1/*
| 1/*
|
2 * Copyright (c) 2004-2005 The Regents of The University of Michigan
| 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.
| 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are 7 * met: redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer; 9 * redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution; 12 * neither the name of the copyright holders nor the names of its 13 * contributors may be used to endorse or promote products derived from 14 * this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
27 * 28 * Authors: Kevin Lim
| |
29 */ 30
| 27 */ 28
|
31// Remove this later; used only for debugging. 32#define OPCODE(X) (X >> 26) & 0x3f 33
| |
34#include "arch/isa_traits.hh"
| 29#include "arch/isa_traits.hh"
|
35#include "sim/byteswap.hh"
| 30#include "arch/utility.hh"
|
36#include "cpu/exetrace.hh"
| 31#include "cpu/exetrace.hh"
|
37#include "mem/base_mem.hh" 38#include "mem/mem_interface.hh" 39#include "mem/mem_req.hh"
| |
40#include "cpu/o3/fetch.hh"
| 32#include "cpu/o3/fetch.hh"
|
41
| 33#include "mem/packet.hh" 34#include "mem/request.hh" 35#include "sim/byteswap.hh" 36#include "sim/host.hh"
|
42#include "sim/root.hh" 43
| 37#include "sim/root.hh" 38
|
| 39#if FULL_SYSTEM 40#include "arch/tlb.hh" 41#include "arch/vtophys.hh" 42#include "base/remote_gdb.hh" 43#include "mem/functional/memory_control.hh" 44#include "mem/functional/physical.hh" 45#include "sim/system.hh" 46#endif // FULL_SYSTEM 47 48#include <algorithm> 49 50using namespace std; 51using namespace TheISA; 52
|
44template<class Impl>
| 53template<class Impl>
|
45SimpleFetch<Impl>::CacheCompletionEvent 46::CacheCompletionEvent(SimpleFetch *_fetch) 47 : Event(&mainEventQueue), 48 fetch(_fetch)
| 54Tick 55DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt)
|
49{
| 56{
|
| 57 panic("DefaultFetch doesn't expect recvAtomic callback!"); 58 return curTick;
|
50} 51 52template<class Impl> 53void
| 59} 60 61template<class Impl> 62void
|
54SimpleFetch<Impl>::CacheCompletionEvent::process()
| 63DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt)
|
55{
| 64{
|
56 fetch->processCacheCompletion();
| 65 panic("DefaultFetch doesn't expect recvFunctional callback!");
|
57} 58 59template<class Impl>
| 66} 67 68template<class Impl>
|
60const char * 61SimpleFetch<Impl>::CacheCompletionEvent::description()
| 69void 70DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status)
|
62{
| 71{
|
63 return "SimpleFetch cache completion event";
| 72 if (status == RangeChange) 73 return; 74 75 panic("DefaultFetch doesn't expect recvStatusChange callback!");
|
64} 65 66template<class Impl>
| 76} 77 78template<class Impl>
|
67SimpleFetch<Impl>::SimpleFetch(Params ¶ms) 68 : icacheInterface(params.icacheInterface), 69 branchPred(params), 70 decodeToFetchDelay(params.decodeToFetchDelay), 71 renameToFetchDelay(params.renameToFetchDelay), 72 iewToFetchDelay(params.iewToFetchDelay), 73 commitToFetchDelay(params.commitToFetchDelay), 74 fetchWidth(params.fetchWidth)
| 79bool 80DefaultFetch<Impl>::IcachePort::recvTiming(Packet *pkt)
|
75{
| 81{
|
76 DPRINTF(Fetch, "Fetch: Fetch constructor called\n");
| 82 fetch->processCacheCompletion(pkt); 83 return true; 84}
|
77
| 85
|
78 // Set status to idle. 79 _status = Idle;
| 86template<class Impl> 87void 88DefaultFetch<Impl>::IcachePort::recvRetry() 89{ 90 panic("DefaultFetch doesn't support retry yet."); 91 // we shouldn't get a retry unless we have a packet that we're 92 // waiting to transmit 93/* 94 assert(cpu->dcache_pkt != NULL); 95 assert(cpu->_status == DcacheRetry); 96 Packet *tmp = cpu->dcache_pkt; 97 if (sendTiming(tmp)) { 98 cpu->_status = DcacheWaitResponse; 99 cpu->dcache_pkt = NULL; 100 } 101*/ 102}
|
80
| 103
|
81 // Create a new memory request. 82 memReq = new MemReq(); 83 // Not sure of this parameter. I think it should be based on the 84 // thread number. 85#if !FULL_SYSTEM 86 memReq->asid = 0; 87#else 88 memReq->asid = 0; 89#endif // FULL_SYSTEM 90 memReq->data = new uint8_t[64];
| 104template<class Impl> 105DefaultFetch<Impl>::DefaultFetch(Params *params) 106 : branchPred(params), 107 decodeToFetchDelay(params->decodeToFetchDelay), 108 renameToFetchDelay(params->renameToFetchDelay), 109 iewToFetchDelay(params->iewToFetchDelay), 110 commitToFetchDelay(params->commitToFetchDelay), 111 fetchWidth(params->fetchWidth), 112 numThreads(params->numberOfThreads), 113 numFetchingThreads(params->smtNumFetchingThreads), 114 interruptPending(false) 115{ 116 if (numThreads > Impl::MaxThreads) 117 fatal("numThreads is not a valid value\n");
|
91
| 118
|
| 119 DPRINTF(Fetch, "Fetch constructor called\n"); 120 121 // Set fetch stage's status to inactive. 122 _status = Inactive; 123 124 string policy = params->smtFetchPolicy; 125 126 // Convert string to lowercase 127 std::transform(policy.begin(), policy.end(), policy.begin(), 128 (int(*)(int)) tolower); 129 130 // Figure out fetch policy 131 if (policy == "singlethread") { 132 fetchPolicy = SingleThread; 133 } else if (policy == "roundrobin") { 134 fetchPolicy = RoundRobin; 135 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 136 } else if (policy == "branch") { 137 fetchPolicy = Branch; 138 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 139 } else if (policy == "iqcount") { 140 fetchPolicy = IQ; 141 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 142 } else if (policy == "lsqcount") { 143 fetchPolicy = LSQ; 144 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 145 } else { 146 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 147 " RoundRobin,LSQcount,IQcount}\n"); 148 } 149
|
92 // Size of cache block.
| 150 // Size of cache block.
|
93 cacheBlkSize = icacheInterface ? icacheInterface->getBlockSize() : 64;
| 151 cacheBlkSize = 64;
|
94 95 // Create mask to get rid of offset bits. 96 cacheBlkMask = (cacheBlkSize - 1); 97
| 152 153 // Create mask to get rid of offset bits. 154 cacheBlkMask = (cacheBlkSize - 1); 155
|
| 156 for (int tid=0; tid < numThreads; tid++) { 157 158 fetchStatus[tid] = Running; 159 160 priorityList.push_back(tid); 161 162 memPkt[tid] = NULL; 163 164 // Create space to store a cache line. 165 cacheData[tid] = new uint8_t[cacheBlkSize]; 166 167 stalls[tid].decode = 0; 168 stalls[tid].rename = 0; 169 stalls[tid].iew = 0; 170 stalls[tid].commit = 0; 171 } 172
|
98 // Get the size of an instruction. 99 instSize = sizeof(MachInst);
| 173 // Get the size of an instruction. 174 instSize = sizeof(MachInst);
|
| 175}
|
100
| 176
|
101 // Create space to store a cache line. 102 cacheData = new uint8_t[cacheBlkSize];
| 177template <class Impl> 178std::string 179DefaultFetch<Impl>::name() const 180{ 181 return cpu->name() + ".fetch";
|
103} 104 105template <class Impl> 106void
| 182} 183 184template <class Impl> 185void
|
107SimpleFetch<Impl>::regStats()
| 186DefaultFetch<Impl>::regStats()
|
108{ 109 icacheStallCycles
| 187{ 188 icacheStallCycles
|
110 .name(name() + ".icacheStallCycles")
| 189 .name(name() + ".FETCH:icacheStallCycles")
|
111 .desc("Number of cycles fetch is stalled on an Icache miss") 112 .prereq(icacheStallCycles); 113 114 fetchedInsts
| 190 .desc("Number of cycles fetch is stalled on an Icache miss") 191 .prereq(icacheStallCycles); 192 193 fetchedInsts
|
115 .name(name() + ".fetchedInsts")
| 194 .name(name() + ".FETCH:Insts")
|
116 .desc("Number of instructions fetch has processed") 117 .prereq(fetchedInsts);
| 195 .desc("Number of instructions fetch has processed") 196 .prereq(fetchedInsts);
|
| 197 198 fetchedBranches 199 .name(name() + ".FETCH:Branches") 200 .desc("Number of branches that fetch encountered") 201 .prereq(fetchedBranches); 202
|
118 predictedBranches
| 203 predictedBranches
|
119 .name(name() + ".predictedBranches")
| 204 .name(name() + ".FETCH:predictedBranches")
|
120 .desc("Number of branches that fetch has predicted taken") 121 .prereq(predictedBranches);
| 205 .desc("Number of branches that fetch has predicted taken") 206 .prereq(predictedBranches);
|
| 207
|
122 fetchCycles
| 208 fetchCycles
|
123 .name(name() + ".fetchCycles")
| 209 .name(name() + ".FETCH:Cycles")
|
124 .desc("Number of cycles fetch has run and was not squashing or" 125 " blocked") 126 .prereq(fetchCycles);
| 210 .desc("Number of cycles fetch has run and was not squashing or" 211 " blocked") 212 .prereq(fetchCycles);
|
| 213
|
127 fetchSquashCycles
| 214 fetchSquashCycles
|
128 .name(name() + ".fetchSquashCycles")
| 215 .name(name() + ".FETCH:SquashCycles")
|
129 .desc("Number of cycles fetch has spent squashing") 130 .prereq(fetchSquashCycles);
| 216 .desc("Number of cycles fetch has spent squashing") 217 .prereq(fetchSquashCycles);
|
| 218 219 fetchIdleCycles 220 .name(name() + ".FETCH:IdleCycles") 221 .desc("Number of cycles fetch was idle") 222 .prereq(fetchIdleCycles); 223
|
131 fetchBlockedCycles
| 224 fetchBlockedCycles
|
132 .name(name() + ".fetchBlockedCycles")
| 225 .name(name() + ".FETCH:BlockedCycles")
|
133 .desc("Number of cycles fetch has spent blocked") 134 .prereq(fetchBlockedCycles);
| 226 .desc("Number of cycles fetch has spent blocked") 227 .prereq(fetchBlockedCycles);
|
| 228
|
135 fetchedCacheLines
| 229 fetchedCacheLines
|
136 .name(name() + ".fetchedCacheLines")
| 230 .name(name() + ".FETCH:CacheLines")
|
137 .desc("Number of cache lines fetched") 138 .prereq(fetchedCacheLines); 139
| 231 .desc("Number of cache lines fetched") 232 .prereq(fetchedCacheLines); 233
|
140 fetch_nisn_dist
| 234 fetchMiscStallCycles 235 .name(name() + ".FETCH:MiscStallCycles") 236 .desc("Number of cycles fetch has spent waiting on interrupts, or " 237 "bad addresses, or out of MSHRs") 238 .prereq(fetchMiscStallCycles); 239 240 fetchIcacheSquashes 241 .name(name() + ".FETCH:IcacheSquashes") 242 .desc("Number of outstanding Icache misses that were squashed") 243 .prereq(fetchIcacheSquashes); 244 245 fetchNisnDist
|
141 .init(/* base value */ 0, 142 /* last value */ fetchWidth, 143 /* bucket size */ 1)
| 246 .init(/* base value */ 0, 247 /* last value */ fetchWidth, 248 /* bucket size */ 1)
|
144 .name(name() + ".FETCH:rate_dist")
| 249 .name(name() + ".FETCH:rateDist")
|
145 .desc("Number of instructions fetched each cycle (Total)")
| 250 .desc("Number of instructions fetched each cycle (Total)")
|
146 .flags(Stats::pdf) 147 ;
| 251 .flags(Stats::pdf);
|
148
| 252
|
| 253 idleRate 254 .name(name() + ".FETCH:idleRate") 255 .desc("Percent of cycles fetch was idle") 256 .prereq(idleRate); 257 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 258 259 branchRate 260 .name(name() + ".FETCH:branchRate") 261 .desc("Number of branch fetches per cycle") 262 .flags(Stats::total); 263 branchRate = predictedBranches / cpu->numCycles; 264 265 fetchRate 266 .name(name() + ".FETCH:rate") 267 .desc("Number of inst fetches per cycle") 268 .flags(Stats::total); 269 fetchRate = fetchedInsts / cpu->numCycles; 270
|
149 branchPred.regStats(); 150} 151 152template<class Impl> 153void
| 271 branchPred.regStats(); 272} 273 274template<class Impl> 275void
|
154SimpleFetch<Impl>::setCPU(FullCPU *cpu_ptr)
| 276DefaultFetch<Impl>::setCPU(FullCPU *cpu_ptr)
|
155{
| 277{
|
156 DPRINTF(Fetch, "Fetch: Setting the CPU pointer.\n");
| 278 DPRINTF(Fetch, "Setting the CPU pointer.\n");
|
157 cpu = cpu_ptr;
| 279 cpu = cpu_ptr;
|
158 // This line will be removed eventually. 159 memReq->xc = cpu->xcBase();
| 280 281 // Name is finally available, so create the port. 282 icachePort = new IcachePort(this); 283 284 // Fetch needs to start fetching instructions at the very beginning, 285 // so it must start up in active state. 286 switchToActive();
|
160} 161 162template<class Impl> 163void
| 287} 288 289template<class Impl> 290void
|
164SimpleFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer)
| 291DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer)
|
165{
| 292{
|
166 DPRINTF(Fetch, "Fetch: Setting the time buffer pointer.\n");
| 293 DPRINTF(Fetch, "Setting the time buffer pointer.\n");
|
167 timeBuffer = time_buffer; 168 169 // Create wires to get information from proper places in time buffer. 170 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 171 fromRename = timeBuffer->getWire(-renameToFetchDelay); 172 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 173 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 174} 175 176template<class Impl> 177void
| 294 timeBuffer = time_buffer; 295 296 // Create wires to get information from proper places in time buffer. 297 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 298 fromRename = timeBuffer->getWire(-renameToFetchDelay); 299 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 300 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 301} 302 303template<class Impl> 304void
|
178SimpleFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr)
| 305DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr)
|
179{
| 306{
|
180 DPRINTF(Fetch, "Fetch: Setting the fetch queue pointer.\n");
| 307 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 308 activeThreads = at_ptr; 309} 310 311template<class Impl> 312void 313DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 314{ 315 DPRINTF(Fetch, "Setting the fetch queue pointer.\n");
|
181 fetchQueue = fq_ptr; 182 183 // Create wire to write information to proper place in fetch queue. 184 toDecode = fetchQueue->getWire(0); 185} 186
| 316 fetchQueue = fq_ptr; 317 318 // Create wire to write information to proper place in fetch queue. 319 toDecode = fetchQueue->getWire(0); 320} 321
|
| 322#if 0
|
187template<class Impl> 188void
| 323template<class Impl> 324void
|
189SimpleFetch<Impl>::processCacheCompletion()
| 325DefaultFetch<Impl>::setPageTable(PageTable *pt_ptr)
|
190{
| 326{
|
191 DPRINTF(Fetch, "Fetch: Waking up from cache miss.\n");
| 327 DPRINTF(Fetch, "Setting the page table pointer.\n"); 328#if !FULL_SYSTEM 329 pTable = pt_ptr; 330#endif 331} 332#endif
|
192
| 333
|
| 334template<class Impl> 335void 336DefaultFetch<Impl>::initStage() 337{ 338 for (int tid = 0; tid < numThreads; tid++) { 339 PC[tid] = cpu->readPC(tid); 340 nextPC[tid] = cpu->readNextPC(tid); 341 } 342} 343 344template<class Impl> 345void 346DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 347{ 348 unsigned tid = pkt->req->getThreadNum(); 349 350 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 351
|
193 // Only change the status if it's still waiting on the icache access 194 // to return. 195 // Can keep track of how many cache accesses go unused due to 196 // misspeculation here.
| 352 // Only change the status if it's still waiting on the icache access 353 // to return. 354 // Can keep track of how many cache accesses go unused due to 355 // misspeculation here.
|
197 if (_status == IcacheMissStall) 198 _status = IcacheMissComplete;
| 356 if (fetchStatus[tid] != IcacheWaitResponse || 357 pkt != memPkt[tid] || 358 isSwitchedOut()) { 359 ++fetchIcacheSquashes; 360 delete pkt; 361 return; 362 } 363 364 // Wake up the CPU (if it went to sleep and was waiting on this completion 365 // event). 366 cpu->wakeCPU(); 367 368 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 369 tid); 370 371 switchToActive(); 372 373 // Only switch to IcacheAccessComplete if we're not stalled as well. 374 if (checkStall(tid)) { 375 fetchStatus[tid] = Blocked; 376 } else { 377 fetchStatus[tid] = IcacheAccessComplete; 378 } 379 380// memcpy(cacheData[tid], memReq[tid]->data, memReq[tid]->size); 381 382 // Reset the mem req to NULL. 383 delete pkt->req; 384 delete pkt; 385 memPkt[tid] = NULL;
|
199} 200 201template <class Impl>
| 386} 387 388template <class Impl>
|
| 389void 390DefaultFetch<Impl>::switchOut() 391{ 392 switchedOut = true; 393 cpu->signalSwitched(); 394} 395 396template <class Impl> 397void 398DefaultFetch<Impl>::doSwitchOut() 399{ 400 branchPred.switchOut(); 401} 402 403template <class Impl> 404void 405DefaultFetch<Impl>::takeOverFrom() 406{ 407 // Reset all state 408 for (int i = 0; i < Impl::MaxThreads; ++i) { 409 stalls[i].decode = 0; 410 stalls[i].rename = 0; 411 stalls[i].iew = 0; 412 stalls[i].commit = 0; 413 PC[i] = cpu->readPC(i); 414 nextPC[i] = cpu->readNextPC(i); 415 fetchStatus[i] = Running; 416 } 417 numInst = 0; 418 wroteToTimeBuffer = false; 419 _status = Inactive; 420 switchedOut = false; 421 branchPred.takeOverFrom(); 422} 423 424template <class Impl> 425void 426DefaultFetch<Impl>::wakeFromQuiesce() 427{ 428 DPRINTF(Fetch, "Waking up from quiesce\n"); 429 // Hopefully this is safe 430 fetchStatus[0] = Running; 431} 432 433template <class Impl> 434inline void 435DefaultFetch<Impl>::switchToActive() 436{ 437 if (_status == Inactive) { 438 DPRINTF(Activity, "Activating stage.\n"); 439 440 cpu->activateStage(FullCPU::FetchIdx); 441 442 _status = Active; 443 } 444} 445 446template <class Impl> 447inline void 448DefaultFetch<Impl>::switchToInactive() 449{ 450 if (_status == Active) { 451 DPRINTF(Activity, "Deactivating stage.\n"); 452 453 cpu->deactivateStage(FullCPU::FetchIdx); 454 455 _status = Inactive; 456 } 457} 458 459template <class Impl>
|
202bool
| 460bool
|
203SimpleFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC)
| 461DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC)
|
204{ 205 // Do branch prediction check here. 206 // A bit of a misnomer...next_PC is actually the current PC until 207 // this function updates it. 208 bool predict_taken; 209 210 if (!inst->isControl()) { 211 next_PC = next_PC + instSize; 212 inst->setPredTarg(next_PC); 213 return false; 214 } 215
| 462{ 463 // Do branch prediction check here. 464 // A bit of a misnomer...next_PC is actually the current PC until 465 // this function updates it. 466 bool predict_taken; 467 468 if (!inst->isControl()) { 469 next_PC = next_PC + instSize; 470 inst->setPredTarg(next_PC); 471 return false; 472 } 473
|
216 predict_taken = branchPred.predict(inst, next_PC);
| 474 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber);
|
217
| 475
|
| 476 ++fetchedBranches; 477
|
218 if (predict_taken) { 219 ++predictedBranches; 220 } 221 222 return predict_taken; 223} 224 225template <class Impl>
| 478 if (predict_taken) { 479 ++predictedBranches; 480 } 481 482 return predict_taken; 483} 484 485template <class Impl>
|
226Fault 227SimpleFetch<Impl>::fetchCacheLine(Addr fetch_PC)
| 486bool 487DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid)
|
228{
| 488{
|
229 // Check if the instruction exists within the cache. 230 // If it does, then proceed on to read the instruction and the rest 231 // of the instructions in the cache line until either the end of the 232 // cache line or a predicted taken branch is encountered.
| 489 Fault fault = NoFault;
|
233 234#if FULL_SYSTEM 235 // Flag to say whether or not address is physical addr.
| 490 491#if FULL_SYSTEM 492 // Flag to say whether or not address is physical addr.
|
236 unsigned flags = cpu->inPalMode() ? PHYSICAL : 0;
| 493 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0;
|
237#else 238 unsigned flags = 0; 239#endif // FULL_SYSTEM 240
| 494#else 495 unsigned flags = 0; 496#endif // FULL_SYSTEM 497
|
241 Fault fault = NoFault;
| 498 if (interruptPending && flags == 0 || switchedOut) { 499 // Hold off fetch from getting new instructions while an interrupt 500 // is pending. 501 return false; 502 }
|
242 243 // Align the fetch PC so it's at the start of a cache block. 244 fetch_PC = icacheBlockAlignPC(fetch_PC); 245
| 503 504 // Align the fetch PC so it's at the start of a cache block. 505 fetch_PC = icacheBlockAlignPC(fetch_PC); 506
|
246 // Setup the memReq to do a read of the first isntruction's address.
| 507 // Setup the memReq to do a read of the first instruction's address.
|
247 // Set the appropriate read size and flags as well.
| 508 // Set the appropriate read size and flags as well.
|
248 memReq->cmd = Read; 249 memReq->reset(fetch_PC, cacheBlkSize, flags);
| 509 // Build request here. 510 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 511 fetch_PC, cpu->readCpuId(), tid);
|
250
| 512
|
| 513 memPkt[tid] = NULL; 514
|
251 // Translate the instruction request.
| 515 // Translate the instruction request.
|
252 // Should this function be 253 // in the CPU class ? Probably...ITB/DTB should exist within the 254 // CPU.
| 516//#if FULL_SYSTEM 517 fault = cpu->translateInstReq(mem_req); 518//#else 519// fault = pTable->translate(memReq[tid]); 520//#endif
|
255
| 521
|
256 fault = cpu->translateInstReq(memReq); 257
| |
258 // In the case of faults, the fetch stage may need to stall and wait
| 522 // In the case of faults, the fetch stage may need to stall and wait
|
259 // on what caused the fetch (ITB or Icache miss).
| 523 // for the ITB miss to be handled.
|
260 261 // If translation was successful, attempt to read the first 262 // instruction. 263 if (fault == NoFault) {
| 524 525 // If translation was successful, attempt to read the first 526 // instruction. 527 if (fault == NoFault) {
|
| 528#if FULL_SYSTEM 529 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 530 memReq[tid]->flags & UNCACHEABLE) { 531 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 532 "misspeculating path)!", 533 memReq[tid]->paddr); 534 ret_fault = TheISA::genMachineCheckFault(); 535 return false; 536 } 537#endif 538 539 // Build packet here. 540 PacketPtr data_pkt = new Packet(mem_req, 541 Packet::ReadReq, Packet::Broadcast); 542 data_pkt->dataStatic(cacheData[tid]); 543
|
264 DPRINTF(Fetch, "Fetch: Doing instruction read.\n");
| 544 DPRINTF(Fetch, "Fetch: Doing instruction read.\n");
|
265 fault = cpu->mem->read(memReq, cacheData); 266 // This read may change when the mem interface changes.
| |
267 268 fetchedCacheLines++;
| 545 546 fetchedCacheLines++;
|
269 }
| |
270
| 547
|
271 // Now do the timing access to see whether or not the instruction 272 // exists within the cache. 273 if (icacheInterface && fault == NoFault) { 274 DPRINTF(Fetch, "Fetch: Doing timing memory access.\n"); 275 memReq->completionEvent = NULL;
| 548 // Now do the timing access to see whether or not the instruction 549 // exists within the cache. 550 if (!icachePort->sendTiming(data_pkt)) { 551 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 552 ret_fault = NoFault; 553 return false; 554 }
|
276
| 555
|
277 memReq->time = curTick;
| 556 DPRINTF(Fetch, "Doing cache access.\n");
|
278
| 557
|
279 MemAccessResult result = icacheInterface->access(memReq);
| 558 lastIcacheStall[tid] = curTick;
|
280
| 559
|
281 // If the cache missed (in this model functional and timing 282 // memories are different), then schedule an event to wake 283 // up this stage once the cache miss completes. 284 if (result != MA_HIT && icacheInterface->doEvents()) { 285 memReq->completionEvent = new CacheCompletionEvent(this);
| 560 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 561 "response.\n", tid);
|
286
| 562
|
287 // How does current model work as far as individual 288 // stages scheduling/unscheduling? 289 // Perhaps have only the main CPU scheduled/unscheduled, 290 // and have it choose what stages to run appropriately. 291 292 DPRINTF(Fetch, "Fetch: Stalling due to icache miss.\n"); 293 _status = IcacheMissStall; 294 }
| 563 fetchStatus[tid] = IcacheWaitResponse;
|
295 } 296
| 564 } 565
|
297 return fault;
| 566 ret_fault = fault; 567 return true;
|
298} 299 300template <class Impl> 301inline void
| 568} 569 570template <class Impl> 571inline void
|
302SimpleFetch<Impl>::doSquash(const Addr &new_PC)
| 572DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid)
|
303{
| 573{
|
304 DPRINTF(Fetch, "Fetch: Squashing, setting PC to: %#x.\n", new_PC);
| 574 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 575 tid, new_PC);
|
305
| 576
|
306 cpu->setNextPC(new_PC + instSize); 307 cpu->setPC(new_PC);
| 577 PC[tid] = new_PC; 578 nextPC[tid] = new_PC + instSize;
|
308 309 // Clear the icache miss if it's outstanding.
| 579 580 // Clear the icache miss if it's outstanding.
|
310 if (_status == IcacheMissStall && icacheInterface) { 311 DPRINTF(Fetch, "Fetch: Squashing outstanding Icache miss.\n"); 312 // @todo: Use an actual thread number here. 313 icacheInterface->squash(0);
| 581 if (fetchStatus[tid] == IcacheWaitResponse) { 582 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 583 tid); 584 delete memPkt[tid]; 585 memPkt[tid] = NULL;
|
314 } 315
| 586 } 587
|
316 _status = Squashing;
| 588 fetchStatus[tid] = Squashing;
|
317 318 ++fetchSquashCycles; 319} 320 321template<class Impl> 322void
| 589 590 ++fetchSquashCycles; 591} 592 593template<class Impl> 594void
|
323SimpleFetch<Impl>::squashFromDecode(const Addr &new_PC, 324 const InstSeqNum &seq_num)
| 595DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 596 const InstSeqNum &seq_num, 597 unsigned tid)
|
325{
| 598{
|
326 DPRINTF(Fetch, "Fetch: Squashing from decode.\n");
| 599 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid);
|
327
| 600
|
328 doSquash(new_PC);
| 601 doSquash(new_PC, tid);
|
329 330 // Tell the CPU to remove any instructions that are in flight between 331 // fetch and decode.
| 602 603 // Tell the CPU to remove any instructions that are in flight between 604 // fetch and decode.
|
332 cpu->removeInstsUntil(seq_num);
| 605 cpu->removeInstsUntil(seq_num, tid);
|
333} 334
| 606} 607
|
| 608template<class Impl> 609bool 610DefaultFetch<Impl>::checkStall(unsigned tid) const 611{ 612 bool ret_val = false; 613 614 if (cpu->contextSwitch) { 615 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 616 ret_val = true; 617 } else if (stalls[tid].decode) { 618 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 619 ret_val = true; 620 } else if (stalls[tid].rename) { 621 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 622 ret_val = true; 623 } else if (stalls[tid].iew) { 624 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 625 ret_val = true; 626 } else if (stalls[tid].commit) { 627 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 628 ret_val = true; 629 } 630 631 return ret_val; 632} 633 634template<class Impl> 635typename DefaultFetch<Impl>::FetchStatus 636DefaultFetch<Impl>::updateFetchStatus() 637{ 638 //Check Running 639 list<unsigned>::iterator threads = (*activeThreads).begin(); 640 641 while (threads != (*activeThreads).end()) { 642 643 unsigned tid = *threads++; 644 645 if (fetchStatus[tid] == Running || 646 fetchStatus[tid] == Squashing || 647 fetchStatus[tid] == IcacheAccessComplete) { 648 649 if (_status == Inactive) { 650 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 651 652 if (fetchStatus[tid] == IcacheAccessComplete) { 653 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 654 "completion\n",tid); 655 } 656 657 cpu->activateStage(FullCPU::FetchIdx); 658 } 659 660 return Active; 661 } 662 } 663 664 // Stage is switching from active to inactive, notify CPU of it. 665 if (_status == Active) { 666 DPRINTF(Activity, "Deactivating stage.\n"); 667 668 cpu->deactivateStage(FullCPU::FetchIdx); 669 } 670 671 return Inactive; 672} 673
|
335template <class Impl> 336void
| 674template <class Impl> 675void
|
337SimpleFetch<Impl>::squash(const Addr &new_PC)
| 676DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid)
|
338{
| 677{
|
339 DPRINTF(Fetch, "Fetch: Squash from commit.\n");
| 678 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid);
|
340
| 679
|
341 doSquash(new_PC);
| 680 doSquash(new_PC, tid);
|
342 343 // Tell the CPU to remove any instructions that are not in the ROB.
| 681 682 // Tell the CPU to remove any instructions that are not in the ROB.
|
344 cpu->removeInstsNotInROB();
| 683 cpu->removeInstsNotInROB(tid);
|
345} 346
| 684} 685
|
347template
| 686template <class Impl>
|
348void
| 687void
|
349SimpleFetch<Impl>::tick()
| 688DefaultFetch<Impl>::tick()
|
350{
| 689{
|
| 690 list<unsigned>::iterator threads = (*activeThreads).begin(); 691 bool status_change = false; 692 693 wroteToTimeBuffer = false; 694 695 while (threads != (*activeThreads).end()) { 696 unsigned tid = *threads++; 697 698 // Check the signals for each thread to determine the proper status 699 // for each thread. 700 bool updated_status = checkSignalsAndUpdate(tid); 701 status_change = status_change || updated_status; 702 } 703 704 DPRINTF(Fetch, "Running stage.\n"); 705 706 // Reset the number of the instruction we're fetching. 707 numInst = 0; 708 709 if (fromCommit->commitInfo[0].interruptPending) { 710 interruptPending = true; 711 } 712 if (fromCommit->commitInfo[0].clearInterrupt) { 713 interruptPending = false; 714 } 715 716 for (threadFetched = 0; threadFetched < numFetchingThreads; 717 threadFetched++) { 718 // Fetch each of the actively fetching threads. 719 fetch(status_change); 720 } 721 722 // Record number of instructions fetched this cycle for distribution. 723 fetchNisnDist.sample(numInst); 724 725 if (status_change) { 726 // Change the fetch stage status if there was a status change. 727 _status = updateFetchStatus(); 728 } 729 730 // If there was activity this cycle, inform the CPU of it. 731 if (wroteToTimeBuffer || cpu->contextSwitch) { 732 DPRINTF(Activity, "Activity this cycle.\n"); 733 734 cpu->activityThisCycle(); 735 } 736} 737 738template <class Impl> 739bool 740DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 741{ 742 // Update the per thread stall statuses. 743 if (fromDecode->decodeBlock[tid]) { 744 stalls[tid].decode = true; 745 } 746 747 if (fromDecode->decodeUnblock[tid]) { 748 assert(stalls[tid].decode); 749 assert(!fromDecode->decodeBlock[tid]); 750 stalls[tid].decode = false; 751 } 752 753 if (fromRename->renameBlock[tid]) { 754 stalls[tid].rename = true; 755 } 756 757 if (fromRename->renameUnblock[tid]) { 758 assert(stalls[tid].rename); 759 assert(!fromRename->renameBlock[tid]); 760 stalls[tid].rename = false; 761 } 762 763 if (fromIEW->iewBlock[tid]) { 764 stalls[tid].iew = true; 765 } 766 767 if (fromIEW->iewUnblock[tid]) { 768 assert(stalls[tid].iew); 769 assert(!fromIEW->iewBlock[tid]); 770 stalls[tid].iew = false; 771 } 772 773 if (fromCommit->commitBlock[tid]) { 774 stalls[tid].commit = true; 775 } 776 777 if (fromCommit->commitUnblock[tid]) { 778 assert(stalls[tid].commit); 779 assert(!fromCommit->commitBlock[tid]); 780 stalls[tid].commit = false; 781 } 782
|
351 // Check squash signals from commit.
| 783 // Check squash signals from commit.
|
352 if (fromCommit->commitInfo.squash) { 353 DPRINTF(Fetch, "Fetch: Squashing instructions due to squash " 354 "from commit.\n");
| 784 if (fromCommit->commitInfo[tid].squash) {
|
355
| 785
|
| 786 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 787 "from commit.\n",tid); 788
|
356 // In any case, squash.
| 789 // In any case, squash.
|
357 squash(fromCommit->commitInfo.nextPC);
| 790 squash(fromCommit->commitInfo[tid].nextPC,tid);
|
358 359 // Also check if there's a mispredict that happened.
| 791 792 // Also check if there's a mispredict that happened.
|
360 if (fromCommit->commitInfo.branchMispredict) { 361 branchPred.squash(fromCommit->commitInfo.doneSeqNum, 362 fromCommit->commitInfo.nextPC, 363 fromCommit->commitInfo.branchTaken);
| 793 if (fromCommit->commitInfo[tid].branchMispredict) { 794 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 795 fromCommit->commitInfo[tid].nextPC, 796 fromCommit->commitInfo[tid].branchTaken, 797 tid);
|
364 } else {
| 798 } else {
|
365 branchPred.squash(fromCommit->commitInfo.doneSeqNum);
| 799 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 800 tid);
|
366 } 367
| 801 } 802
|
368 return; 369 } else if (fromCommit->commitInfo.doneSeqNum) {
| 803 return true; 804 } else if (fromCommit->commitInfo[tid].doneSeqNum) {
|
370 // Update the branch predictor if it wasn't a squashed instruction
| 805 // Update the branch predictor if it wasn't a squashed instruction
|
371 // that was braodcasted. 372 branchPred.update(fromCommit->commitInfo.doneSeqNum);
| 806 // that was broadcasted. 807 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid);
|
373 } 374 375 // Check ROB squash signals from commit.
| 808 } 809 810 // Check ROB squash signals from commit.
|
376 if (fromCommit->commitInfo.robSquashing) { 377 DPRINTF(Fetch, "Fetch: ROB is still squashing.\n");
| 811 if (fromCommit->commitInfo[tid].robSquashing) { 812 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing Thread %u.\n", tid);
|
378 379 // Continue to squash.
| 813 814 // Continue to squash.
|
380 _status = Squashing;
| 815 fetchStatus[tid] = Squashing;
|
381
| 816
|
382 ++fetchSquashCycles; 383 return;
| 817 return true;
|
384 } 385 386 // Check squash signals from decode.
| 818 } 819 820 // Check squash signals from decode.
|
387 if (fromDecode->decodeInfo.squash) { 388 DPRINTF(Fetch, "Fetch: Squashing instructions due to squash " 389 "from decode.\n");
| 821 if (fromDecode->decodeInfo[tid].squash) { 822 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 823 "from decode.\n",tid);
|
390 391 // Update the branch predictor.
| 824 825 // Update the branch predictor.
|
392 if (fromDecode->decodeInfo.branchMispredict) { 393 branchPred.squash(fromDecode->decodeInfo.doneSeqNum, 394 fromDecode->decodeInfo.nextPC, 395 fromDecode->decodeInfo.branchTaken);
| 826 if (fromDecode->decodeInfo[tid].branchMispredict) { 827 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 828 fromDecode->decodeInfo[tid].nextPC, 829 fromDecode->decodeInfo[tid].branchTaken, 830 tid);
|
396 } else {
| 831 } else {
|
397 branchPred.squash(fromDecode->decodeInfo.doneSeqNum);
| 832 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 833 tid);
|
398 } 399
| 834 } 835
|
400 if (_status != Squashing) { 401 // Squash unless we're already squashing? 402 squashFromDecode(fromDecode->decodeInfo.nextPC, 403 fromDecode->decodeInfo.doneSeqNum); 404 return;
| 836 if (fetchStatus[tid] != Squashing) { 837 // Squash unless we're already squashing 838 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 839 fromDecode->decodeInfo[tid].doneSeqNum, 840 tid); 841 842 return true;
|
405 } 406 } 407
| 843 } 844 } 845
|
408 // Check if any of the stall signals are high. 409 if (fromDecode->decodeInfo.stall || 410 fromRename->renameInfo.stall || 411 fromIEW->iewInfo.stall || 412 fromCommit->commitInfo.stall) 413 { 414 // Block stage, regardless of current status.
| 846 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 847 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid);
|
415
| 848
|
416 DPRINTF(Fetch, "Fetch: Stalling stage.\n"); 417 DPRINTF(Fetch, "Fetch: Statuses: Decode: %i Rename: %i IEW: %i " 418 "Commit: %i\n", 419 fromDecode->decodeInfo.stall, 420 fromRename->renameInfo.stall, 421 fromIEW->iewInfo.stall, 422 fromCommit->commitInfo.stall);
| 849 fetchStatus[tid] = Blocked;
|
423
| 850
|
424 _status = Blocked; 425 426 ++fetchBlockedCycles; 427 return; 428 } else if (_status == Blocked) { 429 // Unblock stage if status is currently blocked and none of the 430 // stall signals are being held high. 431 _status = Running; 432 433 ++fetchBlockedCycles; 434 return;
| 851 return true;
|
435 } 436
| 852 } 853
|
437 // If fetch has reached this point, then there are no squash signals 438 // still being held high. Check if fetch is in the squashing state; 439 // if so, fetch can switch to running. 440 // Similarly, there are no blocked signals still being held high. 441 // Check if fetch is in the blocked state; if so, fetch can switch to 442 // running. 443 if (_status == Squashing) { 444 DPRINTF(Fetch, "Fetch: Done squashing, switching to running.\n");
| 854 if (fetchStatus[tid] == Blocked || 855 fetchStatus[tid] == Squashing) { 856 // Switch status to running if fetch isn't being told to block or 857 // squash this cycle. 858 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 859 tid);
|
445
| 860
|
446 // Switch status to running 447 _status = Running;
| 861 fetchStatus[tid] = Running;
|
448
| 862
|
449 ++fetchCycles; 450 451 fetch(); 452 } else if (_status != IcacheMissStall) { 453 DPRINTF(Fetch, "Fetch: Running stage.\n"); 454 455 ++fetchCycles; 456 457 fetch();
| 863 return true;
|
458 }
| 864 }
|
| 865 866 // If we've reached this point, we have not gotten any signals that 867 // cause fetch to change its status. Fetch remains the same as before. 868 return false;
|
459} 460 461template<class Impl> 462void
| 869} 870 871template<class Impl> 872void
|
463SimpleFetch<Impl>::fetch()
| 873DefaultFetch<Impl>::fetch(bool &status_change)
|
464{ 465 ////////////////////////////////////////// 466 // Start actual fetch 467 //////////////////////////////////////////
| 874{ 875 ////////////////////////////////////////// 876 // Start actual fetch 877 //////////////////////////////////////////
|
| 878 int tid = getFetchingThread(fetchPolicy);
|
468
| 879
|
| 880 if (tid == -1) { 881 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 882 883 // Breaks looping condition in tick() 884 threadFetched = numFetchingThreads; 885 return; 886 } 887
|
469 // The current PC.
| 888 // The current PC.
|
470 Addr fetch_PC = cpu->readPC();
| 889 Addr &fetch_PC = PC[tid];
|
471 472 // Fault code for memory access. 473 Fault fault = NoFault; 474 475 // If returning from the delay of a cache miss, then update the status 476 // to running, otherwise do the cache access. Possibly move this up 477 // to tick() function.
| 890 891 // Fault code for memory access. 892 Fault fault = NoFault; 893 894 // If returning from the delay of a cache miss, then update the status 895 // to running, otherwise do the cache access. Possibly move this up 896 // to tick() function.
|
478 if (_status == IcacheMissComplete) { 479 DPRINTF(Fetch, "Fetch: Icache miss is complete.\n");
| 897 if (fetchStatus[tid] == IcacheAccessComplete) { 898 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 899 tid);
|
480
| 900
|
481 // Reset the completion event to NULL. 482 memReq->completionEvent = NULL;
| 901 fetchStatus[tid] = Running; 902 status_change = true; 903 } else if (fetchStatus[tid] == Running) { 904 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 905 "instruction, starting at PC %08p.\n", 906 tid, fetch_PC);
|
483
| 907
|
484 _status = Running;
| 908 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 909 if (!fetch_success) { 910 ++fetchMiscStallCycles; 911 return; 912 }
|
485 } else {
| 913 } else {
|
486 DPRINTF(Fetch, "Fetch: Attempting to translate and read " 487 "instruction, starting at PC %08p.\n", 488 fetch_PC);
| 914 if (fetchStatus[tid] == Idle) { 915 ++fetchIdleCycles; 916 } else if (fetchStatus[tid] == Blocked) { 917 ++fetchBlockedCycles; 918 } else if (fetchStatus[tid] == Squashing) { 919 ++fetchSquashCycles; 920 } else if (fetchStatus[tid] == IcacheWaitResponse) { 921 ++icacheStallCycles; 922 }
|
489
| 923
|
490 fault = fetchCacheLine(fetch_PC);
| 924 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 925 // fetch should do nothing. 926 return;
|
491 } 492
| 927 } 928
|
493 // If we had a stall due to an icache miss, then return. It'd 494 // be nicer if this were handled through the kind of fault that 495 // is returned by the function. 496 if (_status == IcacheMissStall) {
| 929 ++fetchCycles; 930 931 // If we had a stall due to an icache miss, then return. 932 if (fetchStatus[tid] == IcacheWaitResponse) { 933 ++icacheStallCycles; 934 status_change = true;
|
497 return; 498 } 499
| 935 return; 936 } 937
|
500 // As far as timing goes, the CPU will need to send an event through 501 // the MemReq in order to be woken up once the memory access completes. 502 // Probably have a status on a per thread basis so each thread can 503 // block independently and be woken up independently. 504
| |
505 Addr next_PC = fetch_PC; 506 InstSeqNum inst_seq; 507 MachInst inst;
| 938 Addr next_PC = fetch_PC; 939 InstSeqNum inst_seq; 940 MachInst inst;
|
508 unsigned offset = fetch_PC & cacheBlkMask; 509 unsigned fetched;
| 941 ExtMachInst ext_inst; 942 // @todo: Fix this hack. 943 unsigned offset = (fetch_PC & cacheBlkMask) & ~3;
|
510 511 if (fault == NoFault) { 512 // If the read of the first instruction was successful, then grab the 513 // instructions from the rest of the cache line and put them into the 514 // queue heading to decode. 515
| 944 945 if (fault == NoFault) { 946 // If the read of the first instruction was successful, then grab the 947 // instructions from the rest of the cache line and put them into the 948 // queue heading to decode. 949
|
516 DPRINTF(Fetch, "Fetch: Adding instructions to queue to decode.\n");
| 950 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 951 "decode.\n",tid);
|
517
| 952
|
518 ////////////////////////// 519 // Fetch first instruction 520 ////////////////////////// 521
| |
522 // Need to keep track of whether or not a predicted branch 523 // ended this fetch block. 524 bool predicted_branch = false; 525
| 953 // Need to keep track of whether or not a predicted branch 954 // ended this fetch block. 955 bool predicted_branch = false; 956
|
526 for (fetched = 0;
| 957 for (;
|
527 offset < cacheBlkSize &&
| 958 offset < cacheBlkSize &&
|
528 fetched < fetchWidth &&
| 959 numInst < fetchWidth &&
|
529 !predicted_branch;
| 960 !predicted_branch;
|
530 ++fetched) 531 {
| 961 ++numInst) {
|
532 533 // Get a sequence number. 534 inst_seq = cpu->getAndIncrementInstSeq(); 535 536 // Make sure this is a valid index. 537 assert(offset <= cacheBlkSize - instSize); 538 539 // Get the instruction from the array of the cache line. 540 inst = gtoh(*reinterpret_cast<MachInst *>
| 962 963 // Get a sequence number. 964 inst_seq = cpu->getAndIncrementInstSeq(); 965 966 // Make sure this is a valid index. 967 assert(offset <= cacheBlkSize - instSize); 968 969 // Get the instruction from the array of the cache line. 970 inst = gtoh(*reinterpret_cast<MachInst *>
|
541 (&cacheData[offset]));
| 971 (&cacheData[tid][offset]));
|
542
| 972
|
| 973 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 974
|
543 // Create a new DynInst from the instruction fetched.
| 975 // Create a new DynInst from the instruction fetched.
|
544 DynInstPtr instruction = new DynInst(inst, fetch_PC, next_PC,
| 976 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 977 next_PC,
|
545 inst_seq, cpu);
| 978 inst_seq, cpu);
|
| 979 instruction->setThread(tid);
|
546
| 980
|
547 DPRINTF(Fetch, "Fetch: Instruction %i created, with PC %#x\n", 548 inst_seq, instruction->readPC());
| 981 instruction->setASID(tid);
|
549
| 982
|
550 DPRINTF(Fetch, "Fetch: Instruction opcode is: %03p\n", 551 OPCODE(inst));
| 983 instruction->setState(cpu->thread[tid]);
|
552
| 984
|
| 985 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 986 "[sn:%lli]\n", 987 tid, instruction->readPC(), inst_seq); 988 989 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 990 tid, instruction->staticInst->disassemble(fetch_PC)); 991
|
553 instruction->traceData =
| 992 instruction->traceData =
|
554 Trace::getInstRecord(curTick, cpu->xcBase(), cpu,
| 993 Trace::getInstRecord(curTick, cpu->xcBase(tid), cpu,
|
555 instruction->staticInst,
| 994 instruction->staticInst,
|
556 instruction->readPC(), 0);
| 995 instruction->readPC(),tid);
|
557 558 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 559 560 // Add instruction to the CPU's list of instructions.
| 996 997 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 998 999 // Add instruction to the CPU's list of instructions.
|
561 cpu->addInst(instruction);
| 1000 instruction->setInstListIt(cpu->addInst(instruction));
|
562 563 // Write the instruction to the first slot in the queue 564 // that heads to decode.
| 1001 1002 // Write the instruction to the first slot in the queue 1003 // that heads to decode.
|
565 toDecode->insts[fetched] = instruction;
| 1004 toDecode->insts[numInst] = instruction;
|
566 567 toDecode->size++; 568 569 // Increment stat of fetched instructions. 570 ++fetchedInsts; 571 572 // Move to the next instruction, unless we have a branch. 573 fetch_PC = next_PC; 574
| 1005 1006 toDecode->size++; 1007 1008 // Increment stat of fetched instructions. 1009 ++fetchedInsts; 1010 1011 // Move to the next instruction, unless we have a branch. 1012 fetch_PC = next_PC; 1013
|
| 1014 if (instruction->isQuiesce()) { 1015 warn("%lli: Quiesce instruction encountered, halting fetch!", 1016 curTick); 1017 fetchStatus[tid] = QuiescePending; 1018 ++numInst; 1019 status_change = true; 1020 break; 1021 } 1022
|
575 offset+= instSize; 576 }
| 1023 offset+= instSize; 1024 }
|
| 1025 }
|
577
| 1026
|
578 fetch_nisn_dist.sample(fetched);
| 1027 if (numInst > 0) { 1028 wroteToTimeBuffer = true;
|
579 } 580 581 // Now that fetching is completed, update the PC to signify what the next
| 1029 } 1030 1031 // Now that fetching is completed, update the PC to signify what the next
|
582 // cycle will be. Might want to move this to the beginning of this 583 // function so that the PC updates at the beginning of everything. 584 // Or might want to leave setting the PC to the main CPU, with fetch 585 // only changing the nextPC (will require correct determination of 586 // next PC).
| 1032 // cycle will be.
|
587 if (fault == NoFault) {
| 1033 if (fault == NoFault) {
|
588 DPRINTF(Fetch, "Fetch: Setting PC to %08p.\n", next_PC); 589 cpu->setPC(next_PC); 590 cpu->setNextPC(next_PC + instSize);
| 1034 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1035 1036 PC[tid] = next_PC; 1037 nextPC[tid] = next_PC + instSize;
|
591 } else {
| 1038 } else {
|
592 // If the issue was an icache miss, then we can just return and 593 // wait until it is handled. 594 if (_status == IcacheMissStall) { 595 return;
| 1039 // We shouldn't be in an icache miss and also have a fault (an ITB 1040 // miss) 1041 if (fetchStatus[tid] == IcacheWaitResponse) { 1042 panic("Fetch should have exited prior to this!");
|
596 } 597
| 1043 } 1044
|
598 // Handle the fault. 599 // This stage will not be able to continue until all the ROB 600 // slots are empty, at which point the fault can be handled. 601 // The only other way it can wake up is if a squash comes along 602 // and changes the PC. Not sure how to handle that case...perhaps 603 // have it handled by the upper level CPU class which peeks into the 604 // time buffer and sees if a squash comes along, in which case it 605 // changes the status.
| 1045 // Send the fault to commit. This thread will not do anything 1046 // until commit handles the fault. The only other way it can 1047 // wake up is if a squash comes along and changes the PC. 1048#if FULL_SYSTEM 1049 assert(numInst != fetchWidth); 1050 // Get a sequence number. 1051 inst_seq = cpu->getAndIncrementInstSeq(); 1052 // We will use a nop in order to carry the fault. 1053 ext_inst = TheISA::NoopMachInst;
|
606
| 1054
|
607 DPRINTF(Fetch, "Fetch: Blocked, need to handle the trap.\n");
| 1055 // Create a new DynInst from the dummy nop. 1056 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1057 next_PC, 1058 inst_seq, cpu); 1059 instruction->setPredTarg(next_PC + instSize); 1060 instruction->setThread(tid);
|
608
| 1061
|
609 _status = Blocked; 610#if FULL_SYSTEM 611// cpu->trap(fault); 612 // Send a signal to the ROB indicating that there's a trap from the 613 // fetch stage that needs to be handled. Need to indicate that 614 // there's a fault, and the fault type.
| 1062 instruction->setASID(tid); 1063 1064 instruction->setState(cpu->thread[tid]); 1065 1066 instruction->traceData = NULL; 1067 1068 instruction->setInstListIt(cpu->addInst(instruction)); 1069 1070 instruction->fault = fault; 1071 1072 toDecode->insts[numInst] = instruction; 1073 toDecode->size++; 1074 1075 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1076 1077 fetchStatus[tid] = TrapPending; 1078 status_change = true; 1079 1080 warn("%lli fault (%d) detected @ PC %08p", curTick, fault, PC[tid]);
|
615#else // !FULL_SYSTEM
| 1081#else // !FULL_SYSTEM
|
616 fatal("fault (%d) detected @ PC %08p", fault, cpu->readPC());
| 1082 fatal("fault (%d) detected @ PC %08p", fault, PC[tid]);
|
617#endif // FULL_SYSTEM 618 } 619}
| 1083#endif // FULL_SYSTEM 1084 } 1085}
|
| 1086 1087 1088/////////////////////////////////////// 1089// // 1090// SMT FETCH POLICY MAINTAINED HERE // 1091// // 1092/////////////////////////////////////// 1093template<class Impl> 1094int 1095DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1096{ 1097 if (numThreads > 1) { 1098 switch (fetch_priority) { 1099 1100 case SingleThread: 1101 return 0; 1102 1103 case RoundRobin: 1104 return roundRobin(); 1105 1106 case IQ: 1107 return iqCount(); 1108 1109 case LSQ: 1110 return lsqCount(); 1111 1112 case Branch: 1113 return branchCount(); 1114 1115 default: 1116 return -1; 1117 } 1118 } else { 1119 int tid = *((*activeThreads).begin()); 1120 1121 if (fetchStatus[tid] == Running || 1122 fetchStatus[tid] == IcacheAccessComplete || 1123 fetchStatus[tid] == Idle) { 1124 return tid; 1125 } else { 1126 return -1; 1127 } 1128 } 1129 1130} 1131 1132 1133template<class Impl> 1134int 1135DefaultFetch<Impl>::roundRobin() 1136{ 1137 list<unsigned>::iterator pri_iter = priorityList.begin(); 1138 list<unsigned>::iterator end = priorityList.end(); 1139 1140 int high_pri; 1141 1142 while (pri_iter != end) { 1143 high_pri = *pri_iter; 1144 1145 assert(high_pri <= numThreads); 1146 1147 if (fetchStatus[high_pri] == Running || 1148 fetchStatus[high_pri] == IcacheAccessComplete || 1149 fetchStatus[high_pri] == Idle) { 1150 1151 priorityList.erase(pri_iter); 1152 priorityList.push_back(high_pri); 1153 1154 return high_pri; 1155 } 1156 1157 pri_iter++; 1158 } 1159 1160 return -1; 1161} 1162 1163template<class Impl> 1164int 1165DefaultFetch<Impl>::iqCount() 1166{ 1167 priority_queue<unsigned> PQ; 1168 1169 list<unsigned>::iterator threads = (*activeThreads).begin(); 1170 1171 while (threads != (*activeThreads).end()) { 1172 unsigned tid = *threads++; 1173 1174 PQ.push(fromIEW->iewInfo[tid].iqCount); 1175 } 1176 1177 while (!PQ.empty()) { 1178 1179 unsigned high_pri = PQ.top(); 1180 1181 if (fetchStatus[high_pri] == Running || 1182 fetchStatus[high_pri] == IcacheAccessComplete || 1183 fetchStatus[high_pri] == Idle) 1184 return high_pri; 1185 else 1186 PQ.pop(); 1187 1188 } 1189 1190 return -1; 1191} 1192 1193template<class Impl> 1194int 1195DefaultFetch<Impl>::lsqCount() 1196{ 1197 priority_queue<unsigned> PQ; 1198 1199 1200 list<unsigned>::iterator threads = (*activeThreads).begin(); 1201 1202 while (threads != (*activeThreads).end()) { 1203 unsigned tid = *threads++; 1204 1205 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1206 } 1207 1208 while (!PQ.empty()) { 1209 1210 unsigned high_pri = PQ.top(); 1211 1212 if (fetchStatus[high_pri] == Running || 1213 fetchStatus[high_pri] == IcacheAccessComplete || 1214 fetchStatus[high_pri] == Idle) 1215 return high_pri; 1216 else 1217 PQ.pop(); 1218 1219 } 1220 1221 return -1; 1222} 1223 1224template<class Impl> 1225int 1226DefaultFetch<Impl>::branchCount() 1227{ 1228 list<unsigned>::iterator threads = (*activeThreads).begin(); 1229 1230 return *threads; 1231}
|
| |