPRINTF(Fetch, "Running stage.\n"); 864 865 // Reset the number of the instruction we're fetching. 866 numInst = 0; 867 868#if FULL_SYSTEM 869 if (fromCommit->commitInfo[0].interruptPending) { 870 interruptPending = true; 871 } 872 873 if (fromCommit->commitInfo[0].clearInterrupt) { 874 interruptPending = false; 875 } 876#endif 877 878 for (threadFetched = 0; threadFetched < numFetchingThreads; 879 threadFetched++) { 880 // Fetch each of the actively fetching threads. 881 fetch(status_change); 882 } 883 884 // Record number of instructions fetched this cycle for distribution. 885 fetchNisnDist.sample(numInst); 886 887 if (status_change) { 888 // Change the fetch stage status if there was a status change. 889 _status = updateFetchStatus(); 890 } 891 892 // If there was activity this cycle, inform the CPU of it. 893 if (wroteToTimeBuffer || cpu->contextSwitch) { 894 DPRINTF(Activity, "Activity this cycle.\n"); 895 896 cpu->activityThisCycle(); 897 } 898} 899 900template <class Impl> 901bool 902DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid) 903{ 904 // Update the per thread stall statuses. 905 if (fromDecode->decodeBlock[tid]) { 906 stalls[tid].decode = true; 907 } 908 909 if (fromDecode->decodeUnblock[tid]) { 910 assert(stalls[tid].decode); 911 assert(!fromDecode->decodeBlock[tid]); 912 stalls[tid].decode = false; 913 } 914 915 if (fromRename->renameBlock[tid]) { 916 stalls[tid].rename = true; 917 } 918 919 if (fromRename->renameUnblock[tid]) { 920 assert(stalls[tid].rename); 921 assert(!fromRename->renameBlock[tid]); 922 stalls[tid].rename = false; 923 } 924 925 if (fromIEW->iewBlock[tid]) { 926 stalls[tid].iew = true; 927 } 928 929 if (fromIEW->iewUnblock[tid]) { 930 assert(stalls[tid].iew); 931 assert(!fromIEW->iewBlock[tid]); 932 stalls[tid].iew = false; 933 } 934 935 if (fromCommit->commitBlock[tid]) { 936 stalls[tid].commit = true; 937 } 938 939 if (fromCommit->commitUnblock[tid]) { 940 assert(stalls[tid].commit); 941 assert(!fromCommit->commitBlock[tid]); 942 stalls[tid].commit = false; 943 } 944 945 // Check squash signals from commit. 946 if (fromCommit->commitInfo[tid].squash) { 947 948 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 949 "from commit.\n",tid); 950 // In any case, squash. 951 squash(fromCommit->commitInfo[tid].pc, 952 fromCommit->commitInfo[tid].doneSeqNum, 953 fromCommit->commitInfo[tid].squashInst, tid); 954 955 // If it was a branch mispredict on a control instruction, update the 956 // branch predictor with that instruction, otherwise just kill the 957 // invalid state we generated in after sequence number 958 if (fromCommit->commitInfo[tid].mispredictInst && 959 fromCommit->commitInfo[tid].mispredictInst->isControl()) { 960 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 961 fromCommit->commitInfo[tid].pc, 962 fromCommit->commitInfo[tid].branchTaken, 963 tid); 964 } else { 965 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 966 tid); 967 } 968 969 return true; 970 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 971 // Update the branch predictor if it wasn't a squashed instruction 972 // that was broadcasted. 973 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 974 } 975 976 // Check ROB squash signals from commit. 977 if (fromCommit->commitInfo[tid].robSquashing) { 978 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 979 980 // Continue to squash. 981 fetchStatus[tid] = Squashing; 982 983 return true; 984 } 985 986 // Check squash signals from decode. 987 if (fromDecode->decodeInfo[tid].squash) { 988 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 989 "from decode.\n",tid); 990 991 // Update the branch predictor. 992 if (fromDecode->decodeInfo[tid].branchMispredict) { 993 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 994 fromDecode->decodeInfo[tid].nextPC, 995 fromDecode->decodeInfo[tid].branchTaken, 996 tid); 997 } else { 998 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 999 tid); 1000 } 1001 1002 if (fetchStatus[tid] != Squashing) { 1003 1004 TheISA::PCState nextPC = fromDecode->decodeInfo[tid].nextPC; 1005 DPRINTF(Fetch, "Squashing from decode with PC = %s\n", nextPC); 1006 // Squash unless we're already squashing 1007 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 1008 fromDecode->decodeInfo[tid].doneSeqNum, 1009 tid); 1010 1011 return true; 1012 } 1013 } 1014 1015 if (checkStall(tid) && 1016 fetchStatus[tid] != IcacheWaitResponse && 1017 fetchStatus[tid] != IcacheWaitRetry) { 1018 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 1019 1020 fetchStatus[tid] = Blocked; 1021 1022 return true; 1023 } 1024 1025 if (fetchStatus[tid] == Blocked || 1026 fetchStatus[tid] == Squashing) { 1027 // Switch status to running if fetch isn't being told to block or 1028 // squash this cycle. 1029 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1030 tid); 1031 1032 fetchStatus[tid] = Running; 1033 1034 return true; 1035 } 1036 1037 // If we've reached this point, we have not gotten any signals that 1038 // cause fetch to change its status. Fetch remains the same as before. 1039 return false; 1040} 1041 1042template<class Impl> 1043typename Impl::DynInstPtr 1044DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst, 1045 StaticInstPtr curMacroop, TheISA::PCState thisPC, 1046 TheISA::PCState nextPC, bool trace) 1047{ 1048 // Get a sequence number. 1049 InstSeqNum seq = cpu->getAndIncrementInstSeq(); 1050 1051 // Create a new DynInst from the instruction fetched. 1052 DynInstPtr instruction = 1053 new DynInst(staticInst, thisPC, nextPC, seq, cpu); 1054 instruction->setTid(tid); 1055 1056 instruction->setASID(tid); 1057 1058 instruction->setThreadState(cpu->thread[tid]); 1059 1060 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created " 1061 "[sn:%lli].\n", tid, thisPC.instAddr(), 1062 thisPC.microPC(), seq); 1063 1064 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid, 1065 instruction->staticInst-> 1066 disassemble(thisPC.instAddr())); 1067 1068#if TRACING_ON 1069 if (trace) { 1070 instruction->traceData = 1071 cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid), 1072 instruction->staticInst, thisPC, curMacroop); 1073 } 1074#else 1075 instruction->traceData = NULL; 1076#endif 1077 1078 // Add instruction to the CPU's list of instructions. 1079 instruction->setInstListIt(cpu->addInst(instruction)); 1080 1081 // Write the instruction to the first slot in the queue 1082 // that heads to decode. 1083 assert(numInst < fetchWidth); 1084 toDecode->insts[toDecode->size++] = instruction; 1085 1086 // Keep track of if we can take an interrupt at this boundary 1087 delayedCommit[tid] = instruction->isDelayedCommit(); 1088 1089 return instruction; 1090} 1091 1092template<class Impl> 1093void 1094DefaultFetch<Impl>::fetch(bool &status_change) 1095{ 1096 ////////////////////////////////////////// 1097 // Start actual fetch 1098 ////////////////////////////////////////// 1099 ThreadID tid = getFetchingThread(fetchPolicy); 1100 1101 if (tid == InvalidThreadID || drainPending) { 1102 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1103 1104 // Breaks looping condition in tick() 1105 threadFetched = numFetchingThreads; 1106 return; 1107 } 1108 1109 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1110 1111 // The current PC. 1112 TheISA::PCState thisPC = pc[tid]; 1113 1114 Addr pcOffset = fetchOffset[tid]; 1115 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1116 1117 bool inRom = isRomMicroPC(thisPC.microPC()); 1118 1119 // If returning from the delay of a cache miss, then update the status 1120 // to running, otherwise do the cache access. Possibly move this up 1121 // to tick() function. 1122 if (fetchStatus[tid] == IcacheAccessComplete) { 1123 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid); 1124 1125 fetchStatus[tid] = Running; 1126 status_change = true; 1127 } else if (fetchStatus[tid] == Running) { 1128 // Align the fetch PC so its at the start of a cache block. 1129 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1130 1131 // If buffer is no longer valid or fetchAddr has moved to point 1132 // to the next cache block, AND we have no remaining ucode 1133 // from a macro-op, then start fetch from icache. 1134 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid]) 1135 && !inRom && !macroop[tid]) { 1136 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1137 "instruction, starting at PC %s.\n", tid, thisPC); 1138 1139 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1140 1141 if (fetchStatus[tid] == IcacheWaitResponse) 1142 ++icacheStallCycles; 1143 else if (fetchStatus[tid] == ItlbWait) 1144 ++fetchTlbCycles; 1145 else 1146 ++fetchMiscStallCycles; 1147 return; 1148 } else if ((checkInterrupt(thisPC.instAddr()) && !delayedCommit[tid]) 1149 || isSwitchedOut()) { 1150 // Stall CPU if an interrupt is posted and we're not issuing 1151 // an delayed commit micro-op currently (delayed commit instructions 1152 // are not interruptable by interrupts, only faults) 1153 ++fetchMiscStallCycles; 1154 return; 1155 } 1156 } else { 1157 if (fetchStatus[tid] == Idle) { 1158 ++fetchIdleCycles; 1159 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1160 } else if (fetchStatus[tid] == Blocked) { 1161 ++fetchBlockedCycles; 1162 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1163 } else if (fetchStatus[tid] == Squashing) { 1164 ++fetchSquashCycles; 1165 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1166 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1167 ++icacheStallCycles; 1168 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", 1169 tid); 1170 } else if (fetchStatus[tid] == ItlbWait) { 1171 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to " 1172 "finish! \n", tid); 1173 ++fetchTlbCycles; 1174 } else if (fetchStatus[tid] == TrapPending) { 1175 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending trap\n", 1176 tid); 1177 } else if (fetchStatus[tid] == NoGoodAddr) { 1178 DPRINTF(Fetch, "[tid:%i]: Fetch predicted non-executable address\n", 1179 tid); 1180 } 1181 1182 1183 1184 // Status is Idle, Squashing, Blocked, ItlbWait or IcacheWaitResponse 1185 // so fetch should do nothing. 1186 return; 1187 } 1188 1189 ++fetchCycles; 1190 1191 TheISA::PCState nextPC = thisPC; 1192 1193 StaticInstPtr staticInst = NULL; 1194 StaticInstPtr curMacroop = macroop[tid]; 1195 1196 // If the read of the first instruction was successful, then grab the 1197 // instructions from the rest of the cache line and put them into the 1198 // queue heading to decode. 1199 1200 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1201 "decode.\n", tid); 1202 1203 // Need to keep track of whether or not a predicted branch 1204 // ended this fetch block. 1205 bool predictedBranch = false; 1206 1207 TheISA::MachInst *cacheInsts = 1208 reinterpret_cast<TheISA::MachInst *>(cacheData[tid]); 1209 1210 const unsigned numInsts = cacheBlkSize / instSize; 1211 unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1212 1213 // Loop through instruction memory from the cache. 1214 // Keep issuing while we have not reached the end of the block or a 1215 // macroop is active and fetchWidth is available and branch is not 1216 // predicted taken 1217 while ((blkOffset < numInsts || curMacroop) && 1218 numInst < fetchWidth && !predictedBranch) { 1219 1220 // If we need to process more memory, do it now. 1221 if (!(curMacroop || inRom) && !predecoder.extMachInstReady()) { 1222 if (ISA_HAS_DELAY_SLOT && pcOffset == 0) { 1223 // Walk past any annulled delay slot instructions. 1224 Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask; 1225 while (fetchAddr != pcAddr && blkOffset < numInsts) { 1226 blkOffset++; 1227 fetchAddr += instSize; 1228 } 1229 if (blkOffset >= numInsts) 1230 break; 1231 } 1232 MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]); 1233 1234 predecoder.setTC(cpu->thread[tid]->getTC()); 1235 predecoder.moreBytes(thisPC, fetchAddr, inst); 1236 1237 if (predecoder.needMoreBytes()) { 1238 blkOffset++; 1239 fetchAddr += instSize; 1240 pcOffset += instSize; 1241 } 1242 } 1243 1244 // Extract as many instructions and/or microops as we can from 1245 // the memory we've processed so far. 1246 do { 1247 if (!(curMacroop || inRom)) { 1248 if (predecoder.extMachInstReady()) { 1249 ExtMachInst extMachInst; 1250 1251 extMachInst = predecoder.getExtMachInst(thisPC); 1252 staticInst = StaticInstPtr(extMachInst, 1253 thisPC.instAddr()); 1254 1255 // Increment stat of fetched instructions. 1256 ++fetchedInsts; 1257 1258 if (staticInst->isMacroop()) { 1259 curMacroop = staticInst; 1260 } else { 1261 pcOffset = 0; 1262 } 1263 } else { 1264 // We need more bytes for this instruction so blkOffset and 1265 // pcOffset will be updated 1266 break; 1267 } 1268 } 1269 if (curMacroop || inRom) { 1270 if (inRom) { 1271 staticInst = cpu->microcodeRom.fetchMicroop( 1272 thisPC.microPC(), curMacroop); 1273 } else { 1274 staticInst = curMacroop->fetchMicroop(thisPC.microPC()); 1275 } 1276 if (staticInst->isLastMicroop()) { 1277 curMacroop = NULL; 1278 pcOffset = 0; 1279 } 1280 } 1281 1282 DynInstPtr instruction = 1283 buildInst(tid, staticInst, curMacroop, 1284 thisPC, nextPC, true); 1285 1286 numInst++; 1287 1288 nextPC = thisPC; 1289 1290 // If we're branching after this instruction, quite fetching 1291 // from the same block then. 1292 predictedBranch |= thisPC.branching(); 1293 predictedBranch |= 1294 lookupAndUpdateNextPC(instruction, nextPC); 1295 if (predictedBranch) { 1296 DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC); 1297 } 1298 1299 // Move to the next instruction, unless we have a branch. 1300 thisPC = nextPC; 1301 1302 if (instruction->isQuiesce()) { 1303 DPRINTF(Fetch, 1304 "Quiesce instruction encountered, halting fetch!"); 1305 fetchStatus[tid] = QuiescePending; 1306 status_change = true; 1307 break; 1308 } 1309 } while ((curMacroop || predecoder.extMachInstReady()) && 1310 numInst < fetchWidth); 1311 } 1312 1313 if (predictedBranch) { 1314 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1315 "instruction encountered.\n", tid); 1316 } else if (numInst >= fetchWidth) { 1317 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1318 "for this cycle.\n", tid); 1319 } else if (blkOffset >= cacheBlkSize) { 1320 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1321 "block.\n", tid); 1322 } 1323 1324 macroop[tid] = curMacroop; 1325 fetchOffset[tid] = pcOffset; 1326 1327 if (numInst > 0) { 1328 wroteToTimeBuffer = true; 1329 } 1330 1331 pc[tid] = thisPC; 1332} 1333 1334template<class Impl> 1335void 1336DefaultFetch<Impl>::recvRetry() 1337{ 1338 if (retryPkt != NULL) { 1339 assert(cacheBlocked); 1340 assert(retryTid != InvalidThreadID); 1341 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1342 1343 if (icachePort->sendTiming(retryPkt)) { 1344 fetchStatus[retryTid] = IcacheWaitResponse; 1345 retryPkt = NULL; 1346 retryTid = InvalidThreadID; 1347 cacheBlocked = false; 1348 } 1349 } else { 1350 assert(retryTid == InvalidThreadID); 1351 // Access has been squashed since it was sent out. Just clear 1352 // the cache being blocked. 1353 cacheBlocked = false; 1354 } 1355} 1356 1357/////////////////////////////////////// 1358// // 1359// SMT FETCH POLICY MAINTAINED HERE // 1360// // 1361/////////////////////////////////////// 1362template<class Impl> 1363ThreadID 1364DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1365{ 1366 if (numThreads > 1) { 1367 switch (fetch_priority) { 1368 1369 case SingleThread: 1370 return 0; 1371 1372 case RoundRobin: 1373 return roundRobin(); 1374 1375 case IQ: 1376 return iqCount(); 1377 1378 case LSQ: 1379 return lsqCount(); 1380 1381 case Branch: 1382 return branchCount(); 1383 1384 default: 1385 return InvalidThreadID; 1386 } 1387 } else { 1388 list<ThreadID>::iterator thread = activeThreads->begin(); 1389 if (thread == activeThreads->end()) { 1390 return InvalidThreadID; 1391 } 1392 1393 ThreadID tid = *thread; 1394 1395 if (fetchStatus[tid] == Running || 1396 fetchStatus[tid] == IcacheAccessComplete || 1397 fetchStatus[tid] == Idle) { 1398 return tid; 1399 } else { 1400 return InvalidThreadID; 1401 } 1402 } 1403} 1404 1405 1406template<class Impl> 1407ThreadID 1408DefaultFetch<Impl>::roundRobin() 1409{ 1410 list<ThreadID>::iterator pri_iter = priorityList.begin(); 1411 list<ThreadID>::iterator end = priorityList.end(); 1412 1413 ThreadID high_pri; 1414 1415 while (pri_iter != end) { 1416 high_pri = *pri_iter; 1417 1418 assert(high_pri <= numThreads); 1419 1420 if (fetchStatus[high_pri] == Running || 1421 fetchStatus[high_pri] == IcacheAccessComplete || 1422 fetchStatus[high_pri] == Idle) { 1423 1424 priorityList.erase(pri_iter); 1425 priorityList.push_back(high_pri); 1426 1427 return high_pri; 1428 } 1429 1430 pri_iter++; 1431 } 1432 1433 return InvalidThreadID; 1434} 1435 1436template<class Impl> 1437ThreadID 1438DefaultFetch<Impl>::iqCount() 1439{ 1440 std::priority_queue<unsigned> PQ; 1441 std::map<unsigned, ThreadID> threadMap; 1442 1443 list<ThreadID>::iterator threads = activeThreads->begin(); 1444 list<ThreadID>::iterator end = activeThreads->end(); 1445 1446 while (threads != end) { 1447 ThreadID tid = *threads++; 1448 unsigned iqCount = fromIEW->iewInfo[tid].iqCount; 1449 1450 PQ.push(iqCount); 1451 threadMap[iqCount] = tid; 1452 } 1453 1454 while (!PQ.empty()) { 1455 ThreadID high_pri = threadMap[PQ.top()]; 1456 1457 if (fetchStatus[high_pri] == Running || 1458 fetchStatus[high_pri] == IcacheAccessComplete || 1459 fetchStatus[high_pri] == Idle) 1460 return high_pri; 1461 else 1462 PQ.pop(); 1463 1464 } 1465 1466 return InvalidThreadID; 1467} 1468 1469template<class Impl> 1470ThreadID 1471DefaultFetch<Impl>::lsqCount() 1472{ 1473 std::priority_queue<unsigned> PQ; 1474 std::map<unsigned, ThreadID> threadMap; 1475 1476 list<ThreadID>::iterator threads = activeThreads->begin(); 1477 list<ThreadID>::iterator end = activeThreads->end(); 1478 1479 while (threads != end) { 1480 ThreadID tid = *threads++; 1481 unsigned ldstqCount = fromIEW->iewInfo[tid].ldstqCount; 1482 1483 PQ.push(ldstqCount); 1484 threadMap[ldstqCount] = tid; 1485 } 1486 1487 while (!PQ.empty()) { 1488 ThreadID high_pri = threadMap[PQ.top()]; 1489 1490 if (fetchStatus[high_pri] == Running || 1491 fetchStatus[high_pri] == IcacheAccessComplete || 1492 fetchStatus[high_pri] == Idle) 1493 return high_pri; 1494 else 1495 PQ.pop(); 1496 } 1497 1498 return InvalidThreadID; 1499} 1500 1501template<class Impl> 1502ThreadID 1503DefaultFetch<Impl>::branchCount() 1504{ 1505#if 0 1506 list<ThreadID>::iterator thread = activeThreads->begin(); 1507 assert(thread != activeThreads->end()); 1508 ThreadID tid = *thread; 1509#endif 1510 1511 panic("Branch Count Fetch policy unimplemented\n"); 1512 return InvalidThreadID; 1513} 1514