rename.hh revision 2863 
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 */
30
31#ifndef __CPU_O3_RENAME_HH__
32#define __CPU_O3_RENAME_HH__
33
34#include <list>
35
36#include "base/statistics.hh"
37#include "base/timebuf.hh"
38
39/**
40 * DefaultRename handles both single threaded and SMT rename. Its
41 * width is specified by the parameters; each cycle it tries to rename
42 * that many instructions. It holds onto the rename history of all
43 * instructions with destination registers, storing the
44 * arch. register, the new physical register, and the old physical
45 * register, to allow for undoing of mappings if squashing happens, or
46 * freeing up registers upon commit. Rename handles blocking if the
47 * ROB, IQ, or LSQ is going to be full. Rename also handles barriers,
48 * and does so by stalling on the instruction until the ROB is empty
49 * and there are no instructions in flight to the ROB.
50 */
51template<class Impl>
52class DefaultRename
53{
54  public:
55    // Typedefs from the Impl.
56    typedef typename Impl::CPUPol CPUPol;
57    typedef typename Impl::DynInstPtr DynInstPtr;
58    typedef typename Impl::O3CPU O3CPU;
59    typedef typename Impl::Params Params;
60
61    // Typedefs from the CPUPol
62    typedef typename CPUPol::DecodeStruct DecodeStruct;
63    typedef typename CPUPol::RenameStruct RenameStruct;
64    typedef typename CPUPol::TimeStruct TimeStruct;
65    typedef typename CPUPol::FreeList FreeList;
66    typedef typename CPUPol::RenameMap RenameMap;
67    // These are used only for initialization.
68    typedef typename CPUPol::IEW IEW;
69    typedef typename CPUPol::Commit Commit;
70
71    // Typedefs from the ISA.
72    typedef TheISA::RegIndex RegIndex;
73
74    // A list is used to queue the instructions.  Barrier insts must
75    // be added to the front of the list, which is the only reason for
76    // using a list instead of a queue. (Most other stages use a
77    // queue)
78    typedef std::list<DynInstPtr> InstQueue;
79
80  public:
81    /** Overall rename status. Used to determine if the CPU can
82     * deschedule itself due to a lack of activity.
83     */
84    enum RenameStatus {
85        Active,
86        Inactive
87    };
88
89    /** Individual thread status. */
90    enum ThreadStatus {
91        Running,
92        Idle,
93        StartSquash,
94        Squashing,
95        Blocked,
96        Unblocking,
97        SerializeStall
98    };
99
100  private:
101    /** Rename status. */
102    RenameStatus _status;
103
104    /** Per-thread status. */
105    ThreadStatus renameStatus[Impl::MaxThreads];
106
107  public:
108    /** DefaultRename constructor. */
109    DefaultRename(Params *params);
110
111    /** Returns the name of rename. */
112    std::string name() const;
113
114    /** Registers statistics. */
115    void regStats();
116
117    /** Sets CPU pointer. */
118    void setCPU(O3CPU *cpu_ptr);
119
120    /** Sets the main backwards communication time buffer pointer. */
121    void setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr);
122
123    /** Sets pointer to time buffer used to communicate to the next stage. */
124    void setRenameQueue(TimeBuffer<RenameStruct> *rq_ptr);
125
126    /** Sets pointer to time buffer coming from decode. */
127    void setDecodeQueue(TimeBuffer<DecodeStruct> *dq_ptr);
128
129    /** Sets pointer to IEW stage. Used only for initialization. */
130    void setIEWStage(IEW *iew_stage)
131    { iew_ptr = iew_stage; }
132
133    /** Sets pointer to commit stage. Used only for initialization. */
134    void setCommitStage(Commit *commit_stage)
135    { commit_ptr = commit_stage; }
136
137  private:
138    /** Pointer to IEW stage. Used only for initialization. */
139    IEW *iew_ptr;
140
141    /** Pointer to commit stage. Used only for initialization. */
142    Commit *commit_ptr;
143
144  public:
145    /** Initializes variables for the stage. */
146    void initStage();
147
148    /** Sets pointer to list of active threads. */
149    void setActiveThreads(std::list<unsigned> *at_ptr);
150
151    /** Sets pointer to rename maps (per-thread structures). */
152    void setRenameMap(RenameMap rm_ptr[Impl::MaxThreads]);
153
154    /** Sets pointer to the free list. */
155    void setFreeList(FreeList *fl_ptr);
156
157    /** Sets pointer to the scoreboard. */
158    void setScoreboard(Scoreboard *_scoreboard);
159
160    /** Drains the rename stage. */
161    bool drain();
162
163    /** Resumes execution after a drain. */
164    void resume() { }
165
166    /** Switches out the rename stage. */
167    void switchOut();
168
169    /** Takes over from another CPU's thread. */
170    void takeOverFrom();
171
172    /** Squashes all instructions in a thread. */
173    void squash(unsigned tid);
174
175    /** Ticks rename, which processes all input signals and attempts to rename
176     * as many instructions as possible.
177     */
178    void tick();
179
180    /** Debugging function used to dump history buffer of renamings. */
181    void dumpHistory();
182
183  private:
184    /** Determines what to do based on rename's current status.
185     * @param status_change rename() sets this variable if there was a status
186     * change (ie switching from blocking to unblocking).
187     * @param tid Thread id to rename instructions from.
188     */
189    void rename(bool &status_change, unsigned tid);
190
191    /** Renames instructions for the given thread. Also handles serializing
192     * instructions.
193     */
194    void renameInsts(unsigned tid);
195
196    /** Inserts unused instructions from a given thread into the skid buffer,
197     * to be renamed once rename unblocks.
198     */
199    void skidInsert(unsigned tid);
200
201    /** Separates instructions from decode into individual lists of instructions
202     * sorted by thread.
203     */
204    void sortInsts();
205
206    /** Returns if all of the skid buffers are empty. */
207    bool skidsEmpty();
208
209    /** Updates overall rename status based on all of the threads' statuses. */
210    void updateStatus();
211
212    /** Switches rename to blocking, and signals back that rename has become
213     * blocked.
214     * @return Returns true if there is a status change.
215     */
216    bool block(unsigned tid);
217
218    /** Switches rename to unblocking if the skid buffer is empty, and signals
219     * back that rename has unblocked.
220     * @return Returns true if there is a status change.
221     */
222    bool unblock(unsigned tid);
223
224    /** Executes actual squash, removing squashed instructions. */
225    void doSquash(unsigned tid);
226
227    /** Removes a committed instruction's rename history. */
228    void removeFromHistory(InstSeqNum inst_seq_num, unsigned tid);
229
230    /** Renames the source registers of an instruction. */
231    inline void renameSrcRegs(DynInstPtr &inst, unsigned tid);
232
233    /** Renames the destination registers of an instruction. */
234    inline void renameDestRegs(DynInstPtr &inst, unsigned tid);
235
236    /** Calculates the number of free ROB entries for a specific thread. */
237    inline int calcFreeROBEntries(unsigned tid);
238
239    /** Calculates the number of free IQ entries for a specific thread. */
240    inline int calcFreeIQEntries(unsigned tid);
241
242    /** Calculates the number of free LSQ entries for a specific thread. */
243    inline int calcFreeLSQEntries(unsigned tid);
244
245    /** Returns the number of valid instructions coming from decode. */
246    unsigned validInsts();
247
248    /** Reads signals telling rename to block/unblock. */
249    void readStallSignals(unsigned tid);
250
251    /** Checks if any stages are telling rename to block. */
252    bool checkStall(unsigned tid);
253
254    /** Gets the number of free entries for a specific thread. */
255    void readFreeEntries(unsigned tid);
256
257    /** Checks the signals and updates the status. */
258    bool checkSignalsAndUpdate(unsigned tid);
259
260    /** Either serializes on the next instruction available in the InstQueue,
261     * or records that it must serialize on the next instruction to enter
262     * rename.
263     * @param inst_list The list of younger, unprocessed instructions for the
264     * thread that has the serializeAfter instruction.
265     * @param tid The thread id.
266     */
267    void serializeAfter(InstQueue &inst_list, unsigned tid);
268
269    /** Holds the information for each destination register rename. It holds
270     * the instruction's sequence number, the arch register, the old physical
271     * register for that arch. register, and the new physical register.
272     */
273    struct RenameHistory {
274        RenameHistory(InstSeqNum _instSeqNum, RegIndex _archReg,
275                      PhysRegIndex _newPhysReg, PhysRegIndex _prevPhysReg)
276            : instSeqNum(_instSeqNum), archReg(_archReg),
277              newPhysReg(_newPhysReg), prevPhysReg(_prevPhysReg)
278        {
279        }
280
281        /** The sequence number of the instruction that renamed. */
282        InstSeqNum instSeqNum;
283        /** The architectural register index that was renamed. */
284        RegIndex archReg;
285        /** The new physical register that the arch. register is renamed to. */
286        PhysRegIndex newPhysReg;
287        /** The old physical register that the arch. register was renamed to. */
288        PhysRegIndex prevPhysReg;
289    };
290
291    /** A per-thread list of all destination register renames, used to either
292     * undo rename mappings or free old physical registers.
293     */
294    std::list<RenameHistory> historyBuffer[Impl::MaxThreads];
295
296    /** Pointer to CPU. */
297    O3CPU *cpu;
298
299    /** Pointer to main time buffer used for backwards communication. */
300    TimeBuffer<TimeStruct> *timeBuffer;
301
302    /** Wire to get IEW's output from backwards time buffer. */
303    typename TimeBuffer<TimeStruct>::wire fromIEW;
304
305    /** Wire to get commit's output from backwards time buffer. */
306    typename TimeBuffer<TimeStruct>::wire fromCommit;
307
308    /** Wire to write infromation heading to previous stages. */
309    typename TimeBuffer<TimeStruct>::wire toDecode;
310
311    /** Rename instruction queue. */
312    TimeBuffer<RenameStruct> *renameQueue;
313
314    /** Wire to write any information heading to IEW. */
315    typename TimeBuffer<RenameStruct>::wire toIEW;
316
317    /** Decode instruction queue interface. */
318    TimeBuffer<DecodeStruct> *decodeQueue;
319
320    /** Wire to get decode's output from decode queue. */
321    typename TimeBuffer<DecodeStruct>::wire fromDecode;
322
323    /** Queue of all instructions coming from decode this cycle. */
324    InstQueue insts[Impl::MaxThreads];
325
326    /** Skid buffer between rename and decode. */
327    InstQueue skidBuffer[Impl::MaxThreads];
328
329    /** Rename map interface. */
330    RenameMap *renameMap[Impl::MaxThreads];
331
332    /** Free list interface. */
333    FreeList *freeList;
334
335    /** Pointer to the list of active threads. */
336    std::list<unsigned> *activeThreads;
337
338    /** Pointer to the scoreboard. */
339    Scoreboard *scoreboard;
340
341    /** Count of instructions in progress that have been sent off to the IQ
342     * and ROB, but are not yet included in their occupancy counts.
343     */
344    int instsInProgress[Impl::MaxThreads];
345
346    /** Variable that tracks if decode has written to the time buffer this
347     * cycle. Used to tell CPU if there is activity this cycle.
348     */
349    bool wroteToTimeBuffer;
350
351    /** Structures whose free entries impact the amount of instructions that
352     * can be renamed.
353     */
354    struct FreeEntries {
355        unsigned iqEntries;
356        unsigned lsqEntries;
357        unsigned robEntries;
358    };
359
360    /** Per-thread tracking of the number of free entries of back-end
361     * structures.
362     */
363    FreeEntries freeEntries[Impl::MaxThreads];
364
365    /** Records if the ROB is empty. In SMT mode the ROB may be dynamically
366     * partitioned between threads, so the ROB must tell rename when it is
367     * empty.
368     */
369    bool emptyROB[Impl::MaxThreads];
370
371    /** Source of possible stalls. */
372    struct Stalls {
373        bool iew;
374        bool commit;
375    };
376
377    /** Tracks which stages are telling decode to stall. */
378    Stalls stalls[Impl::MaxThreads];
379
380    /** The serialize instruction that rename has stalled on. */
381    DynInstPtr serializeInst[Impl::MaxThreads];
382
383    /** Records if rename needs to serialize on the next instruction for any
384     * thread.
385     */
386    bool serializeOnNextInst[Impl::MaxThreads];
387
388    /** Delay between iew and rename, in ticks. */
389    int iewToRenameDelay;
390
391    /** Delay between decode and rename, in ticks. */
392    int decodeToRenameDelay;
393
394    /** Delay between commit and rename, in ticks. */
395    unsigned commitToRenameDelay;
396
397    /** Rename width, in instructions. */
398    unsigned renameWidth;
399
400    /** Commit width, in instructions.  Used so rename knows how many
401     *  instructions might have freed registers in the previous cycle.
402     */
403    unsigned commitWidth;
404
405    /** The index of the instruction in the time buffer to IEW that rename is
406     * currently using.
407     */
408    unsigned toIEWIndex;
409
410    /** Whether or not rename needs to block this cycle. */
411    bool blockThisCycle;
412
413    /** The number of threads active in rename. */
414    unsigned numThreads;
415
416    /** The maximum skid buffer size. */
417    unsigned skidBufferMax;
418
419    /** Enum to record the source of a structure full stall.  Can come from
420     * either ROB, IQ, LSQ, and it is priortized in that order.
421     */
422    enum FullSource {
423        ROB,
424        IQ,
425        LSQ,
426        NONE
427    };
428
429    /** Function used to increment the stat that corresponds to the source of
430     * the stall.
431     */
432    inline void incrFullStat(const FullSource &source);
433
434    /** Stat for total number of cycles spent squashing. */
435    Stats::Scalar<> renameSquashCycles;
436    /** Stat for total number of cycles spent idle. */
437    Stats::Scalar<> renameIdleCycles;
438    /** Stat for total number of cycles spent blocking. */
439    Stats::Scalar<> renameBlockCycles;
440    /** Stat for total number of cycles spent stalling for a serializing inst. */
441    Stats::Scalar<> renameSerializeStallCycles;
442    /** Stat for total number of cycles spent running normally. */
443    Stats::Scalar<> renameRunCycles;
444    /** Stat for total number of cycles spent unblocking. */
445    Stats::Scalar<> renameUnblockCycles;
446    /** Stat for total number of renamed instructions. */
447    Stats::Scalar<> renameRenamedInsts;
448    /** Stat for total number of squashed instructions that rename discards. */
449    Stats::Scalar<> renameSquashedInsts;
450    /** Stat for total number of times that the ROB starts a stall in rename. */
451    Stats::Scalar<> renameROBFullEvents;
452    /** Stat for total number of times that the IQ starts a stall in rename. */
453    Stats::Scalar<> renameIQFullEvents;
454    /** Stat for total number of times that the LSQ starts a stall in rename. */
455    Stats::Scalar<> renameLSQFullEvents;
456    /** Stat for total number of times that rename runs out of free registers
457     * to use to rename. */
458    Stats::Scalar<> renameFullRegistersEvents;
459    /** Stat for total number of renamed destination registers. */
460    Stats::Scalar<> renameRenamedOperands;
461    /** Stat for total number of source register rename lookups. */
462    Stats::Scalar<> renameRenameLookups;
463    /** Stat for total number of committed renaming mappings. */
464    Stats::Scalar<> renameCommittedMaps;
465    /** Stat for total number of mappings that were undone due to a squash. */
466    Stats::Scalar<> renameUndoneMaps;
467    /** Number of serialize instructions handled. */
468    Stats::Scalar<> renamedSerializing;
469    /** Number of instructions marked as temporarily serializing. */
470    Stats::Scalar<> renamedTempSerializing;
471    /** Number of instructions inserted into skid buffers. */
472    Stats::Scalar<> renameSkidInsts;
473};
474
475#endif // __CPU_O3_RENAME_HH__
476