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;

--- 305 unchanged lines hidden (view full) ---

314template<class Impl>
315void
316DefaultFetch<Impl>::initStage()
317{
318 // Setup PC and nextPC with initial state.
319 for (int tid = 0; tid < numThreads; tid++) {
320 PC[tid] = cpu->readPC(tid);
321 nextPC[tid] = cpu->readNextPC(tid);
322 nextNPC[tid] = cpu->readNextNPC(tid);
323 }
324
325 // Size of cache block.
326 cacheBlkSize = icachePort->peerBlockSize();
327
328 // Create mask to get rid of offset bits.
329 cacheBlkMask = (cacheBlkSize - 1);
330

--- 166 unchanged lines hidden (view full) ---

497{
498 // Do branch prediction check here.
499 // A bit of a misnomer...next_PC is actually the current PC until
500 // this function updates it.
501 bool predict_taken;
502
503 if (!inst->isControl()) {
504#if ISA_HAS_DELAY_SLOT
505 next_PC = next_NPC;
506 next_NPC = next_NPC + instSize;
507 inst->setPredTarg(next_PC, next_NPC);
508#else
509 next_PC = next_PC + instSize;
510 inst->setPredTarg(next_PC, next_PC + sizeof(TheISA::MachInst));
511#endif
512 inst->setPredTaken(false);
513 return false;
514 }
515
516 int tid = inst->threadNumber;
517#if ISA_HAS_DELAY_SLOT
518 Addr pred_PC = next_PC;
519 predict_taken = branchPred.predict(inst, pred_PC, tid);
520
521 if (predict_taken) {
522 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be taken.\n", tid);
523 } else {
524 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be not taken.\n", tid);
525 }
526
527 next_PC = next_NPC;
528 if (predict_taken) {
529 next_NPC = pred_PC;
530 // Update delay slot info
531 ++delaySlotInfo[tid].numInsts;
532 delaySlotInfo[tid].targetAddr = pred_PC;
533 DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) to process.\n", tid,
534 delaySlotInfo[tid].numInsts);
535 } else {
536 next_NPC = next_NPC + instSize;
537 }
538#else
539 predict_taken = branchPred.predict(inst, next_PC, tid);
540#endif
541 DPRINTF(Fetch, "[tid:%i]: Branch predicted to go to %#x and then %#x.\n",
542 tid, next_PC, next_NPC);
543 inst->setPredTarg(next_PC, next_NPC);
544 inst->setPredTaken(predict_taken);
545
546 ++fetchedBranches;
547
548 if (predict_taken) {
549 ++predictedBranches;
550 }
551
552 return predict_taken;

--- 104 unchanged lines hidden (view full) ---

657 }
658
659 ret_fault = fault;
660 return true;
661}
662
663template <class Impl>
664inline void
665DefaultFetch::doSquash(const Addr &new_PC,
666 const Addr &new_NPC, unsigned tid)
667{
668 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x, NPC to: %#x.\n",
669 tid, new_PC, new_NPC);
670
671 PC[tid] = new_PC;
672 nextPC[tid] = new_NPC;
673 nextNPC[tid] = new_NPC + instSize;
674
675 // Clear the icache miss if it's outstanding.
676 if (fetchStatus[tid] == IcacheWaitResponse) {
677 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n",
678 tid);
679 memReq[tid] = NULL;
680 }
681

--- 9 unchanged lines hidden (view full) ---

691
692 fetchStatus[tid] = Squashing;
693
694 ++fetchSquashCycles;
695}
696
697template<class Impl>
698void
699DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, const Addr &new_NPC,
700 const InstSeqNum &seq_num,
701 unsigned tid)
702{
703 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid);
704
705 doSquash(new_PC, new_NPC, tid);
706
707#if ISA_HAS_DELAY_SLOT
708 if (seq_num <= delaySlotInfo[tid].branchSeqNum) {
709 delaySlotInfo[tid].numInsts = 0;
710 delaySlotInfo[tid].targetAddr = 0;
711 delaySlotInfo[tid].targetReady = false;
712 }
713#endif

--- 29 unchanged lines hidden (view full) ---

743 return ret_val;
744}
745
746template<class Impl>
747typename DefaultFetch<Impl>::FetchStatus
748DefaultFetch<Impl>::updateFetchStatus()
749{
750 //Check Running
751 std::list<unsigned>::iterator threads = (*activeThreads).begin();
752
753 while (threads != (*activeThreads).end()) {
754
755 unsigned tid = *threads++;
756
757 if (fetchStatus[tid] == Running ||
758 fetchStatus[tid] == Squashing ||
759 fetchStatus[tid] == IcacheAccessComplete) {
760
761 if (_status == Inactive) {
762 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid);

--- 17 unchanged lines hidden (view full) ---

780 cpu->deactivateStage(O3CPU::FetchIdx);
781 }
782
783 return Inactive;
784}
785
786template <class Impl>
787void
788DefaultFetch<Impl>::squash(const Addr &new_PC, const Addr &new_NPC,
789 const InstSeqNum &seq_num,
790 bool squash_delay_slot, unsigned tid)
791{
792 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid);
793
794 doSquash(new_PC, new_NPC, tid);
795
796#if ISA_HAS_DELAY_SLOT
797 if (seq_num <= delaySlotInfo[tid].branchSeqNum) {
798 delaySlotInfo[tid].numInsts = 0;
799 delaySlotInfo[tid].targetAddr = 0;
800 delaySlotInfo[tid].targetReady = false;
801 }
802

--- 4 unchanged lines hidden (view full) ---

807 cpu->removeInstsNotInROB(tid, true, 0);
808#endif
809}
810
811template <class Impl>
812void
813DefaultFetch<Impl>::tick()
814{
815 std::list<unsigned>::iterator threads = (*activeThreads).begin();
816 bool status_change = false;
817
818 wroteToTimeBuffer = false;
819
820 while (threads != (*activeThreads).end()) {
821 unsigned tid = *threads++;
822
823 // Check the signals for each thread to determine the proper status
824 // for each thread.
825 bool updated_status = checkSignalsAndUpdate(tid);
826 status_change = status_change || updated_status;
827 }
828

--- 87 unchanged lines hidden (view full) ---

916
917#if ISA_HAS_DELAY_SLOT
918 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].bdelayDoneSeqNum;
919#else
920 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].doneSeqNum;
921#endif
922 // In any case, squash.
923 squash(fromCommit->commitInfo[tid].nextPC,
924 fromCommit->commitInfo[tid].nextNPC,
925 doneSeqNum,
926 fromCommit->commitInfo[tid].squashDelaySlot,
927 tid);
928
929 // Also check if there's a mispredict that happened.
930 if (fromCommit->commitInfo[tid].branchMispredict) {
931 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
932 fromCommit->commitInfo[tid].nextPC,

--- 41 unchanged lines hidden (view full) ---

974
975#if ISA_HAS_DELAY_SLOT
976 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].bdelayDoneSeqNum;
977#else
978 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].doneSeqNum;
979#endif
980 // Squash unless we're already squashing
981 squashFromDecode(fromDecode->decodeInfo[tid].nextPC,
982 fromDecode->decodeInfo[tid].nextNPC,
983 doneSeqNum,
984 tid);
985
986 return true;
987 }
988 }
989
990 if (checkStall(tid) &&

--- 40 unchanged lines hidden (view full) ---

1031 return;
1032 }
1033
1034 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid);
1035
1036 // The current PC.
1037 Addr &fetch_PC = PC[tid];
1038
1039 Addr &fetch_NPC = nextPC[tid];
1040
1041 // Fault code for memory access.
1042 Fault fault = NoFault;
1043
1044 // If returning from the delay of a cache miss, then update the status
1045 // to running, otherwise do the cache access. Possibly move this up
1046 // to tick() function.
1047 if (fetchStatus[tid] == IcacheAccessComplete) {
1048 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n",

--- 40 unchanged lines hidden (view full) ---

1089 // If we had a stall due to an icache miss, then return.
1090 if (fetchStatus[tid] == IcacheWaitResponse) {
1091 ++icacheStallCycles;
1092 status_change = true;
1093 return;
1094 }
1095
1096 Addr next_PC = fetch_PC;
1097 Addr next_NPC = fetch_NPC;
1098
1099 InstSeqNum inst_seq;
1100 MachInst inst;
1101 ExtMachInst ext_inst;
1102 // @todo: Fix this hack.
1103 unsigned offset = (fetch_PC & cacheBlkMask) & ~3;
1104
1105 if (fault == NoFault) {
1106 // If the read of the first instruction was successful, then grab the
1107 // instructions from the rest of the cache line and put them into the
1108 // queue heading to decode.
1109
1110 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to "
1111 "decode.\n",tid);
1112
1113 // Need to keep track of whether or not a predicted branch
1114 // ended this fetch block.
1115 bool predicted_branch = false;
1116
1117 for (;
1118 offset < cacheBlkSize &&
1119 numInst < fetchWidth &&
1120 !predicted_branch;
1121 ++numInst) {
1122
1123 // If we're branching after this instruction, quite fetching
1124 // from the same block then.
1125 predicted_branch =
1126 (fetch_PC + sizeof(TheISA::MachInst) != fetch_NPC);
1127
1128 // Get a sequence number.
1129 inst_seq = cpu->getAndIncrementInstSeq();
1130
1131 // Make sure this is a valid index.
1132 assert(offset <= cacheBlkSize - instSize);
1133
1134 // Get the instruction from the array of the cache line.
1135 inst = TheISA::gtoh(*reinterpret_cast<TheISA::MachInst *>
1136 (&cacheData[tid][offset]));
1137
1138#if THE_ISA == ALPHA_ISA
1139 ext_inst = TheISA::makeExtMI(inst, fetch_PC);
1140#elif THE_ISA == SPARC_ISA
1141 ext_inst = TheISA::makeExtMI(inst, cpu->thread[tid]->getTC());
1142#elif THE_ISA == MIPS_ISA
1143 ext_inst = TheISA::makeExtMI(inst, cpu->thread[tid]->getTC());
1144#endif
1145
1146 // Create a new DynInst from the instruction fetched.
1147 DynInstPtr instruction = new DynInst(ext_inst,
1148 fetch_PC, fetch_NPC,
1149 next_PC, next_NPC,
1150 inst_seq, cpu);
1151 instruction->setTid(tid);
1152
1153 instruction->setASID(tid);
1154
1155 instruction->setThreadState(cpu->thread[tid]);
1156
1157 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created "
1158 "[sn:%lli]\n",
1159 tid, instruction->readPC(), inst_seq);
1160
1161 DPRINTF(Fetch, "[tid:%i]: MachInst is %#x\n", tid, ext_inst);
1162
1163 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n",
1164 tid, instruction->staticInst->disassemble(fetch_PC));
1165
1166 instruction->traceData =
1167 Trace::getInstRecord(curTick, cpu->tcBase(tid),
1168 instruction->staticInst,
1169 instruction->readPC());
1170
1171 lookupAndUpdateNextPC(instruction, next_PC, next_NPC);
1172
1173 // Add instruction to the CPU's list of instructions.
1174 instruction->setInstListIt(cpu->addInst(instruction));
1175
1176 // Write the instruction to the first slot in the queue
1177 // that heads to decode.
1178 toDecode->insts[numInst] = instruction;
1179
1180 toDecode->size++;
1181
1182 // Increment stat of fetched instructions.
1183 ++fetchedInsts;
1184
1185 // Move to the next instruction, unless we have a branch.
1186 fetch_PC = next_PC;
1187 fetch_NPC = next_NPC;
1188
1189 if (instruction->isQuiesce()) {
1190 DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!",
1191 curTick);
1192 fetchStatus[tid] = QuiescePending;
1193 ++numInst;
1194 status_change = true;
1195 break;
1196 }
1197
1198 offset += instSize;
1199 }
1200
1201 if (offset >= cacheBlkSize) {
1202 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache "
1203 "block.\n", tid);
1204 } else if (numInst >= fetchWidth) {
1205 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth "
1206 "for this cycle.\n", tid);
1207 } else if (predicted_branch) {
1208 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch "
1209 "instruction encountered.\n", tid);
1210 }
1211 }
1212
1213 if (numInst > 0) {
1214 wroteToTimeBuffer = true;
1215 }
1216
1217 // Now that fetching is completed, update the PC to signify what the next
1218 // cycle will be.
1219 if (fault == NoFault) {
1220#if ISA_HAS_DELAY_SLOT
1221 if (delaySlotInfo[tid].targetReady &&
1222 delaySlotInfo[tid].numInsts == 0) {
1223 // Set PC to target
1224 PC[tid] = next_PC;
1225 nextPC[tid] = next_NPC;
1226 nextNPC[tid] = next_NPC + instSize;
1227
1228 delaySlotInfo[tid].targetReady = false;
1229 } else {
1230 PC[tid] = next_PC;
1231 nextPC[tid] = next_NPC;
1232 nextNPC[tid] = next_NPC + instSize;
1233 }
1234

--- 103 unchanged lines hidden (view full) ---

1338
1339 case Branch:
1340 return branchCount();
1341
1342 default:
1343 return -1;
1344 }
1345 } else {
1346 int tid = *((*activeThreads).begin());
1347
1348 if (fetchStatus[tid] == Running ||
1349 fetchStatus[tid] == IcacheAccessComplete ||
1350 fetchStatus[tid] == Idle) {
1351 return tid;
1352 } else {
1353 return -1;
1354 }

--- 33 unchanged lines hidden (view full) ---

1388}
1389
1390template<class Impl>
1391int
1392DefaultFetch<Impl>::iqCount()
1393{
1394 std::priority_queue<unsigned> PQ;
1395
1396 std::list<unsigned>::iterator threads = (*activeThreads).begin();
1397
1398 while (threads != (*activeThreads).end()) {
1399 unsigned tid = *threads++;
1400
1401 PQ.push(fromIEW->iewInfo[tid].iqCount);
1402 }
1403
1404 while (!PQ.empty()) {
1405
1406 unsigned high_pri = PQ.top();

--- 11 unchanged lines hidden (view full) ---

1418}
1419
1420template<class Impl>
1421int
1422DefaultFetch<Impl>::lsqCount()
1423{
1424 std::priority_queue<unsigned> PQ;
1425
1426
1427 std::list<unsigned>::iterator threads = (*activeThreads).begin();
1428
1429 while (threads != (*activeThreads).end()) {
1430 unsigned tid = *threads++;
1431
1432 PQ.push(fromIEW->iewInfo[tid].ldstqCount);
1433 }
1434
1435 while (!PQ.empty()) {
1436
1437 unsigned high_pri = PQ.top();

--- 9 unchanged lines hidden (view full) ---

1447
1448 return -1;
1449}
1450
1451template<class Impl>
1452int
1453DefaultFetch<Impl>::branchCount()
1454{
1455 std::list<unsigned>::iterator threads = (*activeThreads).begin();
1456 panic("Branch Count Fetch policy unimplemented\n");
1457 return *threads;
1458}