src/cpu/ozone/lw_back_end.hh

   1 /*
   2  * Copyright (c) 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  */
  30
  31 #ifndef __CPU_OZONE_LW_BACK_END_HH__
  32 #define __CPU_OZONE_LW_BACK_END_HH__
  33
  34 #include <list>
  35 #include <queue>
  36 #include <set>
  37 #include <string>
  38
  39 #include "arch/faults.hh"
  40 #include "base/timebuf.hh"
  41 #include "cpu/inst_seq.hh"
  42 #include "cpu/ozone/rename_table.hh"
  43 #include "cpu/ozone/thread_state.hh"
  44 #include "mem/request.hh"
  45 #include "sim/eventq.hh"
  46
  47 template <class>
  48 class Checker;
  49 class ThreadContext;
  50
  51 template <class Impl>
  52 class OzoneThreadState;
  53
  54 class Port;
  55
  56 template <class Impl>
  57 class LWBackEnd
  58 {
  59   public:
  60     typedef OzoneThreadState<Impl> Thread;
  61
  62     typedef typename Impl::Params Params;
  63     typedef typename Impl::DynInst DynInst;
  64     typedef typename Impl::DynInstPtr DynInstPtr;
  65     typedef typename Impl::OzoneCPU OzoneCPU;
  66     typedef typename Impl::FrontEnd FrontEnd;
  67     typedef typename Impl::OzoneCPU::CommStruct CommStruct;
  68
  69     struct SizeStruct {
  70         int size;
  71     };
  72
  73     typedef SizeStruct DispatchToIssue;
  74     typedef SizeStruct IssueToExec;
  75     typedef SizeStruct ExecToCommit;
  76     typedef SizeStruct Writeback;
  77
  78     TimeBuffer<DispatchToIssue> d2i;
  79     typename TimeBuffer<DispatchToIssue>::wire instsToDispatch;
  80     TimeBuffer<IssueToExec> i2e;
  81     typename TimeBuffer<IssueToExec>::wire instsToExecute;
  82     TimeBuffer<ExecToCommit> e2c;
  83     TimeBuffer<Writeback> numInstsToWB;
  84
  85     TimeBuffer<CommStruct> *comm;
  86     typename TimeBuffer<CommStruct>::wire toIEW;
  87     typename TimeBuffer<CommStruct>::wire fromCommit;
  88
  89     class TrapEvent : public Event {
  90       private:
  91         LWBackEnd<Impl> *be;
  92
  93       public:
  94         TrapEvent(LWBackEnd<Impl> *_be);
  95
  96         void process();
  97         const char *description();
  98     };
  99
 100     LWBackEnd(Params *params);
 101
 102     std::string name() const;
 103
 104     void regStats();
 105
 106     void setCPU(OzoneCPU *cpu_ptr);
 107
 108     void setFrontEnd(FrontEnd *front_end_ptr)
 109     { frontEnd = front_end_ptr; }
 110
 111     void setTC(ThreadContext *tc_ptr)
 112     { tc = tc_ptr; }
 113
 114     void setThreadState(Thread *thread_ptr)
 115     { thread = thread_ptr; }
 116
 117     void setCommBuffer(TimeBuffer<CommStruct> *_comm);
 118
 119     Port *getDcachePort() { return LSQ.getDcachePort(); }
 120
 121     void tick();
 122     void squash();
 123     void generateTCEvent() { tcSquash = true; }
 124     void squashFromTC();
 125     void squashFromTrap();
 126     void checkInterrupts();
 127     bool trapSquash;
 128     bool tcSquash;
 129
 130     template <class T>
 131     Fault read(RequestPtr req, T &data, int load_idx);
 132
 133     template <class T>
 134     Fault write(RequestPtr req, T &data, int store_idx);
 135
 136     Addr readCommitPC() { return commitPC; }
 137
 138     Addr commitPC;
 139
 140     Tick lastCommitCycle;
 141
 142     bool robEmpty() { return instList.empty(); }
 143
 144     bool isFull() { return numInsts >= numROBEntries; }
 145     bool isBlocked() { return status == Blocked || dispatchStatus == Blocked; }
 146
 147     void fetchFault(Fault &fault);
 148
 149     int wakeDependents(DynInstPtr &inst, bool memory_deps = false);
 150
 151     /** Tells memory dependence unit that a memory instruction needs to be
 152      * rescheduled. It will re-execute once replayMemInst() is called.
 153      */
 154     void rescheduleMemInst(DynInstPtr &inst);
 155
 156     /** Re-executes all rescheduled memory instructions. */
 157     void replayMemInst(DynInstPtr &inst);
 158
 159     /** Completes memory instruction. */
 160     void completeMemInst(DynInstPtr &inst) { }
 161
 162     void addDcacheMiss(DynInstPtr &inst)
 163     {
 164         waitingMemOps.insert(inst->seqNum);
 165         numWaitingMemOps++;
 166         DPRINTF(BE, "Adding a Dcache miss mem op [sn:%lli], total %i\n",
 167                 inst->seqNum, numWaitingMemOps);
 168     }
 169
 170     void removeDcacheMiss(DynInstPtr &inst)
 171     {
 172         assert(waitingMemOps.find(inst->seqNum) != waitingMemOps.end());
 173         waitingMemOps.erase(inst->seqNum);
 174         numWaitingMemOps--;
 175         DPRINTF(BE, "Removing a Dcache miss mem op [sn:%lli], total %i\n",
 176                 inst->seqNum, numWaitingMemOps);
 177     }
 178
 179     void addWaitingMemOp(DynInstPtr &inst)
 180     {
 181         waitingMemOps.insert(inst->seqNum);
 182         numWaitingMemOps++;
 183         DPRINTF(BE, "Adding a waiting mem op [sn:%lli], total %i\n",
 184                 inst->seqNum, numWaitingMemOps);
 185     }
 186
 187     void removeWaitingMemOp(DynInstPtr &inst)
 188     {
 189         assert(waitingMemOps.find(inst->seqNum) != waitingMemOps.end());
 190         waitingMemOps.erase(inst->seqNum);
 191         numWaitingMemOps--;
 192         DPRINTF(BE, "Removing a waiting mem op [sn:%lli], total %i\n",
 193                 inst->seqNum, numWaitingMemOps);
 194     }
 195
 196     void instToCommit(DynInstPtr &inst);
 197
 198     void switchOut();
 199     void doSwitchOut();
 200     void takeOverFrom(ThreadContext *old_tc = NULL);
 201
 202     bool isSwitchedOut() { return switchedOut; }
 203
 204   private:
 205     void generateTrapEvent(Tick latency = 0);
 206     void handleFault(Fault &fault, Tick latency = 0);
 207     void updateStructures();
 208     void dispatchInsts();
 209     void dispatchStall();
 210     void checkDispatchStatus();
 211     void executeInsts();
 212     void commitInsts();
 213     void addToLSQ(DynInstPtr &inst);
 214     void writebackInsts();
 215     bool commitInst(int inst_num);
 216     void squash(const InstSeqNum &sn);
 217     void squashDueToBranch(DynInstPtr &inst);
 218     void squashDueToMemViolation(DynInstPtr &inst);
 219     void squashDueToMemBlocked(DynInstPtr &inst);
 220     void updateExeInstStats(DynInstPtr &inst);
 221     void updateComInstStats(DynInstPtr &inst);
 222
 223   public:
 224     OzoneCPU *cpu;
 225
 226     FrontEnd *frontEnd;
 227
 228     ThreadContext *tc;
 229
 230     Thread *thread;
 231
 232     enum Status {
 233         Running,
 234         Idle,
 235         DcacheMissStall,
 236         DcacheMissComplete,
 237         Blocked,
 238         TrapPending
 239     };
 240
 241     Status status;
 242
 243     Status dispatchStatus;
 244
 245     Status commitStatus;
 246
 247     Counter funcExeInst;
 248
 249   private:
 250     typedef typename Impl::LdstQueue LdstQueue;
 251
 252     LdstQueue LSQ;
 253   public:
 254     RenameTable<Impl> commitRenameTable;
 255
 256     RenameTable<Impl> renameTable;
 257   private:
 258     // General back end width. Used if the more specific isn't given.
 259     int width;
 260
 261     // Dispatch width.
 262     int dispatchWidth;
 263     int numDispatchEntries;
 264     int dispatchSize;
 265
 266     int waitingInsts;
 267
 268     int issueWidth;
 269
 270     // Writeback width
 271     int wbWidth;
 272
 273     // Commit width
 274     int commitWidth;
 275
 276     /** Index into queue of instructions being written back. */
 277     unsigned wbNumInst;
 278
 279     /** Cycle number within the queue of instructions being written
 280      * back.  Used in case there are too many instructions writing
 281      * back at the current cycle and writesbacks need to be scheduled
 282      * for the future. See comments in instToCommit().
 283      */
 284     unsigned wbCycle;
 285
 286     int numROBEntries;
 287     int numInsts;
 288
 289     std::set<InstSeqNum> waitingMemOps;
 290     typedef std::set<InstSeqNum>::iterator MemIt;
 291     int numWaitingMemOps;
 292     unsigned maxOutstandingMemOps;
 293
 294     bool squashPending;
 295     InstSeqNum squashSeqNum;
 296     Addr squashNextPC;
 297
 298     Fault faultFromFetch;
 299     bool fetchHasFault;
 300
 301     bool switchedOut;
 302     bool switchPending;
 303
 304     DynInstPtr memBarrier;
 305
 306   private:
 307     struct pqCompare {
 308         bool operator() (const DynInstPtr &lhs, const DynInstPtr &rhs) const
 309         {
 310             return lhs->seqNum > rhs->seqNum;
 311         }
 312     };
 313
 314     typedef typename std::priority_queue<DynInstPtr, std::vector<DynInstPtr>, pqCompare> ReadyInstQueue;
 315     ReadyInstQueue exeList;
 316
 317     typedef typename std::list<DynInstPtr>::iterator InstListIt;
 318
 319     std::list<DynInstPtr> instList;
 320     std::list<DynInstPtr> waitingList;
 321     std::list<DynInstPtr> replayList;
 322     std::list<DynInstPtr> writeback;
 323
 324     int latency;
 325
 326     int squashLatency;
 327
 328     bool exactFullStall;
 329
 330     // number of cycles stalled for D-cache misses
 331 /*    Stats::Scalar<> dcacheStallCycles;
 332       Counter lastDcacheStall;
 333 */
 334     Stats::Vector<> rob_cap_events;
 335     Stats::Vector<> rob_cap_inst_count;
 336     Stats::Vector<> iq_cap_events;
 337     Stats::Vector<> iq_cap_inst_count;
 338     // total number of instructions executed
 339     Stats::Vector<> exe_inst;
 340     Stats::Vector<> exe_swp;
 341     Stats::Vector<> exe_nop;
 342     Stats::Vector<> exe_refs;
 343     Stats::Vector<> exe_loads;
 344     Stats::Vector<> exe_branches;
 345
 346     Stats::Vector<> issued_ops;
 347
 348     // total number of loads forwaded from LSQ stores
 349     Stats::Vector<> lsq_forw_loads;
 350
 351     // total number of loads ignored due to invalid addresses
 352     Stats::Vector<> inv_addr_loads;
 353
 354     // total number of software prefetches ignored due to invalid addresses
 355     Stats::Vector<> inv_addr_swpfs;
 356     // ready loads blocked due to memory disambiguation
 357     Stats::Vector<> lsq_blocked_loads;
 358
 359     Stats::Scalar<> lsqInversion;
 360
 361     Stats::Vector<> n_issued_dist;
 362     Stats::VectorDistribution<> issue_delay_dist;
 363
 364     Stats::VectorDistribution<> queue_res_dist;
 365 /*
 366     Stats::Vector<> stat_fu_busy;
 367     Stats::Vector2d<> stat_fuBusy;
 368     Stats::Vector<> dist_unissued;
 369     Stats::Vector2d<> stat_issued_inst_type;
 370
 371     Stats::Formula misspec_cnt;
 372     Stats::Formula misspec_ipc;
 373     Stats::Formula issue_rate;
 374     Stats::Formula issue_stores;
 375     Stats::Formula issue_op_rate;
 376     Stats::Formula fu_busy_rate;
 377     Stats::Formula commit_stores;
 378     Stats::Formula commit_ipc;
 379     Stats::Formula commit_ipb;
 380     Stats::Formula lsq_inv_rate;
 381 */
 382     Stats::Vector<> writeback_count;
 383     Stats::Vector<> producer_inst;
 384     Stats::Vector<> consumer_inst;
 385     Stats::Vector<> wb_penalized;
 386
 387     Stats::Formula wb_rate;
 388     Stats::Formula wb_fanout;
 389     Stats::Formula wb_penalized_rate;
 390
 391     // total number of instructions committed
 392     Stats::Vector<> stat_com_inst;
 393     Stats::Vector<> stat_com_swp;
 394     Stats::Vector<> stat_com_refs;
 395     Stats::Vector<> stat_com_loads;
 396     Stats::Vector<> stat_com_membars;
 397     Stats::Vector<> stat_com_branches;
 398
 399     Stats::Distribution<> n_committed_dist;
 400
 401     Stats::Scalar<> commit_eligible_samples;
 402     Stats::Vector<> commit_eligible;
 403
 404     Stats::Vector<> squashedInsts;
 405     Stats::Vector<> ROBSquashedInsts;
 406
 407     Stats::Scalar<> ROB_fcount;
 408     Stats::Formula ROB_full_rate;
 409
 410     Stats::Vector<>  ROB_count;  // cumulative ROB occupancy
 411     Stats::Formula ROB_occ_rate;
 412     Stats::VectorDistribution<> ROB_occ_dist;
 413   public:
 414     void dumpInsts();
 415
 416     Checker<DynInstPtr> *checker;
 417 };
 418
 419 template <class Impl>
 420 template <class T>
 421 Fault
 422 LWBackEnd<Impl>::read(RequestPtr req, T &data, int load_idx)
 423 {
 424     return LSQ.read(req, data, load_idx);
 425 }
 426
 427 template <class Impl>
 428 template <class T>
 429 Fault
 430 LWBackEnd<Impl>::write(RequestPtr req, T &data, int store_idx)
 431 {
 432     return LSQ.write(req, data, store_idx);
 433 }
 434
 435 #endif // __CPU_OZONE_LW_BACK_END_HH__