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 * Korey Sewell 30 */ 31
| 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 * Korey Sewell 30 */ 31
|
32#include <algorithm> 33#include <cstring> 34
| |
35#include "config/use_checker.hh" 36 37#include "arch/isa_traits.hh" 38#include "arch/utility.hh" 39#include "cpu/checker/cpu.hh" 40#include "cpu/exetrace.hh" 41#include "cpu/o3/fetch.hh" 42#include "mem/packet.hh" 43#include "mem/request.hh" 44#include "sim/byteswap.hh" 45#include "sim/host.hh" 46#include "sim/core.hh" 47 48#if FULL_SYSTEM 49#include "arch/tlb.hh" 50#include "arch/vtophys.hh" 51#include "sim/system.hh" 52#endif // FULL_SYSTEM 53
| 32#include "config/use_checker.hh" 33 34#include "arch/isa_traits.hh" 35#include "arch/utility.hh" 36#include "cpu/checker/cpu.hh" 37#include "cpu/exetrace.hh" 38#include "cpu/o3/fetch.hh" 39#include "mem/packet.hh" 40#include "mem/request.hh" 41#include "sim/byteswap.hh" 42#include "sim/host.hh" 43#include "sim/core.hh" 44 45#if FULL_SYSTEM 46#include "arch/tlb.hh" 47#include "arch/vtophys.hh" 48#include "sim/system.hh" 49#endif // FULL_SYSTEM 50
|
| 51#include <algorithm> 52
|
54template<class Impl> 55void 56DefaultFetch<Impl>::IcachePort::setPeer(Port *port) 57{ 58 Port::setPeer(port); 59 60 fetch->setIcache(); 61} 62 63template<class Impl> 64Tick 65DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 66{ 67 panic("DefaultFetch doesn't expect recvAtomic callback!"); 68 return curTick; 69} 70 71template<class Impl> 72void 73DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 74{ 75 DPRINTF(Fetch, "DefaultFetch doesn't update its state from a " 76 "functional call."); 77} 78 79template<class Impl> 80void 81DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 82{ 83 if (status == RangeChange) { 84 if (!snoopRangeSent) { 85 snoopRangeSent = true; 86 sendStatusChange(Port::RangeChange); 87 } 88 return; 89 } 90 91 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 92} 93 94template<class Impl> 95bool 96DefaultFetch<Impl>::IcachePort::recvTiming(PacketPtr pkt) 97{ 98 DPRINTF(Fetch, "Received timing\n"); 99 if (pkt->isResponse()) { 100 fetch->processCacheCompletion(pkt); 101 } 102 //else Snooped a coherence request, just return 103 return true; 104} 105 106template<class Impl> 107void 108DefaultFetch<Impl>::IcachePort::recvRetry() 109{ 110 fetch->recvRetry(); 111} 112 113template<class Impl> 114DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, Params *params) 115 : cpu(_cpu), 116 branchPred(params), 117 predecoder(NULL), 118 decodeToFetchDelay(params->decodeToFetchDelay), 119 renameToFetchDelay(params->renameToFetchDelay), 120 iewToFetchDelay(params->iewToFetchDelay), 121 commitToFetchDelay(params->commitToFetchDelay), 122 fetchWidth(params->fetchWidth), 123 cacheBlocked(false), 124 retryPkt(NULL), 125 retryTid(-1), 126 numThreads(params->numberOfThreads), 127 numFetchingThreads(params->smtNumFetchingThreads), 128 interruptPending(false), 129 drainPending(false), 130 switchedOut(false) 131{ 132 if (numThreads > Impl::MaxThreads) 133 fatal("numThreads is not a valid value\n"); 134 135 // Set fetch stage's status to inactive. 136 _status = Inactive; 137 138 std::string policy = params->smtFetchPolicy; 139 140 // Convert string to lowercase 141 std::transform(policy.begin(), policy.end(), policy.begin(), 142 (int(*)(int)) tolower); 143 144 // Figure out fetch policy 145 if (policy == "singlethread") { 146 fetchPolicy = SingleThread; 147 if (numThreads > 1) 148 panic("Invalid Fetch Policy for a SMT workload."); 149 } else if (policy == "roundrobin") { 150 fetchPolicy = RoundRobin; 151 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 152 } else if (policy == "branch") { 153 fetchPolicy = Branch; 154 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 155 } else if (policy == "iqcount") { 156 fetchPolicy = IQ; 157 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 158 } else if (policy == "lsqcount") { 159 fetchPolicy = LSQ; 160 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 161 } else { 162 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 163 " RoundRobin,LSQcount,IQcount}\n"); 164 } 165 166 // Get the size of an instruction. 167 instSize = sizeof(TheISA::MachInst); 168 169 // Name is finally available, so create the port. 170 icachePort = new IcachePort(this); 171 172 icachePort->snoopRangeSent = false; 173 174#if USE_CHECKER 175 if (cpu->checker) { 176 cpu->checker->setIcachePort(icachePort); 177 } 178#endif 179} 180 181template <class Impl> 182std::string 183DefaultFetch<Impl>::name() const 184{ 185 return cpu->name() + ".fetch"; 186} 187 188template <class Impl> 189void 190DefaultFetch<Impl>::regStats() 191{ 192 icacheStallCycles 193 .name(name() + ".icacheStallCycles") 194 .desc("Number of cycles fetch is stalled on an Icache miss") 195 .prereq(icacheStallCycles); 196 197 fetchedInsts 198 .name(name() + ".Insts") 199 .desc("Number of instructions fetch has processed") 200 .prereq(fetchedInsts); 201 202 fetchedBranches 203 .name(name() + ".Branches") 204 .desc("Number of branches that fetch encountered") 205 .prereq(fetchedBranches); 206 207 predictedBranches 208 .name(name() + ".predictedBranches") 209 .desc("Number of branches that fetch has predicted taken") 210 .prereq(predictedBranches); 211 212 fetchCycles 213 .name(name() + ".Cycles") 214 .desc("Number of cycles fetch has run and was not squashing or" 215 " blocked") 216 .prereq(fetchCycles); 217 218 fetchSquashCycles 219 .name(name() + ".SquashCycles") 220 .desc("Number of cycles fetch has spent squashing") 221 .prereq(fetchSquashCycles); 222 223 fetchIdleCycles 224 .name(name() + ".IdleCycles") 225 .desc("Number of cycles fetch was idle") 226 .prereq(fetchIdleCycles); 227 228 fetchBlockedCycles 229 .name(name() + ".BlockedCycles") 230 .desc("Number of cycles fetch has spent blocked") 231 .prereq(fetchBlockedCycles); 232 233 fetchedCacheLines 234 .name(name() + ".CacheLines") 235 .desc("Number of cache lines fetched") 236 .prereq(fetchedCacheLines); 237 238 fetchMiscStallCycles 239 .name(name() + ".MiscStallCycles") 240 .desc("Number of cycles fetch has spent waiting on interrupts, or " 241 "bad addresses, or out of MSHRs") 242 .prereq(fetchMiscStallCycles); 243 244 fetchIcacheSquashes 245 .name(name() + ".IcacheSquashes") 246 .desc("Number of outstanding Icache misses that were squashed") 247 .prereq(fetchIcacheSquashes); 248 249 fetchNisnDist 250 .init(/* base value */ 0, 251 /* last value */ fetchWidth, 252 /* bucket size */ 1) 253 .name(name() + ".rateDist") 254 .desc("Number of instructions fetched each cycle (Total)") 255 .flags(Stats::pdf); 256 257 idleRate 258 .name(name() + ".idleRate") 259 .desc("Percent of cycles fetch was idle") 260 .prereq(idleRate); 261 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 262 263 branchRate 264 .name(name() + ".branchRate") 265 .desc("Number of branch fetches per cycle") 266 .flags(Stats::total); 267 branchRate = fetchedBranches / cpu->numCycles; 268 269 fetchRate 270 .name(name() + ".rate") 271 .desc("Number of inst fetches per cycle") 272 .flags(Stats::total); 273 fetchRate = fetchedInsts / cpu->numCycles; 274 275 branchPred.regStats(); 276} 277 278template<class Impl> 279void 280DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 281{ 282 timeBuffer = time_buffer; 283 284 // Create wires to get information from proper places in time buffer. 285 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 286 fromRename = timeBuffer->getWire(-renameToFetchDelay); 287 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 288 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 289} 290 291template<class Impl> 292void 293DefaultFetch<Impl>::setActiveThreads(std::list<unsigned> *at_ptr) 294{ 295 activeThreads = at_ptr; 296} 297 298template<class Impl> 299void 300DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 301{ 302 fetchQueue = fq_ptr; 303 304 // Create wire to write information to proper place in fetch queue. 305 toDecode = fetchQueue->getWire(0); 306} 307 308template<class Impl> 309void 310DefaultFetch<Impl>::initStage() 311{ 312 // Setup PC and nextPC with initial state. 313 for (int tid = 0; tid < numThreads; tid++) { 314 PC[tid] = cpu->readPC(tid); 315 nextPC[tid] = cpu->readNextPC(tid); 316 nextNPC[tid] = cpu->readNextNPC(tid); 317 } 318 319 for (int tid=0; tid < numThreads; tid++) { 320 321 fetchStatus[tid] = Running; 322 323 priorityList.push_back(tid); 324 325 memReq[tid] = NULL; 326 327 stalls[tid].decode = false; 328 stalls[tid].rename = false; 329 stalls[tid].iew = false; 330 stalls[tid].commit = false; 331 } 332 333 // Schedule fetch to get the correct PC from the CPU 334 // scheduleFetchStartupEvent(1); 335 336 // Fetch needs to start fetching instructions at the very beginning, 337 // so it must start up in active state. 338 switchToActive(); 339} 340 341template<class Impl> 342void 343DefaultFetch<Impl>::setIcache() 344{ 345 // Size of cache block. 346 cacheBlkSize = icachePort->peerBlockSize(); 347 348 // Create mask to get rid of offset bits. 349 cacheBlkMask = (cacheBlkSize - 1); 350 351 for (int tid=0; tid < numThreads; tid++) { 352 // Create space to store a cache line. 353 cacheData[tid] = new uint8_t[cacheBlkSize]; 354 cacheDataPC[tid] = 0; 355 cacheDataValid[tid] = false; 356 } 357} 358 359template<class Impl> 360void 361DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 362{ 363 unsigned tid = pkt->req->getThreadNum(); 364 365 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 366 367 // Only change the status if it's still waiting on the icache access 368 // to return. 369 if (fetchStatus[tid] != IcacheWaitResponse || 370 pkt->req != memReq[tid] || 371 isSwitchedOut()) { 372 ++fetchIcacheSquashes; 373 delete pkt->req; 374 delete pkt; 375 return; 376 } 377
| 53template<class Impl> 54void 55DefaultFetch<Impl>::IcachePort::setPeer(Port *port) 56{ 57 Port::setPeer(port); 58 59 fetch->setIcache(); 60} 61 62template<class Impl> 63Tick 64DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 65{ 66 panic("DefaultFetch doesn't expect recvAtomic callback!"); 67 return curTick; 68} 69 70template<class Impl> 71void 72DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 73{ 74 DPRINTF(Fetch, "DefaultFetch doesn't update its state from a " 75 "functional call."); 76} 77 78template<class Impl> 79void 80DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 81{ 82 if (status == RangeChange) { 83 if (!snoopRangeSent) { 84 snoopRangeSent = true; 85 sendStatusChange(Port::RangeChange); 86 } 87 return; 88 } 89 90 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 91} 92 93template<class Impl> 94bool 95DefaultFetch<Impl>::IcachePort::recvTiming(PacketPtr pkt) 96{ 97 DPRINTF(Fetch, "Received timing\n"); 98 if (pkt->isResponse()) { 99 fetch->processCacheCompletion(pkt); 100 } 101 //else Snooped a coherence request, just return 102 return true; 103} 104 105template<class Impl> 106void 107DefaultFetch<Impl>::IcachePort::recvRetry() 108{ 109 fetch->recvRetry(); 110} 111 112template<class Impl> 113DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, Params *params) 114 : cpu(_cpu), 115 branchPred(params), 116 predecoder(NULL), 117 decodeToFetchDelay(params->decodeToFetchDelay), 118 renameToFetchDelay(params->renameToFetchDelay), 119 iewToFetchDelay(params->iewToFetchDelay), 120 commitToFetchDelay(params->commitToFetchDelay), 121 fetchWidth(params->fetchWidth), 122 cacheBlocked(false), 123 retryPkt(NULL), 124 retryTid(-1), 125 numThreads(params->numberOfThreads), 126 numFetchingThreads(params->smtNumFetchingThreads), 127 interruptPending(false), 128 drainPending(false), 129 switchedOut(false) 130{ 131 if (numThreads > Impl::MaxThreads) 132 fatal("numThreads is not a valid value\n"); 133 134 // Set fetch stage's status to inactive. 135 _status = Inactive; 136 137 std::string policy = params->smtFetchPolicy; 138 139 // Convert string to lowercase 140 std::transform(policy.begin(), policy.end(), policy.begin(), 141 (int(*)(int)) tolower); 142 143 // Figure out fetch policy 144 if (policy == "singlethread") { 145 fetchPolicy = SingleThread; 146 if (numThreads > 1) 147 panic("Invalid Fetch Policy for a SMT workload."); 148 } else if (policy == "roundrobin") { 149 fetchPolicy = RoundRobin; 150 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 151 } else if (policy == "branch") { 152 fetchPolicy = Branch; 153 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 154 } else if (policy == "iqcount") { 155 fetchPolicy = IQ; 156 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 157 } else if (policy == "lsqcount") { 158 fetchPolicy = LSQ; 159 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 160 } else { 161 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 162 " RoundRobin,LSQcount,IQcount}\n"); 163 } 164 165 // Get the size of an instruction. 166 instSize = sizeof(TheISA::MachInst); 167 168 // Name is finally available, so create the port. 169 icachePort = new IcachePort(this); 170 171 icachePort->snoopRangeSent = false; 172 173#if USE_CHECKER 174 if (cpu->checker) { 175 cpu->checker->setIcachePort(icachePort); 176 } 177#endif 178} 179 180template <class Impl> 181std::string 182DefaultFetch<Impl>::name() const 183{ 184 return cpu->name() + ".fetch"; 185} 186 187template <class Impl> 188void 189DefaultFetch<Impl>::regStats() 190{ 191 icacheStallCycles 192 .name(name() + ".icacheStallCycles") 193 .desc("Number of cycles fetch is stalled on an Icache miss") 194 .prereq(icacheStallCycles); 195 196 fetchedInsts 197 .name(name() + ".Insts") 198 .desc("Number of instructions fetch has processed") 199 .prereq(fetchedInsts); 200 201 fetchedBranches 202 .name(name() + ".Branches") 203 .desc("Number of branches that fetch encountered") 204 .prereq(fetchedBranches); 205 206 predictedBranches 207 .name(name() + ".predictedBranches") 208 .desc("Number of branches that fetch has predicted taken") 209 .prereq(predictedBranches); 210 211 fetchCycles 212 .name(name() + ".Cycles") 213 .desc("Number of cycles fetch has run and was not squashing or" 214 " blocked") 215 .prereq(fetchCycles); 216 217 fetchSquashCycles 218 .name(name() + ".SquashCycles") 219 .desc("Number of cycles fetch has spent squashing") 220 .prereq(fetchSquashCycles); 221 222 fetchIdleCycles 223 .name(name() + ".IdleCycles") 224 .desc("Number of cycles fetch was idle") 225 .prereq(fetchIdleCycles); 226 227 fetchBlockedCycles 228 .name(name() + ".BlockedCycles") 229 .desc("Number of cycles fetch has spent blocked") 230 .prereq(fetchBlockedCycles); 231 232 fetchedCacheLines 233 .name(name() + ".CacheLines") 234 .desc("Number of cache lines fetched") 235 .prereq(fetchedCacheLines); 236 237 fetchMiscStallCycles 238 .name(name() + ".MiscStallCycles") 239 .desc("Number of cycles fetch has spent waiting on interrupts, or " 240 "bad addresses, or out of MSHRs") 241 .prereq(fetchMiscStallCycles); 242 243 fetchIcacheSquashes 244 .name(name() + ".IcacheSquashes") 245 .desc("Number of outstanding Icache misses that were squashed") 246 .prereq(fetchIcacheSquashes); 247 248 fetchNisnDist 249 .init(/* base value */ 0, 250 /* last value */ fetchWidth, 251 /* bucket size */ 1) 252 .name(name() + ".rateDist") 253 .desc("Number of instructions fetched each cycle (Total)") 254 .flags(Stats::pdf); 255 256 idleRate 257 .name(name() + ".idleRate") 258 .desc("Percent of cycles fetch was idle") 259 .prereq(idleRate); 260 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 261 262 branchRate 263 .name(name() + ".branchRate") 264 .desc("Number of branch fetches per cycle") 265 .flags(Stats::total); 266 branchRate = fetchedBranches / cpu->numCycles; 267 268 fetchRate 269 .name(name() + ".rate") 270 .desc("Number of inst fetches per cycle") 271 .flags(Stats::total); 272 fetchRate = fetchedInsts / cpu->numCycles; 273 274 branchPred.regStats(); 275} 276 277template<class Impl> 278void 279DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 280{ 281 timeBuffer = time_buffer; 282 283 // Create wires to get information from proper places in time buffer. 284 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 285 fromRename = timeBuffer->getWire(-renameToFetchDelay); 286 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 287 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 288} 289 290template<class Impl> 291void 292DefaultFetch<Impl>::setActiveThreads(std::list<unsigned> *at_ptr) 293{ 294 activeThreads = at_ptr; 295} 296 297template<class Impl> 298void 299DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 300{ 301 fetchQueue = fq_ptr; 302 303 // Create wire to write information to proper place in fetch queue. 304 toDecode = fetchQueue->getWire(0); 305} 306 307template<class Impl> 308void 309DefaultFetch<Impl>::initStage() 310{ 311 // Setup PC and nextPC with initial state. 312 for (int tid = 0; tid < numThreads; tid++) { 313 PC[tid] = cpu->readPC(tid); 314 nextPC[tid] = cpu->readNextPC(tid); 315 nextNPC[tid] = cpu->readNextNPC(tid); 316 } 317 318 for (int tid=0; tid < numThreads; tid++) { 319 320 fetchStatus[tid] = Running; 321 322 priorityList.push_back(tid); 323 324 memReq[tid] = NULL; 325 326 stalls[tid].decode = false; 327 stalls[tid].rename = false; 328 stalls[tid].iew = false; 329 stalls[tid].commit = false; 330 } 331 332 // Schedule fetch to get the correct PC from the CPU 333 // scheduleFetchStartupEvent(1); 334 335 // Fetch needs to start fetching instructions at the very beginning, 336 // so it must start up in active state. 337 switchToActive(); 338} 339 340template<class Impl> 341void 342DefaultFetch<Impl>::setIcache() 343{ 344 // Size of cache block. 345 cacheBlkSize = icachePort->peerBlockSize(); 346 347 // Create mask to get rid of offset bits. 348 cacheBlkMask = (cacheBlkSize - 1); 349 350 for (int tid=0; tid < numThreads; tid++) { 351 // Create space to store a cache line. 352 cacheData[tid] = new uint8_t[cacheBlkSize]; 353 cacheDataPC[tid] = 0; 354 cacheDataValid[tid] = false; 355 } 356} 357 358template<class Impl> 359void 360DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 361{ 362 unsigned tid = pkt->req->getThreadNum(); 363 364 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 365 366 // Only change the status if it's still waiting on the icache access 367 // to return. 368 if (fetchStatus[tid] != IcacheWaitResponse || 369 pkt->req != memReq[tid] || 370 isSwitchedOut()) { 371 ++fetchIcacheSquashes; 372 delete pkt->req; 373 delete pkt; 374 return; 375 } 376
|
378 memcpy(cacheData[tid], pkt->getPtr(), cacheBlkSize);
| 377 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize);
|
379 cacheDataValid[tid] = true; 380 381 if (!drainPending) { 382 // Wake up the CPU (if it went to sleep and was waiting on 383 // this completion event). 384 cpu->wakeCPU(); 385 386 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 387 tid); 388 389 switchToActive(); 390 } 391 392 // Only switch to IcacheAccessComplete if we're not stalled as well. 393 if (checkStall(tid)) { 394 fetchStatus[tid] = Blocked; 395 } else { 396 fetchStatus[tid] = IcacheAccessComplete; 397 } 398 399 // Reset the mem req to NULL. 400 delete pkt->req; 401 delete pkt; 402 memReq[tid] = NULL; 403} 404 405template <class Impl> 406bool 407DefaultFetch<Impl>::drain() 408{ 409 // Fetch is ready to drain at any time. 410 cpu->signalDrained(); 411 drainPending = true; 412 return true; 413} 414 415template <class Impl> 416void 417DefaultFetch<Impl>::resume() 418{ 419 drainPending = false; 420} 421 422template <class Impl> 423void 424DefaultFetch<Impl>::switchOut() 425{ 426 switchedOut = true; 427 // Branch predictor needs to have its state cleared. 428 branchPred.switchOut(); 429} 430 431template <class Impl> 432void 433DefaultFetch<Impl>::takeOverFrom() 434{ 435 // Reset all state 436 for (int i = 0; i < Impl::MaxThreads; ++i) { 437 stalls[i].decode = 0; 438 stalls[i].rename = 0; 439 stalls[i].iew = 0; 440 stalls[i].commit = 0; 441 PC[i] = cpu->readPC(i); 442 nextPC[i] = cpu->readNextPC(i); 443#if ISA_HAS_DELAY_SLOT 444 nextNPC[i] = cpu->readNextNPC(i); 445#else 446 nextNPC[i] = nextPC[i] + sizeof(TheISA::MachInst); 447#endif 448 fetchStatus[i] = Running; 449 } 450 numInst = 0; 451 wroteToTimeBuffer = false; 452 _status = Inactive; 453 switchedOut = false; 454 interruptPending = false; 455 branchPred.takeOverFrom(); 456} 457 458template <class Impl> 459void 460DefaultFetch<Impl>::wakeFromQuiesce() 461{ 462 DPRINTF(Fetch, "Waking up from quiesce\n"); 463 // Hopefully this is safe 464 // @todo: Allow other threads to wake from quiesce. 465 fetchStatus[0] = Running; 466} 467 468template <class Impl> 469inline void 470DefaultFetch<Impl>::switchToActive() 471{ 472 if (_status == Inactive) { 473 DPRINTF(Activity, "Activating stage.\n"); 474 475 cpu->activateStage(O3CPU::FetchIdx); 476 477 _status = Active; 478 } 479} 480 481template <class Impl> 482inline void 483DefaultFetch<Impl>::switchToInactive() 484{ 485 if (_status == Active) { 486 DPRINTF(Activity, "Deactivating stage.\n"); 487 488 cpu->deactivateStage(O3CPU::FetchIdx); 489 490 _status = Inactive; 491 } 492} 493 494template <class Impl> 495bool 496DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC, 497 Addr &next_NPC) 498{ 499 // Do branch prediction check here. 500 // A bit of a misnomer...next_PC is actually the current PC until 501 // this function updates it. 502 bool predict_taken; 503 504 if (!inst->isControl()) { 505 next_PC = next_NPC; 506 next_NPC = next_NPC + instSize; 507 inst->setPredTarg(next_PC, next_NPC); 508 inst->setPredTaken(false); 509 return false; 510 } 511 512 int tid = inst->threadNumber; 513 Addr pred_PC = next_PC; 514 predict_taken = branchPred.predict(inst, pred_PC, tid); 515 516/* if (predict_taken) { 517 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be taken to %#x.\n", 518 tid, pred_PC); 519 } else { 520 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be not taken.\n", tid); 521 }*/ 522 523#if ISA_HAS_DELAY_SLOT 524 next_PC = next_NPC; 525 if (predict_taken) 526 next_NPC = pred_PC; 527 else 528 next_NPC += instSize; 529#else 530 if (predict_taken) 531 next_PC = pred_PC; 532 else 533 next_PC += instSize; 534 next_NPC = next_PC + instSize; 535#endif 536/* DPRINTF(Fetch, "[tid:%i]: Branch predicted to go to %#x and then %#x.\n", 537 tid, next_PC, next_NPC);*/ 538 inst->setPredTarg(next_PC, next_NPC); 539 inst->setPredTaken(predict_taken); 540 541 ++fetchedBranches; 542 543 if (predict_taken) { 544 ++predictedBranches; 545 } 546 547 return predict_taken; 548} 549 550template <class Impl> 551bool 552DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 553{ 554 Fault fault = NoFault; 555 556 //AlphaDep 557 if (cacheBlocked) { 558 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 559 tid); 560 return false; 561 } else if (isSwitchedOut()) { 562 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 563 tid); 564 return false; 565 } else if (interruptPending && !(fetch_PC & 0x3)) { 566 // Hold off fetch from getting new instructions when: 567 // Cache is blocked, or 568 // while an interrupt is pending and we're not in PAL mode, or 569 // fetch is switched out. 570 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 571 tid); 572 return false; 573 } 574 575 // Align the fetch PC so it's at the start of a cache block. 576 Addr block_PC = icacheBlockAlignPC(fetch_PC); 577 578 // If we've already got the block, no need to try to fetch it again. 579 if (cacheDataValid[tid] && block_PC == cacheDataPC[tid]) { 580 return true; 581 } 582 583 // Setup the memReq to do a read of the first instruction's address. 584 // Set the appropriate read size and flags as well. 585 // Build request here. 586 RequestPtr mem_req = new Request(tid, block_PC, cacheBlkSize, 0, 587 fetch_PC, cpu->readCpuId(), tid); 588 589 memReq[tid] = mem_req; 590 591 // Translate the instruction request. 592 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 593 594 // In the case of faults, the fetch stage may need to stall and wait 595 // for the ITB miss to be handled. 596 597 // If translation was successful, attempt to read the first 598 // instruction. 599 if (fault == NoFault) { 600#if 0 601 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 602 memReq[tid]->isUncacheable()) { 603 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 604 "misspeculating path)!", 605 memReq[tid]->paddr); 606 ret_fault = TheISA::genMachineCheckFault(); 607 return false; 608 } 609#endif 610 611 // Build packet here. 612 PacketPtr data_pkt = new Packet(mem_req, 613 MemCmd::ReadReq, Packet::Broadcast); 614 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 615 616 cacheDataPC[tid] = block_PC; 617 cacheDataValid[tid] = false; 618 619 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 620 621 fetchedCacheLines++; 622 623 // Now do the timing access to see whether or not the instruction 624 // exists within the cache. 625 if (!icachePort->sendTiming(data_pkt)) { 626 if (data_pkt->result == Packet::BadAddress) { 627 fault = TheISA::genMachineCheckFault(); 628 delete mem_req; 629 memReq[tid] = NULL; 630 warn("Bad address!\n"); 631 } 632 assert(retryPkt == NULL); 633 assert(retryTid == -1); 634 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 635 fetchStatus[tid] = IcacheWaitRetry; 636 retryPkt = data_pkt; 637 retryTid = tid; 638 cacheBlocked = true; 639 return false; 640 } 641 642 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 643 644 lastIcacheStall[tid] = curTick; 645 646 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 647 "response.\n", tid); 648 649 fetchStatus[tid] = IcacheWaitResponse; 650 } else { 651 delete mem_req; 652 memReq[tid] = NULL; 653 } 654 655 ret_fault = fault; 656 return true; 657} 658 659template <class Impl> 660inline void 661DefaultFetch<Impl>::doSquash(const Addr &new_PC, 662 const Addr &new_NPC, unsigned tid) 663{ 664 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x, NPC to: %#x.\n", 665 tid, new_PC, new_NPC); 666 667 PC[tid] = new_PC; 668 nextPC[tid] = new_NPC; 669 nextNPC[tid] = new_NPC + instSize; 670 671 // Clear the icache miss if it's outstanding. 672 if (fetchStatus[tid] == IcacheWaitResponse) { 673 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 674 tid); 675 memReq[tid] = NULL; 676 } 677 678 // Get rid of the retrying packet if it was from this thread. 679 if (retryTid == tid) { 680 assert(cacheBlocked); 681 if (retryPkt) { 682 delete retryPkt->req; 683 delete retryPkt; 684 } 685 retryPkt = NULL; 686 retryTid = -1; 687 } 688 689 fetchStatus[tid] = Squashing; 690 691 ++fetchSquashCycles; 692} 693 694template<class Impl> 695void 696DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, const Addr &new_NPC, 697 const InstSeqNum &seq_num, 698 unsigned tid) 699{ 700 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 701 702 doSquash(new_PC, new_NPC, tid); 703 704 // Tell the CPU to remove any instructions that are in flight between 705 // fetch and decode. 706 cpu->removeInstsUntil(seq_num, tid); 707} 708 709template<class Impl> 710bool 711DefaultFetch<Impl>::checkStall(unsigned tid) const 712{ 713 bool ret_val = false; 714 715 if (cpu->contextSwitch) { 716 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 717 ret_val = true; 718 } else if (stalls[tid].decode) { 719 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 720 ret_val = true; 721 } else if (stalls[tid].rename) { 722 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 723 ret_val = true; 724 } else if (stalls[tid].iew) { 725 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 726 ret_val = true; 727 } else if (stalls[tid].commit) { 728 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 729 ret_val = true; 730 } 731 732 return ret_val; 733} 734 735template<class Impl> 736typename DefaultFetch<Impl>::FetchStatus 737DefaultFetch<Impl>::updateFetchStatus() 738{ 739 //Check Running 740 std::list<unsigned>::iterator threads = activeThreads->begin(); 741 std::list<unsigned>::iterator end = activeThreads->end(); 742 743 while (threads != end) { 744 unsigned tid = *threads++; 745 746 if (fetchStatus[tid] == Running || 747 fetchStatus[tid] == Squashing || 748 fetchStatus[tid] == IcacheAccessComplete) { 749 750 if (_status == Inactive) { 751 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 752 753 if (fetchStatus[tid] == IcacheAccessComplete) { 754 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 755 "completion\n",tid); 756 } 757 758 cpu->activateStage(O3CPU::FetchIdx); 759 } 760 761 return Active; 762 } 763 } 764 765 // Stage is switching from active to inactive, notify CPU of it. 766 if (_status == Active) { 767 DPRINTF(Activity, "Deactivating stage.\n"); 768 769 cpu->deactivateStage(O3CPU::FetchIdx); 770 } 771 772 return Inactive; 773} 774 775template <class Impl> 776void 777DefaultFetch<Impl>::squash(const Addr &new_PC, const Addr &new_NPC,
| 378 cacheDataValid[tid] = true; 379 380 if (!drainPending) { 381 // Wake up the CPU (if it went to sleep and was waiting on 382 // this completion event). 383 cpu->wakeCPU(); 384 385 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 386 tid); 387 388 switchToActive(); 389 } 390 391 // Only switch to IcacheAccessComplete if we're not stalled as well. 392 if (checkStall(tid)) { 393 fetchStatus[tid] = Blocked; 394 } else { 395 fetchStatus[tid] = IcacheAccessComplete; 396 } 397 398 // Reset the mem req to NULL. 399 delete pkt->req; 400 delete pkt; 401 memReq[tid] = NULL; 402} 403 404template <class Impl> 405bool 406DefaultFetch<Impl>::drain() 407{ 408 // Fetch is ready to drain at any time. 409 cpu->signalDrained(); 410 drainPending = true; 411 return true; 412} 413 414template <class Impl> 415void 416DefaultFetch<Impl>::resume() 417{ 418 drainPending = false; 419} 420 421template <class Impl> 422void 423DefaultFetch<Impl>::switchOut() 424{ 425 switchedOut = true; 426 // Branch predictor needs to have its state cleared. 427 branchPred.switchOut(); 428} 429 430template <class Impl> 431void 432DefaultFetch<Impl>::takeOverFrom() 433{ 434 // Reset all state 435 for (int i = 0; i < Impl::MaxThreads; ++i) { 436 stalls[i].decode = 0; 437 stalls[i].rename = 0; 438 stalls[i].iew = 0; 439 stalls[i].commit = 0; 440 PC[i] = cpu->readPC(i); 441 nextPC[i] = cpu->readNextPC(i); 442#if ISA_HAS_DELAY_SLOT 443 nextNPC[i] = cpu->readNextNPC(i); 444#else 445 nextNPC[i] = nextPC[i] + sizeof(TheISA::MachInst); 446#endif 447 fetchStatus[i] = Running; 448 } 449 numInst = 0; 450 wroteToTimeBuffer = false; 451 _status = Inactive; 452 switchedOut = false; 453 interruptPending = false; 454 branchPred.takeOverFrom(); 455} 456 457template <class Impl> 458void 459DefaultFetch<Impl>::wakeFromQuiesce() 460{ 461 DPRINTF(Fetch, "Waking up from quiesce\n"); 462 // Hopefully this is safe 463 // @todo: Allow other threads to wake from quiesce. 464 fetchStatus[0] = Running; 465} 466 467template <class Impl> 468inline void 469DefaultFetch<Impl>::switchToActive() 470{ 471 if (_status == Inactive) { 472 DPRINTF(Activity, "Activating stage.\n"); 473 474 cpu->activateStage(O3CPU::FetchIdx); 475 476 _status = Active; 477 } 478} 479 480template <class Impl> 481inline void 482DefaultFetch<Impl>::switchToInactive() 483{ 484 if (_status == Active) { 485 DPRINTF(Activity, "Deactivating stage.\n"); 486 487 cpu->deactivateStage(O3CPU::FetchIdx); 488 489 _status = Inactive; 490 } 491} 492 493template <class Impl> 494bool 495DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC, 496 Addr &next_NPC) 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 next_PC = next_NPC; 505 next_NPC = next_NPC + instSize; 506 inst->setPredTarg(next_PC, next_NPC); 507 inst->setPredTaken(false); 508 return false; 509 } 510 511 int tid = inst->threadNumber; 512 Addr pred_PC = next_PC; 513 predict_taken = branchPred.predict(inst, pred_PC, tid); 514 515/* if (predict_taken) { 516 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be taken to %#x.\n", 517 tid, pred_PC); 518 } else { 519 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be not taken.\n", tid); 520 }*/ 521 522#if ISA_HAS_DELAY_SLOT 523 next_PC = next_NPC; 524 if (predict_taken) 525 next_NPC = pred_PC; 526 else 527 next_NPC += instSize; 528#else 529 if (predict_taken) 530 next_PC = pred_PC; 531 else 532 next_PC += instSize; 533 next_NPC = next_PC + instSize; 534#endif 535/* DPRINTF(Fetch, "[tid:%i]: Branch predicted to go to %#x and then %#x.\n", 536 tid, next_PC, next_NPC);*/ 537 inst->setPredTarg(next_PC, next_NPC); 538 inst->setPredTaken(predict_taken); 539 540 ++fetchedBranches; 541 542 if (predict_taken) { 543 ++predictedBranches; 544 } 545 546 return predict_taken; 547} 548 549template <class Impl> 550bool 551DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 552{ 553 Fault fault = NoFault; 554 555 //AlphaDep 556 if (cacheBlocked) { 557 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 558 tid); 559 return false; 560 } else if (isSwitchedOut()) { 561 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 562 tid); 563 return false; 564 } else if (interruptPending && !(fetch_PC & 0x3)) { 565 // Hold off fetch from getting new instructions when: 566 // Cache is blocked, or 567 // while an interrupt is pending and we're not in PAL mode, or 568 // fetch is switched out. 569 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 570 tid); 571 return false; 572 } 573 574 // Align the fetch PC so it's at the start of a cache block. 575 Addr block_PC = icacheBlockAlignPC(fetch_PC); 576 577 // If we've already got the block, no need to try to fetch it again. 578 if (cacheDataValid[tid] && block_PC == cacheDataPC[tid]) { 579 return true; 580 } 581 582 // Setup the memReq to do a read of the first instruction's address. 583 // Set the appropriate read size and flags as well. 584 // Build request here. 585 RequestPtr mem_req = new Request(tid, block_PC, cacheBlkSize, 0, 586 fetch_PC, cpu->readCpuId(), tid); 587 588 memReq[tid] = mem_req; 589 590 // Translate the instruction request. 591 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 592 593 // In the case of faults, the fetch stage may need to stall and wait 594 // for the ITB miss to be handled. 595 596 // If translation was successful, attempt to read the first 597 // instruction. 598 if (fault == NoFault) { 599#if 0 600 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 601 memReq[tid]->isUncacheable()) { 602 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 603 "misspeculating path)!", 604 memReq[tid]->paddr); 605 ret_fault = TheISA::genMachineCheckFault(); 606 return false; 607 } 608#endif 609 610 // Build packet here. 611 PacketPtr data_pkt = new Packet(mem_req, 612 MemCmd::ReadReq, Packet::Broadcast); 613 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 614 615 cacheDataPC[tid] = block_PC; 616 cacheDataValid[tid] = false; 617 618 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 619 620 fetchedCacheLines++; 621 622 // Now do the timing access to see whether or not the instruction 623 // exists within the cache. 624 if (!icachePort->sendTiming(data_pkt)) { 625 if (data_pkt->result == Packet::BadAddress) { 626 fault = TheISA::genMachineCheckFault(); 627 delete mem_req; 628 memReq[tid] = NULL; 629 warn("Bad address!\n"); 630 } 631 assert(retryPkt == NULL); 632 assert(retryTid == -1); 633 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 634 fetchStatus[tid] = IcacheWaitRetry; 635 retryPkt = data_pkt; 636 retryTid = tid; 637 cacheBlocked = true; 638 return false; 639 } 640 641 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 642 643 lastIcacheStall[tid] = curTick; 644 645 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 646 "response.\n", tid); 647 648 fetchStatus[tid] = IcacheWaitResponse; 649 } else { 650 delete mem_req; 651 memReq[tid] = NULL; 652 } 653 654 ret_fault = fault; 655 return true; 656} 657 658template <class Impl> 659inline void 660DefaultFetch<Impl>::doSquash(const Addr &new_PC, 661 const Addr &new_NPC, unsigned tid) 662{ 663 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x, NPC to: %#x.\n", 664 tid, new_PC, new_NPC); 665 666 PC[tid] = new_PC; 667 nextPC[tid] = new_NPC; 668 nextNPC[tid] = new_NPC + instSize; 669 670 // Clear the icache miss if it's outstanding. 671 if (fetchStatus[tid] == IcacheWaitResponse) { 672 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 673 tid); 674 memReq[tid] = NULL; 675 } 676 677 // Get rid of the retrying packet if it was from this thread. 678 if (retryTid == tid) { 679 assert(cacheBlocked); 680 if (retryPkt) { 681 delete retryPkt->req; 682 delete retryPkt; 683 } 684 retryPkt = NULL; 685 retryTid = -1; 686 } 687 688 fetchStatus[tid] = Squashing; 689 690 ++fetchSquashCycles; 691} 692 693template<class Impl> 694void 695DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, const Addr &new_NPC, 696 const InstSeqNum &seq_num, 697 unsigned tid) 698{ 699 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 700 701 doSquash(new_PC, new_NPC, tid); 702 703 // Tell the CPU to remove any instructions that are in flight between 704 // fetch and decode. 705 cpu->removeInstsUntil(seq_num, tid); 706} 707 708template<class Impl> 709bool 710DefaultFetch<Impl>::checkStall(unsigned tid) const 711{ 712 bool ret_val = false; 713 714 if (cpu->contextSwitch) { 715 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 716 ret_val = true; 717 } else if (stalls[tid].decode) { 718 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 719 ret_val = true; 720 } else if (stalls[tid].rename) { 721 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 722 ret_val = true; 723 } else if (stalls[tid].iew) { 724 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 725 ret_val = true; 726 } else if (stalls[tid].commit) { 727 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 728 ret_val = true; 729 } 730 731 return ret_val; 732} 733 734template<class Impl> 735typename DefaultFetch<Impl>::FetchStatus 736DefaultFetch<Impl>::updateFetchStatus() 737{ 738 //Check Running 739 std::list<unsigned>::iterator threads = activeThreads->begin(); 740 std::list<unsigned>::iterator end = activeThreads->end(); 741 742 while (threads != end) { 743 unsigned tid = *threads++; 744 745 if (fetchStatus[tid] == Running || 746 fetchStatus[tid] == Squashing || 747 fetchStatus[tid] == IcacheAccessComplete) { 748 749 if (_status == Inactive) { 750 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 751 752 if (fetchStatus[tid] == IcacheAccessComplete) { 753 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 754 "completion\n",tid); 755 } 756 757 cpu->activateStage(O3CPU::FetchIdx); 758 } 759 760 return Active; 761 } 762 } 763 764 // Stage is switching from active to inactive, notify CPU of it. 765 if (_status == Active) { 766 DPRINTF(Activity, "Deactivating stage.\n"); 767 768 cpu->deactivateStage(O3CPU::FetchIdx); 769 } 770 771 return Inactive; 772} 773 774template <class Impl> 775void 776DefaultFetch<Impl>::squash(const Addr &new_PC, const Addr &new_NPC,
|
778 const InstSeqNum &seq_num, 779 bool squash_delay_slot, unsigned tid)
| 777 const InstSeqNum &seq_num, unsigned tid)
|
780{ 781 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 782 783 doSquash(new_PC, new_NPC, tid); 784
| 778{ 779 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 780 781 doSquash(new_PC, new_NPC, tid); 782
|
785#if ISA_HAS_DELAY_SLOT
| |
786 // Tell the CPU to remove any instructions that are not in the ROB.
| 783 // Tell the CPU to remove any instructions that are not in the ROB.
|
787 cpu->removeInstsNotInROB(tid, squash_delay_slot, seq_num); 788#else 789 // Tell the CPU to remove any instructions that are not in the ROB. 790 cpu->removeInstsNotInROB(tid, true, 0); 791#endif
| 784 cpu->removeInstsNotInROB(tid);
|
792} 793 794template <class Impl> 795void 796DefaultFetch<Impl>::tick() 797{ 798 std::list<unsigned>::iterator threads = activeThreads->begin(); 799 std::list<unsigned>::iterator end = activeThreads->end(); 800 bool status_change = false; 801 802 wroteToTimeBuffer = false; 803 804 while (threads != end) { 805 unsigned tid = *threads++; 806 807 // Check the signals for each thread to determine the proper status 808 // for each thread. 809 bool updated_status = checkSignalsAndUpdate(tid); 810 status_change = status_change || updated_status; 811 } 812 813 DPRINTF(Fetch, "Running stage.\n"); 814 815 // Reset the number of the instruction we're fetching. 816 numInst = 0; 817 818#if FULL_SYSTEM 819 if (fromCommit->commitInfo[0].interruptPending) { 820 interruptPending = true; 821 } 822 823 if (fromCommit->commitInfo[0].clearInterrupt) { 824 interruptPending = false; 825 } 826#endif 827 828 for (threadFetched = 0; threadFetched < numFetchingThreads; 829 threadFetched++) { 830 // Fetch each of the actively fetching threads. 831 fetch(status_change); 832 } 833 834 // Record number of instructions fetched this cycle for distribution. 835 fetchNisnDist.sample(numInst); 836 837 if (status_change) { 838 // Change the fetch stage status if there was a status change. 839 _status = updateFetchStatus(); 840 } 841 842 // If there was activity this cycle, inform the CPU of it. 843 if (wroteToTimeBuffer || cpu->contextSwitch) { 844 DPRINTF(Activity, "Activity this cycle.\n"); 845 846 cpu->activityThisCycle(); 847 } 848} 849 850template <class Impl> 851bool 852DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 853{ 854 // Update the per thread stall statuses. 855 if (fromDecode->decodeBlock[tid]) { 856 stalls[tid].decode = true; 857 } 858 859 if (fromDecode->decodeUnblock[tid]) { 860 assert(stalls[tid].decode); 861 assert(!fromDecode->decodeBlock[tid]); 862 stalls[tid].decode = false; 863 } 864 865 if (fromRename->renameBlock[tid]) { 866 stalls[tid].rename = true; 867 } 868 869 if (fromRename->renameUnblock[tid]) { 870 assert(stalls[tid].rename); 871 assert(!fromRename->renameBlock[tid]); 872 stalls[tid].rename = false; 873 } 874 875 if (fromIEW->iewBlock[tid]) { 876 stalls[tid].iew = true; 877 } 878 879 if (fromIEW->iewUnblock[tid]) { 880 assert(stalls[tid].iew); 881 assert(!fromIEW->iewBlock[tid]); 882 stalls[tid].iew = false; 883 } 884 885 if (fromCommit->commitBlock[tid]) { 886 stalls[tid].commit = true; 887 } 888 889 if (fromCommit->commitUnblock[tid]) { 890 assert(stalls[tid].commit); 891 assert(!fromCommit->commitBlock[tid]); 892 stalls[tid].commit = false; 893 } 894 895 // Check squash signals from commit. 896 if (fromCommit->commitInfo[tid].squash) { 897 898 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 899 "from commit.\n",tid);
| 785} 786 787template <class Impl> 788void 789DefaultFetch<Impl>::tick() 790{ 791 std::list<unsigned>::iterator threads = activeThreads->begin(); 792 std::list<unsigned>::iterator end = activeThreads->end(); 793 bool status_change = false; 794 795 wroteToTimeBuffer = false; 796 797 while (threads != end) { 798 unsigned tid = *threads++; 799 800 // Check the signals for each thread to determine the proper status 801 // for each thread. 802 bool updated_status = checkSignalsAndUpdate(tid); 803 status_change = status_change || updated_status; 804 } 805 806 DPRINTF(Fetch, "Running stage.\n"); 807 808 // Reset the number of the instruction we're fetching. 809 numInst = 0; 810 811#if FULL_SYSTEM 812 if (fromCommit->commitInfo[0].interruptPending) { 813 interruptPending = true; 814 } 815 816 if (fromCommit->commitInfo[0].clearInterrupt) { 817 interruptPending = false; 818 } 819#endif 820 821 for (threadFetched = 0; threadFetched < numFetchingThreads; 822 threadFetched++) { 823 // Fetch each of the actively fetching threads. 824 fetch(status_change); 825 } 826 827 // Record number of instructions fetched this cycle for distribution. 828 fetchNisnDist.sample(numInst); 829 830 if (status_change) { 831 // Change the fetch stage status if there was a status change. 832 _status = updateFetchStatus(); 833 } 834 835 // If there was activity this cycle, inform the CPU of it. 836 if (wroteToTimeBuffer || cpu->contextSwitch) { 837 DPRINTF(Activity, "Activity this cycle.\n"); 838 839 cpu->activityThisCycle(); 840 } 841} 842 843template <class Impl> 844bool 845DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 846{ 847 // Update the per thread stall statuses. 848 if (fromDecode->decodeBlock[tid]) { 849 stalls[tid].decode = true; 850 } 851 852 if (fromDecode->decodeUnblock[tid]) { 853 assert(stalls[tid].decode); 854 assert(!fromDecode->decodeBlock[tid]); 855 stalls[tid].decode = false; 856 } 857 858 if (fromRename->renameBlock[tid]) { 859 stalls[tid].rename = true; 860 } 861 862 if (fromRename->renameUnblock[tid]) { 863 assert(stalls[tid].rename); 864 assert(!fromRename->renameBlock[tid]); 865 stalls[tid].rename = false; 866 } 867 868 if (fromIEW->iewBlock[tid]) { 869 stalls[tid].iew = true; 870 } 871 872 if (fromIEW->iewUnblock[tid]) { 873 assert(stalls[tid].iew); 874 assert(!fromIEW->iewBlock[tid]); 875 stalls[tid].iew = false; 876 } 877 878 if (fromCommit->commitBlock[tid]) { 879 stalls[tid].commit = true; 880 } 881 882 if (fromCommit->commitUnblock[tid]) { 883 assert(stalls[tid].commit); 884 assert(!fromCommit->commitBlock[tid]); 885 stalls[tid].commit = false; 886 } 887 888 // Check squash signals from commit. 889 if (fromCommit->commitInfo[tid].squash) { 890 891 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 892 "from commit.\n",tid);
|
900 901#if ISA_HAS_DELAY_SLOT 902 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].bdelayDoneSeqNum; 903#else 904 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].doneSeqNum; 905#endif
| |
906 // In any case, squash. 907 squash(fromCommit->commitInfo[tid].nextPC, 908 fromCommit->commitInfo[tid].nextNPC,
| 893 // In any case, squash. 894 squash(fromCommit->commitInfo[tid].nextPC, 895 fromCommit->commitInfo[tid].nextNPC,
|
909 doneSeqNum, 910 fromCommit->commitInfo[tid].squashDelaySlot,
| 896 fromCommit->commitInfo[tid].doneSeqNum,
|
911 tid); 912 913 // Also check if there's a mispredict that happened. 914 if (fromCommit->commitInfo[tid].branchMispredict) { 915 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 916 fromCommit->commitInfo[tid].nextPC, 917 fromCommit->commitInfo[tid].branchTaken, 918 tid); 919 } else { 920 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 921 tid); 922 } 923 924 return true; 925 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 926 // Update the branch predictor if it wasn't a squashed instruction 927 // that was broadcasted. 928 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 929 } 930 931 // Check ROB squash signals from commit. 932 if (fromCommit->commitInfo[tid].robSquashing) { 933 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 934 935 // Continue to squash. 936 fetchStatus[tid] = Squashing; 937 938 return true; 939 } 940 941 // Check squash signals from decode. 942 if (fromDecode->decodeInfo[tid].squash) { 943 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 944 "from decode.\n",tid); 945 946 // Update the branch predictor. 947 if (fromDecode->decodeInfo[tid].branchMispredict) { 948 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 949 fromDecode->decodeInfo[tid].nextPC, 950 fromDecode->decodeInfo[tid].branchTaken, 951 tid); 952 } else { 953 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 954 tid); 955 } 956 957 if (fetchStatus[tid] != Squashing) { 958
| 897 tid); 898 899 // Also check if there's a mispredict that happened. 900 if (fromCommit->commitInfo[tid].branchMispredict) { 901 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 902 fromCommit->commitInfo[tid].nextPC, 903 fromCommit->commitInfo[tid].branchTaken, 904 tid); 905 } else { 906 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 907 tid); 908 } 909 910 return true; 911 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 912 // Update the branch predictor if it wasn't a squashed instruction 913 // that was broadcasted. 914 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 915 } 916 917 // Check ROB squash signals from commit. 918 if (fromCommit->commitInfo[tid].robSquashing) { 919 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 920 921 // Continue to squash. 922 fetchStatus[tid] = Squashing; 923 924 return true; 925 } 926 927 // Check squash signals from decode. 928 if (fromDecode->decodeInfo[tid].squash) { 929 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 930 "from decode.\n",tid); 931 932 // Update the branch predictor. 933 if (fromDecode->decodeInfo[tid].branchMispredict) { 934 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 935 fromDecode->decodeInfo[tid].nextPC, 936 fromDecode->decodeInfo[tid].branchTaken, 937 tid); 938 } else { 939 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 940 tid); 941 } 942 943 if (fetchStatus[tid] != Squashing) { 944
|
959#if ISA_HAS_DELAY_SLOT 960 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].bdelayDoneSeqNum; 961#else 962 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].doneSeqNum; 963#endif
| |
964 DPRINTF(Fetch, "Squashing from decode with PC = %#x, NPC = %#x\n", 965 fromDecode->decodeInfo[tid].nextPC, 966 fromDecode->decodeInfo[tid].nextNPC); 967 // Squash unless we're already squashing 968 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 969 fromDecode->decodeInfo[tid].nextNPC,
| 945 DPRINTF(Fetch, "Squashing from decode with PC = %#x, NPC = %#x\n", 946 fromDecode->decodeInfo[tid].nextPC, 947 fromDecode->decodeInfo[tid].nextNPC); 948 // Squash unless we're already squashing 949 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 950 fromDecode->decodeInfo[tid].nextNPC,
|
970 doneSeqNum,
| 951 fromDecode->decodeInfo[tid].doneSeqNum,
|
971 tid); 972 973 return true; 974 } 975 } 976 977 if (checkStall(tid) && 978 fetchStatus[tid] != IcacheWaitResponse && 979 fetchStatus[tid] != IcacheWaitRetry) { 980 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 981 982 fetchStatus[tid] = Blocked; 983 984 return true; 985 } 986 987 if (fetchStatus[tid] == Blocked || 988 fetchStatus[tid] == Squashing) { 989 // Switch status to running if fetch isn't being told to block or 990 // squash this cycle. 991 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 992 tid); 993 994 fetchStatus[tid] = Running; 995 996 return true; 997 } 998 999 // If we've reached this point, we have not gotten any signals that 1000 // cause fetch to change its status. Fetch remains the same as before. 1001 return false; 1002} 1003 1004template<class Impl> 1005void 1006DefaultFetch<Impl>::fetch(bool &status_change) 1007{ 1008 ////////////////////////////////////////// 1009 // Start actual fetch 1010 ////////////////////////////////////////// 1011 int tid = getFetchingThread(fetchPolicy); 1012 1013 if (tid == -1 || drainPending) { 1014 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1015 1016 // Breaks looping condition in tick() 1017 threadFetched = numFetchingThreads; 1018 return; 1019 } 1020 1021 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1022 1023 // The current PC. 1024 Addr &fetch_PC = PC[tid]; 1025 1026 Addr &fetch_NPC = nextPC[tid]; 1027 1028 // Fault code for memory access. 1029 Fault fault = NoFault; 1030 1031 // If returning from the delay of a cache miss, then update the status 1032 // to running, otherwise do the cache access. Possibly move this up 1033 // to tick() function. 1034 if (fetchStatus[tid] == IcacheAccessComplete) { 1035 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 1036 tid); 1037 1038 fetchStatus[tid] = Running; 1039 status_change = true; 1040 } else if (fetchStatus[tid] == Running) { 1041 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1042 "instruction, starting at PC %08p.\n", 1043 tid, fetch_PC); 1044 1045 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 1046 if (!fetch_success) { 1047 if (cacheBlocked) { 1048 ++icacheStallCycles; 1049 } else { 1050 ++fetchMiscStallCycles; 1051 } 1052 return; 1053 } 1054 } else { 1055 if (fetchStatus[tid] == Idle) { 1056 ++fetchIdleCycles; 1057 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1058 } else if (fetchStatus[tid] == Blocked) { 1059 ++fetchBlockedCycles; 1060 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1061 } else if (fetchStatus[tid] == Squashing) { 1062 ++fetchSquashCycles; 1063 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1064 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1065 ++icacheStallCycles; 1066 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", tid); 1067 } 1068 1069 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 1070 // fetch should do nothing. 1071 return; 1072 } 1073 1074 ++fetchCycles; 1075 1076 // If we had a stall due to an icache miss, then return. 1077 if (fetchStatus[tid] == IcacheWaitResponse) { 1078 ++icacheStallCycles; 1079 status_change = true; 1080 return; 1081 } 1082 1083 Addr next_PC = fetch_PC; 1084 Addr next_NPC = fetch_NPC; 1085 1086 InstSeqNum inst_seq; 1087 MachInst inst; 1088 ExtMachInst ext_inst; 1089 // @todo: Fix this hack. 1090 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 1091 1092 if (fault == NoFault) { 1093 // If the read of the first instruction was successful, then grab the 1094 // instructions from the rest of the cache line and put them into the 1095 // queue heading to decode. 1096 1097 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1098 "decode.\n",tid); 1099 1100 // Need to keep track of whether or not a predicted branch 1101 // ended this fetch block. 1102 bool predicted_branch = false; 1103 1104 for (; 1105 offset < cacheBlkSize && 1106 numInst < fetchWidth && 1107 !predicted_branch; 1108 ++numInst) { 1109 1110 // If we're branching after this instruction, quite fetching 1111 // from the same block then. 1112 predicted_branch = 1113 (fetch_PC + sizeof(TheISA::MachInst) != fetch_NPC); 1114 if (predicted_branch) { 1115 DPRINTF(Fetch, "Branch detected with PC = %#x, NPC = %#x\n", 1116 fetch_PC, fetch_NPC); 1117 } 1118 1119 1120 // Get a sequence number. 1121 inst_seq = cpu->getAndIncrementInstSeq(); 1122 1123 // Make sure this is a valid index. 1124 assert(offset <= cacheBlkSize - instSize); 1125 1126 // Get the instruction from the array of the cache line. 1127 inst = TheISA::gtoh(*reinterpret_cast<TheISA::MachInst *> 1128 (&cacheData[tid][offset])); 1129 1130 predecoder.setTC(cpu->thread[tid]->getTC());
| 952 tid); 953 954 return true; 955 } 956 } 957 958 if (checkStall(tid) && 959 fetchStatus[tid] != IcacheWaitResponse && 960 fetchStatus[tid] != IcacheWaitRetry) { 961 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 962 963 fetchStatus[tid] = Blocked; 964 965 return true; 966 } 967 968 if (fetchStatus[tid] == Blocked || 969 fetchStatus[tid] == Squashing) { 970 // Switch status to running if fetch isn't being told to block or 971 // squash this cycle. 972 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 973 tid); 974 975 fetchStatus[tid] = Running; 976 977 return true; 978 } 979 980 // If we've reached this point, we have not gotten any signals that 981 // cause fetch to change its status. Fetch remains the same as before. 982 return false; 983} 984 985template<class Impl> 986void 987DefaultFetch<Impl>::fetch(bool &status_change) 988{ 989 ////////////////////////////////////////// 990 // Start actual fetch 991 ////////////////////////////////////////// 992 int tid = getFetchingThread(fetchPolicy); 993 994 if (tid == -1 || drainPending) { 995 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 996 997 // Breaks looping condition in tick() 998 threadFetched = numFetchingThreads; 999 return; 1000 } 1001 1002 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1003 1004 // The current PC. 1005 Addr &fetch_PC = PC[tid]; 1006 1007 Addr &fetch_NPC = nextPC[tid]; 1008 1009 // Fault code for memory access. 1010 Fault fault = NoFault; 1011 1012 // If returning from the delay of a cache miss, then update the status 1013 // to running, otherwise do the cache access. Possibly move this up 1014 // to tick() function. 1015 if (fetchStatus[tid] == IcacheAccessComplete) { 1016 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 1017 tid); 1018 1019 fetchStatus[tid] = Running; 1020 status_change = true; 1021 } else if (fetchStatus[tid] == Running) { 1022 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1023 "instruction, starting at PC %08p.\n", 1024 tid, fetch_PC); 1025 1026 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 1027 if (!fetch_success) { 1028 if (cacheBlocked) { 1029 ++icacheStallCycles; 1030 } else { 1031 ++fetchMiscStallCycles; 1032 } 1033 return; 1034 } 1035 } else { 1036 if (fetchStatus[tid] == Idle) { 1037 ++fetchIdleCycles; 1038 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1039 } else if (fetchStatus[tid] == Blocked) { 1040 ++fetchBlockedCycles; 1041 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1042 } else if (fetchStatus[tid] == Squashing) { 1043 ++fetchSquashCycles; 1044 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1045 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1046 ++icacheStallCycles; 1047 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", tid); 1048 } 1049 1050 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 1051 // fetch should do nothing. 1052 return; 1053 } 1054 1055 ++fetchCycles; 1056 1057 // If we had a stall due to an icache miss, then return. 1058 if (fetchStatus[tid] == IcacheWaitResponse) { 1059 ++icacheStallCycles; 1060 status_change = true; 1061 return; 1062 } 1063 1064 Addr next_PC = fetch_PC; 1065 Addr next_NPC = fetch_NPC; 1066 1067 InstSeqNum inst_seq; 1068 MachInst inst; 1069 ExtMachInst ext_inst; 1070 // @todo: Fix this hack. 1071 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 1072 1073 if (fault == NoFault) { 1074 // If the read of the first instruction was successful, then grab the 1075 // instructions from the rest of the cache line and put them into the 1076 // queue heading to decode. 1077 1078 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1079 "decode.\n",tid); 1080 1081 // Need to keep track of whether or not a predicted branch 1082 // ended this fetch block. 1083 bool predicted_branch = false; 1084 1085 for (; 1086 offset < cacheBlkSize && 1087 numInst < fetchWidth && 1088 !predicted_branch; 1089 ++numInst) { 1090 1091 // If we're branching after this instruction, quite fetching 1092 // from the same block then. 1093 predicted_branch = 1094 (fetch_PC + sizeof(TheISA::MachInst) != fetch_NPC); 1095 if (predicted_branch) { 1096 DPRINTF(Fetch, "Branch detected with PC = %#x, NPC = %#x\n", 1097 fetch_PC, fetch_NPC); 1098 } 1099 1100 1101 // Get a sequence number. 1102 inst_seq = cpu->getAndIncrementInstSeq(); 1103 1104 // Make sure this is a valid index. 1105 assert(offset <= cacheBlkSize - instSize); 1106 1107 // Get the instruction from the array of the cache line. 1108 inst = TheISA::gtoh(*reinterpret_cast<TheISA::MachInst *> 1109 (&cacheData[tid][offset])); 1110 1111 predecoder.setTC(cpu->thread[tid]->getTC());
|
1131 predecoder.moreBytes(fetch_PC, fetch_PC, inst);
| 1112 predecoder.moreBytes(fetch_PC, 0, inst);
|
1132 1133 ext_inst = predecoder.getExtMachInst(); 1134 1135 // Create a new DynInst from the instruction fetched. 1136 DynInstPtr instruction = new DynInst(ext_inst, 1137 fetch_PC, fetch_NPC, 1138 next_PC, next_NPC, 1139 inst_seq, cpu); 1140 instruction->setTid(tid); 1141 1142 instruction->setASID(tid); 1143 1144 instruction->setThreadState(cpu->thread[tid]); 1145 1146 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1147 "[sn:%lli]\n", 1148 tid, instruction->readPC(), inst_seq); 1149 1150 //DPRINTF(Fetch, "[tid:%i]: MachInst is %#x\n", tid, ext_inst); 1151 1152 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1153 tid, instruction->staticInst->disassemble(fetch_PC)); 1154
| 1113 1114 ext_inst = predecoder.getExtMachInst(); 1115 1116 // Create a new DynInst from the instruction fetched. 1117 DynInstPtr instruction = new DynInst(ext_inst, 1118 fetch_PC, fetch_NPC, 1119 next_PC, next_NPC, 1120 inst_seq, cpu); 1121 instruction->setTid(tid); 1122 1123 instruction->setASID(tid); 1124 1125 instruction->setThreadState(cpu->thread[tid]); 1126 1127 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1128 "[sn:%lli]\n", 1129 tid, instruction->readPC(), inst_seq); 1130 1131 //DPRINTF(Fetch, "[tid:%i]: MachInst is %#x\n", tid, ext_inst); 1132 1133 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1134 tid, instruction->staticInst->disassemble(fetch_PC)); 1135
|
1155#if TRACING_ON
| |
1156 instruction->traceData = 1157 Trace::getInstRecord(curTick, cpu->tcBase(tid), 1158 instruction->staticInst, 1159 instruction->readPC());
| 1136 instruction->traceData = 1137 Trace::getInstRecord(curTick, cpu->tcBase(tid), 1138 instruction->staticInst, 1139 instruction->readPC());
|
1160#else 1161 instruction->traceData = NULL; 1162#endif
| |
1163 1164 ///FIXME This needs to be more robust in dealing with delay slots
| 1140 1141 ///FIXME This needs to be more robust in dealing with delay slots
|
1165#if !ISA_HAS_DELAY_SLOT 1166// predicted_branch |= 1167#endif
| |
1168 lookupAndUpdateNextPC(instruction, next_PC, next_NPC); 1169 predicted_branch |= (next_PC != fetch_NPC); 1170 1171 // Add instruction to the CPU's list of instructions. 1172 instruction->setInstListIt(cpu->addInst(instruction)); 1173 1174 // Write the instruction to the first slot in the queue 1175 // that heads to decode. 1176 toDecode->insts[numInst] = instruction; 1177 1178 toDecode->size++; 1179 1180 // Increment stat of fetched instructions. 1181 ++fetchedInsts; 1182 1183 // Move to the next instruction, unless we have a branch. 1184 fetch_PC = next_PC; 1185 fetch_NPC = next_NPC; 1186 1187 if (instruction->isQuiesce()) { 1188 DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!", 1189 curTick); 1190 fetchStatus[tid] = QuiescePending; 1191 ++numInst; 1192 status_change = true; 1193 break; 1194 } 1195 1196 offset += instSize; 1197 } 1198 1199 if (offset >= cacheBlkSize) { 1200 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1201 "block.\n", tid); 1202 } else if (numInst >= fetchWidth) { 1203 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1204 "for this cycle.\n", tid); 1205 } else if (predicted_branch) { 1206 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1207 "instruction encountered.\n", tid); 1208 } 1209 } 1210 1211 if (numInst > 0) { 1212 wroteToTimeBuffer = true; 1213 } 1214 1215 // Now that fetching is completed, update the PC to signify what the next 1216 // cycle will be. 1217 if (fault == NoFault) { 1218 PC[tid] = next_PC; 1219 nextPC[tid] = next_NPC; 1220 nextNPC[tid] = next_NPC + instSize;
| 1142 lookupAndUpdateNextPC(instruction, next_PC, next_NPC); 1143 predicted_branch |= (next_PC != fetch_NPC); 1144 1145 // Add instruction to the CPU's list of instructions. 1146 instruction->setInstListIt(cpu->addInst(instruction)); 1147 1148 // Write the instruction to the first slot in the queue 1149 // that heads to decode. 1150 toDecode->insts[numInst] = instruction; 1151 1152 toDecode->size++; 1153 1154 // Increment stat of fetched instructions. 1155 ++fetchedInsts; 1156 1157 // Move to the next instruction, unless we have a branch. 1158 fetch_PC = next_PC; 1159 fetch_NPC = next_NPC; 1160 1161 if (instruction->isQuiesce()) { 1162 DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!", 1163 curTick); 1164 fetchStatus[tid] = QuiescePending; 1165 ++numInst; 1166 status_change = true; 1167 break; 1168 } 1169 1170 offset += instSize; 1171 } 1172 1173 if (offset >= cacheBlkSize) { 1174 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1175 "block.\n", tid); 1176 } else if (numInst >= fetchWidth) { 1177 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1178 "for this cycle.\n", tid); 1179 } else if (predicted_branch) { 1180 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1181 "instruction encountered.\n", tid); 1182 } 1183 } 1184 1185 if (numInst > 0) { 1186 wroteToTimeBuffer = true; 1187 } 1188 1189 // Now that fetching is completed, update the PC to signify what the next 1190 // cycle will be. 1191 if (fault == NoFault) { 1192 PC[tid] = next_PC; 1193 nextPC[tid] = next_NPC; 1194 nextNPC[tid] = next_NPC + instSize;
|
1221#if ISA_HAS_DELAY_SLOT 1222 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, PC[tid]); 1223#else
| |
1224 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, next_PC);
| 1195 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, next_PC);
|
1225#endif
| |
1226 } else { 1227 // We shouldn't be in an icache miss and also have a fault (an ITB 1228 // miss) 1229 if (fetchStatus[tid] == IcacheWaitResponse) { 1230 panic("Fetch should have exited prior to this!"); 1231 } 1232 1233 // Send the fault to commit. This thread will not do anything 1234 // until commit handles the fault. The only other way it can 1235 // wake up is if a squash comes along and changes the PC. 1236#if FULL_SYSTEM 1237 assert(numInst < fetchWidth); 1238 // Get a sequence number. 1239 inst_seq = cpu->getAndIncrementInstSeq(); 1240 // We will use a nop in order to carry the fault. 1241 ext_inst = TheISA::NoopMachInst; 1242 1243 // Create a new DynInst from the dummy nop. 1244 DynInstPtr instruction = new DynInst(ext_inst, 1245 fetch_PC, fetch_NPC, 1246 next_PC, next_NPC, 1247 inst_seq, cpu); 1248 instruction->setPredTarg(next_PC, next_NPC); 1249 instruction->setTid(tid); 1250 1251 instruction->setASID(tid); 1252 1253 instruction->setThreadState(cpu->thread[tid]); 1254 1255 instruction->traceData = NULL; 1256 1257 instruction->setInstListIt(cpu->addInst(instruction)); 1258 1259 instruction->fault = fault; 1260 1261 toDecode->insts[numInst] = instruction; 1262 toDecode->size++; 1263 1264 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1265 1266 fetchStatus[tid] = TrapPending; 1267 status_change = true; 1268#else // !FULL_SYSTEM 1269 fetchStatus[tid] = TrapPending; 1270 status_change = true; 1271 1272#endif // FULL_SYSTEM 1273 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %08p", 1274 tid, fault->name(), PC[tid]); 1275 } 1276} 1277 1278template<class Impl> 1279void 1280DefaultFetch<Impl>::recvRetry() 1281{ 1282 if (retryPkt != NULL) { 1283 assert(cacheBlocked); 1284 assert(retryTid != -1); 1285 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1286 1287 if (icachePort->sendTiming(retryPkt)) { 1288 fetchStatus[retryTid] = IcacheWaitResponse; 1289 retryPkt = NULL; 1290 retryTid = -1; 1291 cacheBlocked = false; 1292 } 1293 } else { 1294 assert(retryTid == -1); 1295 // Access has been squashed since it was sent out. Just clear 1296 // the cache being blocked. 1297 cacheBlocked = false; 1298 } 1299} 1300 1301/////////////////////////////////////// 1302// // 1303// SMT FETCH POLICY MAINTAINED HERE // 1304// // 1305/////////////////////////////////////// 1306template<class Impl> 1307int 1308DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1309{ 1310 if (numThreads > 1) { 1311 switch (fetch_priority) { 1312 1313 case SingleThread: 1314 return 0; 1315 1316 case RoundRobin: 1317 return roundRobin(); 1318 1319 case IQ: 1320 return iqCount(); 1321 1322 case LSQ: 1323 return lsqCount(); 1324 1325 case Branch: 1326 return branchCount(); 1327 1328 default: 1329 return -1; 1330 } 1331 } else { 1332 std::list<unsigned>::iterator thread = activeThreads->begin(); 1333 assert(thread != activeThreads->end()); 1334 int tid = *thread; 1335 1336 if (fetchStatus[tid] == Running || 1337 fetchStatus[tid] == IcacheAccessComplete || 1338 fetchStatus[tid] == Idle) { 1339 return tid; 1340 } else { 1341 return -1; 1342 } 1343 } 1344 1345} 1346 1347 1348template<class Impl> 1349int 1350DefaultFetch<Impl>::roundRobin() 1351{ 1352 std::list<unsigned>::iterator pri_iter = priorityList.begin(); 1353 std::list<unsigned>::iterator end = priorityList.end(); 1354 1355 int high_pri; 1356 1357 while (pri_iter != end) { 1358 high_pri = *pri_iter; 1359 1360 assert(high_pri <= numThreads); 1361 1362 if (fetchStatus[high_pri] == Running || 1363 fetchStatus[high_pri] == IcacheAccessComplete || 1364 fetchStatus[high_pri] == Idle) { 1365 1366 priorityList.erase(pri_iter); 1367 priorityList.push_back(high_pri); 1368 1369 return high_pri; 1370 } 1371 1372 pri_iter++; 1373 } 1374 1375 return -1; 1376} 1377 1378template<class Impl> 1379int 1380DefaultFetch<Impl>::iqCount() 1381{ 1382 std::priority_queue<unsigned> PQ; 1383 1384 std::list<unsigned>::iterator threads = activeThreads->begin(); 1385 std::list<unsigned>::iterator end = activeThreads->end(); 1386 1387 while (threads != end) { 1388 unsigned tid = *threads++; 1389 1390 PQ.push(fromIEW->iewInfo[tid].iqCount); 1391 } 1392 1393 while (!PQ.empty()) { 1394 1395 unsigned high_pri = PQ.top(); 1396 1397 if (fetchStatus[high_pri] == Running || 1398 fetchStatus[high_pri] == IcacheAccessComplete || 1399 fetchStatus[high_pri] == Idle) 1400 return high_pri; 1401 else 1402 PQ.pop(); 1403 1404 } 1405 1406 return -1; 1407} 1408 1409template<class Impl> 1410int 1411DefaultFetch<Impl>::lsqCount() 1412{ 1413 std::priority_queue<unsigned> PQ; 1414 1415 std::list<unsigned>::iterator threads = activeThreads->begin(); 1416 std::list<unsigned>::iterator end = activeThreads->end(); 1417 1418 while (threads != end) { 1419 unsigned tid = *threads++; 1420 1421 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1422 } 1423 1424 while (!PQ.empty()) { 1425 1426 unsigned high_pri = PQ.top(); 1427 1428 if (fetchStatus[high_pri] == Running || 1429 fetchStatus[high_pri] == IcacheAccessComplete || 1430 fetchStatus[high_pri] == Idle) 1431 return high_pri; 1432 else 1433 PQ.pop(); 1434 1435 } 1436 1437 return -1; 1438} 1439 1440template<class Impl> 1441int 1442DefaultFetch<Impl>::branchCount() 1443{ 1444 std::list<unsigned>::iterator thread = activeThreads->begin(); 1445 assert(thread != activeThreads->end()); 1446 unsigned tid = *thread; 1447 1448 panic("Branch Count Fetch policy unimplemented\n"); 1449 return 0 * tid; 1450}
| 1196 } else { 1197 // We shouldn't be in an icache miss and also have a fault (an ITB 1198 // miss) 1199 if (fetchStatus[tid] == IcacheWaitResponse) { 1200 panic("Fetch should have exited prior to this!"); 1201 } 1202 1203 // Send the fault to commit. This thread will not do anything 1204 // until commit handles the fault. The only other way it can 1205 // wake up is if a squash comes along and changes the PC. 1206#if FULL_SYSTEM 1207 assert(numInst < fetchWidth); 1208 // Get a sequence number. 1209 inst_seq = cpu->getAndIncrementInstSeq(); 1210 // We will use a nop in order to carry the fault. 1211 ext_inst = TheISA::NoopMachInst; 1212 1213 // Create a new DynInst from the dummy nop. 1214 DynInstPtr instruction = new DynInst(ext_inst, 1215 fetch_PC, fetch_NPC, 1216 next_PC, next_NPC, 1217 inst_seq, cpu); 1218 instruction->setPredTarg(next_PC, next_NPC); 1219 instruction->setTid(tid); 1220 1221 instruction->setASID(tid); 1222 1223 instruction->setThreadState(cpu->thread[tid]); 1224 1225 instruction->traceData = NULL; 1226 1227 instruction->setInstListIt(cpu->addInst(instruction)); 1228 1229 instruction->fault = fault; 1230 1231 toDecode->insts[numInst] = instruction; 1232 toDecode->size++; 1233 1234 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1235 1236 fetchStatus[tid] = TrapPending; 1237 status_change = true; 1238#else // !FULL_SYSTEM 1239 fetchStatus[tid] = TrapPending; 1240 status_change = true; 1241 1242#endif // FULL_SYSTEM 1243 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %08p", 1244 tid, fault->name(), PC[tid]); 1245 } 1246} 1247 1248template<class Impl> 1249void 1250DefaultFetch<Impl>::recvRetry() 1251{ 1252 if (retryPkt != NULL) { 1253 assert(cacheBlocked); 1254 assert(retryTid != -1); 1255 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1256 1257 if (icachePort->sendTiming(retryPkt)) { 1258 fetchStatus[retryTid] = IcacheWaitResponse; 1259 retryPkt = NULL; 1260 retryTid = -1; 1261 cacheBlocked = false; 1262 } 1263 } else { 1264 assert(retryTid == -1); 1265 // Access has been squashed since it was sent out. Just clear 1266 // the cache being blocked. 1267 cacheBlocked = false; 1268 } 1269} 1270 1271/////////////////////////////////////// 1272// // 1273// SMT FETCH POLICY MAINTAINED HERE // 1274// // 1275/////////////////////////////////////// 1276template<class Impl> 1277int 1278DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1279{ 1280 if (numThreads > 1) { 1281 switch (fetch_priority) { 1282 1283 case SingleThread: 1284 return 0; 1285 1286 case RoundRobin: 1287 return roundRobin(); 1288 1289 case IQ: 1290 return iqCount(); 1291 1292 case LSQ: 1293 return lsqCount(); 1294 1295 case Branch: 1296 return branchCount(); 1297 1298 default: 1299 return -1; 1300 } 1301 } else { 1302 std::list<unsigned>::iterator thread = activeThreads->begin(); 1303 assert(thread != activeThreads->end()); 1304 int tid = *thread; 1305 1306 if (fetchStatus[tid] == Running || 1307 fetchStatus[tid] == IcacheAccessComplete || 1308 fetchStatus[tid] == Idle) { 1309 return tid; 1310 } else { 1311 return -1; 1312 } 1313 } 1314 1315} 1316 1317 1318template<class Impl> 1319int 1320DefaultFetch<Impl>::roundRobin() 1321{ 1322 std::list<unsigned>::iterator pri_iter = priorityList.begin(); 1323 std::list<unsigned>::iterator end = priorityList.end(); 1324 1325 int high_pri; 1326 1327 while (pri_iter != end) { 1328 high_pri = *pri_iter; 1329 1330 assert(high_pri <= numThreads); 1331 1332 if (fetchStatus[high_pri] == Running || 1333 fetchStatus[high_pri] == IcacheAccessComplete || 1334 fetchStatus[high_pri] == Idle) { 1335 1336 priorityList.erase(pri_iter); 1337 priorityList.push_back(high_pri); 1338 1339 return high_pri; 1340 } 1341 1342 pri_iter++; 1343 } 1344 1345 return -1; 1346} 1347 1348template<class Impl> 1349int 1350DefaultFetch<Impl>::iqCount() 1351{ 1352 std::priority_queue<unsigned> PQ; 1353 1354 std::list<unsigned>::iterator threads = activeThreads->begin(); 1355 std::list<unsigned>::iterator end = activeThreads->end(); 1356 1357 while (threads != end) { 1358 unsigned tid = *threads++; 1359 1360 PQ.push(fromIEW->iewInfo[tid].iqCount); 1361 } 1362 1363 while (!PQ.empty()) { 1364 1365 unsigned high_pri = PQ.top(); 1366 1367 if (fetchStatus[high_pri] == Running || 1368 fetchStatus[high_pri] == IcacheAccessComplete || 1369 fetchStatus[high_pri] == Idle) 1370 return high_pri; 1371 else 1372 PQ.pop(); 1373 1374 } 1375 1376 return -1; 1377} 1378 1379template<class Impl> 1380int 1381DefaultFetch<Impl>::lsqCount() 1382{ 1383 std::priority_queue<unsigned> PQ; 1384 1385 std::list<unsigned>::iterator threads = activeThreads->begin(); 1386 std::list<unsigned>::iterator end = activeThreads->end(); 1387 1388 while (threads != end) { 1389 unsigned tid = *threads++; 1390 1391 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1392 } 1393 1394 while (!PQ.empty()) { 1395 1396 unsigned high_pri = PQ.top(); 1397 1398 if (fetchStatus[high_pri] == Running || 1399 fetchStatus[high_pri] == IcacheAccessComplete || 1400 fetchStatus[high_pri] == Idle) 1401 return high_pri; 1402 else 1403 PQ.pop(); 1404 1405 } 1406 1407 return -1; 1408} 1409 1410template<class Impl> 1411int 1412DefaultFetch<Impl>::branchCount() 1413{ 1414 std::list<unsigned>::iterator thread = activeThreads->begin(); 1415 assert(thread != activeThreads->end()); 1416 unsigned tid = *thread; 1417 1418 panic("Branch Count Fetch policy unimplemented\n"); 1419 return 0 * tid; 1420}
|