fetch.hh revision 2871
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#ifndef __CPU_O3_FETCH_HH__ 33#define __CPU_O3_FETCH_HH__ 34 35#include "arch/utility.hh" 36#include "base/statistics.hh" 37#include "base/timebuf.hh" 38#include "cpu/pc_event.hh" 39#include "mem/packet_impl.hh" 40#include "mem/port.hh" 41#include "sim/eventq.hh" 42 43/** 44 * DefaultFetch class handles both single threaded and SMT fetch. Its 45 * width is specified by the parameters; each cycle it tries to fetch 46 * that many instructions. It supports using a branch predictor to 47 * predict direction and targets. 48 * It supports the idling functionality of the CPU by indicating to 49 * the CPU when it is active and inactive. 50 */ 51template <class Impl> 52class DefaultFetch 53{ 54 public: 55 /** Typedefs from Impl. */ 56 typedef typename Impl::CPUPol CPUPol; 57 typedef typename Impl::DynInst DynInst; 58 typedef typename Impl::DynInstPtr DynInstPtr; 59 typedef typename Impl::O3CPU O3CPU; 60 typedef typename Impl::Params Params; 61 62 /** Typedefs from the CPU policy. */ 63 typedef typename CPUPol::BPredUnit BPredUnit; 64 typedef typename CPUPol::FetchStruct FetchStruct; 65 typedef typename CPUPol::TimeStruct TimeStruct; 66 67 /** Typedefs from ISA. */ 68 typedef TheISA::MachInst MachInst; 69 typedef TheISA::ExtMachInst ExtMachInst; 70 71 /** IcachePort class for DefaultFetch. Handles doing the 72 * communication with the cache/memory. 73 */ 74 class IcachePort : public Port 75 { 76 protected: 77 /** Pointer to fetch. */ 78 DefaultFetch<Impl> *fetch; 79 80 public: 81 /** Default constructor. */ 82 IcachePort(DefaultFetch<Impl> *_fetch) 83 : Port(_fetch->name() + "-iport"), fetch(_fetch) 84 { } 85 86 protected: 87 /** Atomic version of receive. Panics. */ 88 virtual Tick recvAtomic(PacketPtr pkt); 89 90 /** Functional version of receive. Panics. */ 91 virtual void recvFunctional(PacketPtr pkt); 92 93 /** Receives status change. Other than range changing, panics. */ 94 virtual void recvStatusChange(Status status); 95 96 /** Returns the address ranges of this device. */ 97 virtual void getDeviceAddressRanges(AddrRangeList &resp, 98 AddrRangeList &snoop) 99 { resp.clear(); snoop.clear(); } 100 101 /** Timing version of receive. Handles setting fetch to the 102 * proper status to start fetching. */ 103 virtual bool recvTiming(PacketPtr pkt); 104 105 /** Handles doing a retry of a failed fetch. */ 106 virtual void recvRetry(); 107 }; 108 109 public: 110 /** Overall fetch status. Used to determine if the CPU can 111 * deschedule itsef due to a lack of activity. 112 */ 113 enum FetchStatus { 114 Active, 115 Inactive 116 }; 117 118 /** Individual thread status. */ 119 enum ThreadStatus { 120 Running, 121 Idle, 122 Squashing, 123 Blocked, 124 Fetching, 125 TrapPending, 126 QuiescePending, 127 SwitchOut, 128 IcacheWaitResponse, 129 IcacheWaitRetry, 130 IcacheAccessComplete 131 }; 132 133 /** Fetching Policy, Add new policies here.*/ 134 enum FetchPriority { 135 SingleThread, 136 RoundRobin, 137 Branch, 138 IQ, 139 LSQ 140 }; 141 142 private: 143 /** Fetch status. */ 144 FetchStatus _status; 145 146 /** Per-thread status. */ 147 ThreadStatus fetchStatus[Impl::MaxThreads]; 148 149 /** Fetch policy. */ 150 FetchPriority fetchPolicy; 151 152 /** List that has the threads organized by priority. */ 153 std::list<unsigned> priorityList; 154 155 public: 156 /** DefaultFetch constructor. */ 157 DefaultFetch(Params *params); 158 159 /** Returns the name of fetch. */ 160 std::string name() const; 161 162 /** Registers statistics. */ 163 void regStats(); 164 165 /** Returns the icache port. */ 166 Port *getIcachePort() { return icachePort; } 167 168 /** Sets CPU pointer. */ 169 void setCPU(O3CPU *cpu_ptr); 170 171 /** Sets the main backwards communication time buffer pointer. */ 172 void setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer); 173 174 /** Sets pointer to list of active threads. */ 175 void setActiveThreads(std::list<unsigned> *at_ptr); 176 177 /** Sets pointer to time buffer used to communicate to the next stage. */ 178 void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr); 179 180 /** Initialize stage. */ 181 void initStage(); 182 183 /** Processes cache completion event. */ 184 void processCacheCompletion(PacketPtr pkt); 185 186 /** Begins the drain of the fetch stage. */ 187 bool drain(); 188 189 /** Resumes execution after a drain. */ 190 void resume(); 191 192 /** Tells fetch stage to prepare to be switched out. */ 193 void switchOut(); 194 195 /** Takes over from another CPU's thread. */ 196 void takeOverFrom(); 197 198 /** Checks if the fetch stage is switched out. */ 199 bool isSwitchedOut() { return switchedOut; } 200 201 /** Tells fetch to wake up from a quiesce instruction. */ 202 void wakeFromQuiesce(); 203 204 private: 205 /** Changes the status of this stage to active, and indicates this 206 * to the CPU. 207 */ 208 inline void switchToActive(); 209 210 /** Changes the status of this stage to inactive, and indicates 211 * this to the CPU. 212 */ 213 inline void switchToInactive(); 214 215 /** 216 * Looks up in the branch predictor to see if the next PC should be 217 * either next PC+=MachInst or a branch target. 218 * @param next_PC Next PC variable passed in by reference. It is 219 * expected to be set to the current PC; it will be updated with what 220 * the next PC will be. 221 * @return Whether or not a branch was predicted as taken. 222 */ 223 bool lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC); 224 225 /** 226 * Fetches the cache line that contains fetch_PC. Returns any 227 * fault that happened. Puts the data into the class variable 228 * cacheData. 229 * @param fetch_PC The PC address that is being fetched from. 230 * @param ret_fault The fault reference that will be set to the result of 231 * the icache access. 232 * @param tid Thread id. 233 * @return Any fault that occured. 234 */ 235 bool fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid); 236 237 /** Squashes a specific thread and resets the PC. */ 238 inline void doSquash(const Addr &new_PC, unsigned tid); 239 240 /** Squashes a specific thread and resets the PC. Also tells the CPU to 241 * remove any instructions between fetch and decode that should be sqaushed. 242 */ 243 void squashFromDecode(const Addr &new_PC, const InstSeqNum &seq_num, 244 unsigned tid); 245 246 /** Checks if a thread is stalled. */ 247 bool checkStall(unsigned tid) const; 248 249 /** Updates overall fetch stage status; to be called at the end of each 250 * cycle. */ 251 FetchStatus updateFetchStatus(); 252 253 public: 254 /** Squashes a specific thread and resets the PC. Also tells the CPU to 255 * remove any instructions that are not in the ROB. The source of this 256 * squash should be the commit stage. 257 */ 258 void squash(const Addr &new_PC, unsigned tid); 259 260 /** Ticks the fetch stage, processing all inputs signals and fetching 261 * as many instructions as possible. 262 */ 263 void tick(); 264 265 /** Checks all input signals and updates the status as necessary. 266 * @return: Returns if the status has changed due to input signals. 267 */ 268 bool checkSignalsAndUpdate(unsigned tid); 269 270 /** Does the actual fetching of instructions and passing them on to the 271 * next stage. 272 * @param status_change fetch() sets this variable if there was a status 273 * change (ie switching to IcacheMissStall). 274 */ 275 void fetch(bool &status_change); 276 277 /** Align a PC to the start of an I-cache block. */ 278 Addr icacheBlockAlignPC(Addr addr) 279 { 280 addr = TheISA::realPCToFetchPC(addr); 281 return (addr & ~(cacheBlkMask)); 282 } 283 284 private: 285 /** Handles retrying the fetch access. */ 286 void recvRetry(); 287 288 /** Returns the appropriate thread to fetch, given the fetch policy. */ 289 int getFetchingThread(FetchPriority &fetch_priority); 290 291 /** Returns the appropriate thread to fetch using a round robin policy. */ 292 int roundRobin(); 293 294 /** Returns the appropriate thread to fetch using the IQ count policy. */ 295 int iqCount(); 296 297 /** Returns the appropriate thread to fetch using the LSQ count policy. */ 298 int lsqCount(); 299 300 /** Returns the appropriate thread to fetch using the branch count policy. */ 301 int branchCount(); 302 303 private: 304 /** Pointer to the O3CPU. */ 305 O3CPU *cpu; 306 307 /** Time buffer interface. */ 308 TimeBuffer<TimeStruct> *timeBuffer; 309 310 /** Wire to get decode's information from backwards time buffer. */ 311 typename TimeBuffer<TimeStruct>::wire fromDecode; 312 313 /** Wire to get rename's information from backwards time buffer. */ 314 typename TimeBuffer<TimeStruct>::wire fromRename; 315 316 /** Wire to get iew's information from backwards time buffer. */ 317 typename TimeBuffer<TimeStruct>::wire fromIEW; 318 319 /** Wire to get commit's information from backwards time buffer. */ 320 typename TimeBuffer<TimeStruct>::wire fromCommit; 321 322 /** Internal fetch instruction queue. */ 323 TimeBuffer<FetchStruct> *fetchQueue; 324 325 //Might be annoying how this name is different than the queue. 326 /** Wire used to write any information heading to decode. */ 327 typename TimeBuffer<FetchStruct>::wire toDecode; 328 329 MemObject *mem; 330 331 /** Icache interface. */ 332 IcachePort *icachePort; 333 334 /** BPredUnit. */ 335 BPredUnit branchPred; 336 337 /** Per-thread fetch PC. */ 338 Addr PC[Impl::MaxThreads]; 339 340 /** Per-thread next PC. */ 341 Addr nextPC[Impl::MaxThreads]; 342 343#if THE_ISA != ALPHA_ISA 344 /** Per-thread next Next PC. 345 * This is not a real register but is used for 346 * architectures that use a branch-delay slot. 347 * (such as MIPS or Sparc) 348 */ 349 Addr nextNPC[Impl::MaxThreads]; 350#endif 351 352 /** Memory request used to access cache. */ 353 RequestPtr memReq[Impl::MaxThreads]; 354 355 /** Variable that tracks if fetch has written to the time buffer this 356 * cycle. Used to tell CPU if there is activity this cycle. 357 */ 358 bool wroteToTimeBuffer; 359 360 /** Tracks how many instructions has been fetched this cycle. */ 361 int numInst; 362 363 /** Source of possible stalls. */ 364 struct Stalls { 365 bool decode; 366 bool rename; 367 bool iew; 368 bool commit; 369 }; 370 371 /** Tracks which stages are telling fetch to stall. */ 372 Stalls stalls[Impl::MaxThreads]; 373 374 /** Decode to fetch delay, in ticks. */ 375 unsigned decodeToFetchDelay; 376 377 /** Rename to fetch delay, in ticks. */ 378 unsigned renameToFetchDelay; 379 380 /** IEW to fetch delay, in ticks. */ 381 unsigned iewToFetchDelay; 382 383 /** Commit to fetch delay, in ticks. */ 384 unsigned commitToFetchDelay; 385 386 /** The width of fetch in instructions. */ 387 unsigned fetchWidth; 388 389 /** Is the cache blocked? If so no threads can access it. */ 390 bool cacheBlocked; 391 392 /** The packet that is waiting to be retried. */ 393 PacketPtr retryPkt; 394 395 /** The thread that is waiting on the cache to tell fetch to retry. */ 396 int retryTid; 397 398 /** Cache block size. */ 399 int cacheBlkSize; 400 401 /** Mask to get a cache block's address. */ 402 Addr cacheBlkMask; 403 404 /** The cache line being fetched. */ 405 uint8_t *cacheData[Impl::MaxThreads]; 406 407 /** Size of instructions. */ 408 int instSize; 409 410 /** Icache stall statistics. */ 411 Counter lastIcacheStall[Impl::MaxThreads]; 412 413 /** List of Active Threads */ 414 std::list<unsigned> *activeThreads; 415 416 /** Number of threads. */ 417 unsigned numThreads; 418 419 /** Number of threads that are actively fetching. */ 420 unsigned numFetchingThreads; 421 422 /** Thread ID being fetched. */ 423 int threadFetched; 424 425 /** Checks if there is an interrupt pending. If there is, fetch 426 * must stop once it is not fetching PAL instructions. 427 */ 428 bool interruptPending; 429 430 /** Is there a drain pending. */ 431 bool drainPending; 432 433 /** Records if fetch is switched out. */ 434 bool switchedOut; 435 436 // @todo: Consider making these vectors and tracking on a per thread basis. 437 /** Stat for total number of cycles stalled due to an icache miss. */ 438 Stats::Scalar<> icacheStallCycles; 439 /** Stat for total number of fetched instructions. */ 440 Stats::Scalar<> fetchedInsts; 441 /** Total number of fetched branches. */ 442 Stats::Scalar<> fetchedBranches; 443 /** Stat for total number of predicted branches. */ 444 Stats::Scalar<> predictedBranches; 445 /** Stat for total number of cycles spent fetching. */ 446 Stats::Scalar<> fetchCycles; 447 /** Stat for total number of cycles spent squashing. */ 448 Stats::Scalar<> fetchSquashCycles; 449 /** Stat for total number of cycles spent blocked due to other stages in 450 * the pipeline. 451 */ 452 Stats::Scalar<> fetchIdleCycles; 453 /** Total number of cycles spent blocked. */ 454 Stats::Scalar<> fetchBlockedCycles; 455 /** Total number of cycles spent in any other state. */ 456 Stats::Scalar<> fetchMiscStallCycles; 457 /** Stat for total number of fetched cache lines. */ 458 Stats::Scalar<> fetchedCacheLines; 459 /** Total number of outstanding icache accesses that were dropped 460 * due to a squash. 461 */ 462 Stats::Scalar<> fetchIcacheSquashes; 463 /** Distribution of number of instructions fetched each cycle. */ 464 Stats::Distribution<> fetchNisnDist; 465 /** Rate of how often fetch was idle. */ 466 Stats::Formula idleRate; 467 /** Number of branch fetches per cycle. */ 468 Stats::Formula branchRate; 469 /** Number of instruction fetched per cycle. */ 470 Stats::Formula fetchRate; 471}; 472 473#endif //__CPU_O3_FETCH_HH__ 474