4 * All rights reserved.
5 *
6 * The license below extends only to copyright in the software and shall
7 * not be construed as granting a license to any other intellectual
8 * property including but not limited to intellectual property relating
9 * to a hardware implementation of the functionality of the software
10 * licensed hereunder. You may use the software subject to the license
11 * terms below provided that you ensure that this notice is replicated
12 * unmodified and in its entirety in all distributions of the software,
13 * modified or unmodified, in source code or in binary form.
14 *
15 * Copyright (c) 2004-2006 The Regents of The University of Michigan
16 * All rights reserved.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 *
41 * Authors: Kevin Lim
42 */
43
44#ifndef __CPU_O3_INST_QUEUE_HH__
45#define __CPU_O3_INST_QUEUE_HH__
46
47#include <list>
48#include <map>
49#include <queue>
50#include <vector>
51
52#include "base/statistics.hh"
53#include "base/types.hh"
54#include "cpu/o3/dep_graph.hh"
55#include "cpu/inst_seq.hh"
56#include "cpu/op_class.hh"
57#include "cpu/timebuf.hh"
58#include "sim/eventq.hh"
59
60struct DerivO3CPUParams;
61class FUPool;
62class MemInterface;
63
64/**
65 * A standard instruction queue class. It holds ready instructions, in
66 * order, in seperate priority queues to facilitate the scheduling of
67 * instructions. The IQ uses a separate linked list to track dependencies.
68 * Similar to the rename map and the free list, it expects that
69 * floating point registers have their indices start after the integer
70 * registers (ie with 96 int and 96 fp registers, regs 0-95 are integer
71 * and 96-191 are fp). This remains true even for both logical and
72 * physical register indices. The IQ depends on the memory dependence unit to
73 * track when memory operations are ready in terms of ordering; register
74 * dependencies are tracked normally. Right now the IQ also handles the
75 * execution timing; this is mainly to allow back-to-back scheduling without
76 * requiring IEW to be able to peek into the IQ. At the end of the execution
77 * latency, the instruction is put into the queue to execute, where it will
78 * have the execute() function called on it.
79 * @todo: Make IQ able to handle multiple FU pools.
80 */
81template <class Impl>
82class InstructionQueue
83{
84 public:
85 //Typedefs from the Impl.
86 typedef typename Impl::O3CPU O3CPU;
87 typedef typename Impl::DynInstPtr DynInstPtr;
88
89 typedef typename Impl::CPUPol::IEW IEW;
90 typedef typename Impl::CPUPol::MemDepUnit MemDepUnit;
91 typedef typename Impl::CPUPol::IssueStruct IssueStruct;
92 typedef typename Impl::CPUPol::TimeStruct TimeStruct;
93
94 // Typedef of iterator through the list of instructions.
95 typedef typename std::list<DynInstPtr>::iterator ListIt;
96
97 /** FU completion event class. */
98 class FUCompletion : public Event {
99 private:
100 /** Executing instruction. */
101 DynInstPtr inst;
102
103 /** Index of the FU used for executing. */
104 int fuIdx;
105
106 /** Pointer back to the instruction queue. */
107 InstructionQueue<Impl> *iqPtr;
108
109 /** Should the FU be added to the list to be freed upon
110 * completing this event.
111 */
112 bool freeFU;
113
114 public:
115 /** Construct a FU completion event. */
116 FUCompletion(DynInstPtr &_inst, int fu_idx,
117 InstructionQueue<Impl> *iq_ptr);
118
119 virtual void process();
120 virtual const char *description() const;
121 void setFreeFU() { freeFU = true; }
122 };
123
124 /** Constructs an IQ. */
125 InstructionQueue(O3CPU *cpu_ptr, IEW *iew_ptr, DerivO3CPUParams *params);
126
127 /** Destructs the IQ. */
128 ~InstructionQueue();
129
130 /** Returns the name of the IQ. */
131 std::string name() const;
132
133 /** Registers statistics. */
134 void regStats();
135
136 /** Resets all instruction queue state. */
137 void resetState();
138
139 /** Sets active threads list. */
140 void setActiveThreads(std::list<ThreadID> *at_ptr);
141
142 /** Sets the timer buffer between issue and execute. */
143 void setIssueToExecuteQueue(TimeBuffer<IssueStruct> *i2eQueue);
144
145 /** Sets the global time buffer. */
146 void setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr);
147
148 /** Perform sanity checks after a drain. */
149 void drainSanityCheck() const;
150
151 /** Takes over execution from another CPU's thread. */
152 void takeOverFrom();
153
154 /** Number of entries needed for given amount of threads. */
155 int entryAmount(ThreadID num_threads);
156
157 /** Resets max entries for all threads. */
158 void resetEntries();
159
160 /** Returns total number of free entries. */
161 unsigned numFreeEntries();
162
163 /** Returns number of free entries for a thread. */
164 unsigned numFreeEntries(ThreadID tid);
165
166 /** Returns whether or not the IQ is full. */
167 bool isFull();
168
169 /** Returns whether or not the IQ is full for a specific thread. */
170 bool isFull(ThreadID tid);
171
172 /** Returns if there are any ready instructions in the IQ. */
173 bool hasReadyInsts();
174
175 /** Inserts a new instruction into the IQ. */
176 void insert(DynInstPtr &new_inst);
177
178 /** Inserts a new, non-speculative instruction into the IQ. */
179 void insertNonSpec(DynInstPtr &new_inst);
180
181 /** Inserts a memory or write barrier into the IQ to make sure
182 * loads and stores are ordered properly.
183 */
184 void insertBarrier(DynInstPtr &barr_inst);
185
186 /** Returns the oldest scheduled instruction, and removes it from
187 * the list of instructions waiting to execute.
188 */
189 DynInstPtr getInstToExecute();
190
191 /** Returns a memory instruction that was referred due to a delayed DTB
192 * translation if it is now ready to execute.
193 */
194 DynInstPtr getDeferredMemInstToExecute();
195
196 /**
197 * Records the instruction as the producer of a register without
198 * adding it to the rest of the IQ.
199 */
200 void recordProducer(DynInstPtr &inst)
201 { addToProducers(inst); }
202
203 /** Process FU completion event. */
204 void processFUCompletion(DynInstPtr &inst, int fu_idx);
205
206 /**
207 * Schedules ready instructions, adding the ready ones (oldest first) to
208 * the queue to execute.
209 */
210 void scheduleReadyInsts();
211
212 /** Schedules a single specific non-speculative instruction. */
213 void scheduleNonSpec(const InstSeqNum &inst);
214
215 /**
216 * Commits all instructions up to and including the given sequence number,
217 * for a specific thread.
218 */
219 void commit(const InstSeqNum &inst, ThreadID tid = 0);
220
221 /** Wakes all dependents of a completed instruction. */
222 int wakeDependents(DynInstPtr &completed_inst);
223
224 /** Adds a ready memory instruction to the ready list. */
225 void addReadyMemInst(DynInstPtr &ready_inst);
226
227 /**
228 * Reschedules a memory instruction. It will be ready to issue once
229 * replayMemInst() is called.
230 */
231 void rescheduleMemInst(DynInstPtr &resched_inst);
232
233 /** Replays a memory instruction. It must be rescheduled first. */
234 void replayMemInst(DynInstPtr &replay_inst);
235
236 /** Completes a memory operation. */
237 void completeMemInst(DynInstPtr &completed_inst);
238
239 /**
240 * Defers a memory instruction when its DTB translation incurs a hw
241 * page table walk.
242 */
243 void deferMemInst(DynInstPtr &deferred_inst);
244
245 /** Indicates an ordering violation between a store and a load. */
246 void violation(DynInstPtr &store, DynInstPtr &faulting_load);
247
248 /**
249 * Squashes instructions for a thread. Squashing information is obtained
250 * from the time buffer.
251 */
252 void squash(ThreadID tid);
253
254 /** Returns the number of used entries for a thread. */
255 unsigned getCount(ThreadID tid) { return count[tid]; };
256
257 /** Debug function to print all instructions. */
258 void printInsts();
259
260 private:
261 /** Does the actual squashing. */
262 void doSquash(ThreadID tid);
263
264 /////////////////////////
265 // Various pointers
266 /////////////////////////
267
268 /** Pointer to the CPU. */
269 O3CPU *cpu;
270
271 /** Cache interface. */
272 MemInterface *dcacheInterface;
273
274 /** Pointer to IEW stage. */
275 IEW *iewStage;
276
277 /** The memory dependence unit, which tracks/predicts memory dependences
278 * between instructions.
279 */
280 MemDepUnit memDepUnit[Impl::MaxThreads];
281
282 /** The queue to the execute stage. Issued instructions will be written
283 * into it.
284 */
285 TimeBuffer<IssueStruct> *issueToExecuteQueue;
286
287 /** The backwards time buffer. */
288 TimeBuffer<TimeStruct> *timeBuffer;
289
290 /** Wire to read information from timebuffer. */
291 typename TimeBuffer<TimeStruct>::wire fromCommit;
292
293 /** Function unit pool. */
294 FUPool *fuPool;
295
296 //////////////////////////////////////
297 // Instruction lists, ready queues, and ordering
298 //////////////////////////////////////
299
300 /** List of all the instructions in the IQ (some of which may be issued). */
301 std::list<DynInstPtr> instList[Impl::MaxThreads];
302
303 /** List of instructions that are ready to be executed. */
304 std::list<DynInstPtr> instsToExecute;
305
306 /** List of instructions waiting for their DTB translation to
307 * complete (hw page table walk in progress).
308 */
309 std::list<DynInstPtr> deferredMemInsts;
310
311 /**
312 * Struct for comparing entries to be added to the priority queue.
313 * This gives reverse ordering to the instructions in terms of
314 * sequence numbers: the instructions with smaller sequence
315 * numbers (and hence are older) will be at the top of the
316 * priority queue.
317 */
318 struct pqCompare {
319 bool operator() (const DynInstPtr &lhs, const DynInstPtr &rhs) const
320 {
321 return lhs->seqNum > rhs->seqNum;
322 }
323 };
324
325 typedef std::priority_queue<DynInstPtr, std::vector<DynInstPtr>, pqCompare>
326 ReadyInstQueue;
327
328 /** List of ready instructions, per op class. They are separated by op
329 * class to allow for easy mapping to FUs.
330 */
331 ReadyInstQueue readyInsts[Num_OpClasses];
332
333 /** List of non-speculative instructions that will be scheduled
334 * once the IQ gets a signal from commit. While it's redundant to
335 * have the key be a part of the value (the sequence number is stored
336 * inside of DynInst), when these instructions are woken up only
337 * the sequence number will be available. Thus it is most efficient to be
338 * able to search by the sequence number alone.
339 */
340 std::map<InstSeqNum, DynInstPtr> nonSpecInsts;
341
342 typedef typename std::map<InstSeqNum, DynInstPtr>::iterator NonSpecMapIt;
343
344 /** Entry for the list age ordering by op class. */
345 struct ListOrderEntry {
346 OpClass queueType;
347 InstSeqNum oldestInst;
348 };
349
350 /** List that contains the age order of the oldest instruction of each
351 * ready queue. Used to select the oldest instruction available
352 * among op classes.
353 * @todo: Might be better to just move these entries around instead
354 * of creating new ones every time the position changes due to an
355 * instruction issuing. Not sure std::list supports this.
356 */
357 std::list<ListOrderEntry> listOrder;
358
359 typedef typename std::list<ListOrderEntry>::iterator ListOrderIt;
360
361 /** Tracks if each ready queue is on the age order list. */
362 bool queueOnList[Num_OpClasses];
363
364 /** Iterators of each ready queue. Points to their spot in the age order
365 * list.
366 */
367 ListOrderIt readyIt[Num_OpClasses];
368
369 /** Add an op class to the age order list. */
370 void addToOrderList(OpClass op_class);
371
372 /**
373 * Called when the oldest instruction has been removed from a ready queue;
374 * this places that ready queue into the proper spot in the age order list.
375 */
376 void moveToYoungerInst(ListOrderIt age_order_it);
377
378 DependencyGraph<DynInstPtr> dependGraph;
379
380 //////////////////////////////////////
381 // Various parameters
382 //////////////////////////////////////
383
384 /** IQ Resource Sharing Policy */
385 enum IQPolicy {
386 Dynamic,
387 Partitioned,
388 Threshold
389 };
390
391 /** IQ sharing policy for SMT. */
392 IQPolicy iqPolicy;
393
394 /** Number of Total Threads*/
395 ThreadID numThreads;
396
397 /** Pointer to list of active threads. */
398 std::list<ThreadID> *activeThreads;
399
400 /** Per Thread IQ count */
401 unsigned count[Impl::MaxThreads];
402
403 /** Max IQ Entries Per Thread */
404 unsigned maxEntries[Impl::MaxThreads];
405
406 /** Number of free IQ entries left. */
407 unsigned freeEntries;
408
409 /** The number of entries in the instruction queue. */
410 unsigned numEntries;
411
412 /** The total number of instructions that can be issued in one cycle. */
413 unsigned totalWidth;
414
415 /** The number of physical registers in the CPU. */
416 unsigned numPhysRegs;
417
|
418 /** Delay between commit stage and the IQ.
419 * @todo: Make there be a distinction between the delays within IEW.
420 */
421 Cycles commitToIEWDelay;
422
423 /** The sequence number of the squashed instruction. */
424 InstSeqNum squashedSeqNum[Impl::MaxThreads];
425
426 /** A cache of the recently woken registers. It is 1 if the register
427 * has been woken up recently, and 0 if the register has been added
428 * to the dependency graph and has not yet received its value. It
429 * is basically a secondary scoreboard, and should pretty much mirror
430 * the scoreboard that exists in the rename map.
431 */
432 std::vector<bool> regScoreboard;
433
434 /** Adds an instruction to the dependency graph, as a consumer. */
435 bool addToDependents(DynInstPtr &new_inst);
436
437 /** Adds an instruction to the dependency graph, as a producer. */
438 void addToProducers(DynInstPtr &new_inst);
439
440 /** Moves an instruction to the ready queue if it is ready. */
441 void addIfReady(DynInstPtr &inst);
442
443 /** Debugging function to count how many entries are in the IQ. It does
444 * a linear walk through the instructions, so do not call this function
445 * during normal execution.
446 */
447 int countInsts();
448
449 /** Debugging function to dump all the list sizes, as well as print
450 * out the list of nonspeculative instructions. Should not be used
451 * in any other capacity, but it has no harmful sideaffects.
452 */
453 void dumpLists();
454
455 /** Debugging function to dump out all instructions that are in the
456 * IQ.
457 */
458 void dumpInsts();
459
460 /** Stat for number of instructions added. */
461 Stats::Scalar iqInstsAdded;
462 /** Stat for number of non-speculative instructions added. */
463 Stats::Scalar iqNonSpecInstsAdded;
464
465 Stats::Scalar iqInstsIssued;
466 /** Stat for number of integer instructions issued. */
467 Stats::Scalar iqIntInstsIssued;
468 /** Stat for number of floating point instructions issued. */
469 Stats::Scalar iqFloatInstsIssued;
470 /** Stat for number of branch instructions issued. */
471 Stats::Scalar iqBranchInstsIssued;
472 /** Stat for number of memory instructions issued. */
473 Stats::Scalar iqMemInstsIssued;
474 /** Stat for number of miscellaneous instructions issued. */
475 Stats::Scalar iqMiscInstsIssued;
476 /** Stat for number of squashed instructions that were ready to issue. */
477 Stats::Scalar iqSquashedInstsIssued;
478 /** Stat for number of squashed instructions examined when squashing. */
479 Stats::Scalar iqSquashedInstsExamined;
480 /** Stat for number of squashed instruction operands examined when
481 * squashing.
482 */
483 Stats::Scalar iqSquashedOperandsExamined;
484 /** Stat for number of non-speculative instructions removed due to a squash.
485 */
486 Stats::Scalar iqSquashedNonSpecRemoved;
487 // Also include number of instructions rescheduled and replayed.
488
489 /** Distribution of number of instructions in the queue.
490 * @todo: Need to create struct to track the entry time for each
491 * instruction. */
492// Stats::VectorDistribution queueResDist;
493 /** Distribution of the number of instructions issued. */
494 Stats::Distribution numIssuedDist;
495 /** Distribution of the cycles it takes to issue an instruction.
496 * @todo: Need to create struct to track the ready time for each
497 * instruction. */
498// Stats::VectorDistribution issueDelayDist;
499
500 /** Number of times an instruction could not be issued because a
501 * FU was busy.
502 */
503 Stats::Vector statFuBusy;
504// Stats::Vector dist_unissued;
505 /** Stat for total number issued for each instruction type. */
506 Stats::Vector2d statIssuedInstType;
507
508 /** Number of instructions issued per cycle. */
509 Stats::Formula issueRate;
510
511 /** Number of times the FU was busy. */
512 Stats::Vector fuBusy;
513 /** Number of times the FU was busy per instruction issued. */
514 Stats::Formula fuBusyRate;
515 public:
516 Stats::Scalar intInstQueueReads;
517 Stats::Scalar intInstQueueWrites;
518 Stats::Scalar intInstQueueWakeupAccesses;
519 Stats::Scalar fpInstQueueReads;
520 Stats::Scalar fpInstQueueWrites;
521 Stats::Scalar fpInstQueueWakeupQccesses;
522
523 Stats::Scalar intAluAccesses;
524 Stats::Scalar fpAluAccesses;
525};
526
527#endif //__CPU_O3_INST_QUEUE_HH__
|