inst_queue_impl.hh revision 12106:7784fac1b159
1/*
2 * Copyright (c) 2011-2014 ARM Limited
3 * Copyright (c) 2013 Advanced Micro Devices, Inc.
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 *          Korey Sewell
43 */
44
45#ifndef __CPU_O3_INST_QUEUE_IMPL_HH__
46#define __CPU_O3_INST_QUEUE_IMPL_HH__
47
48#include <limits>
49#include <vector>
50
51#include "cpu/o3/fu_pool.hh"
52#include "cpu/o3/inst_queue.hh"
53#include "debug/IQ.hh"
54#include "enums/OpClass.hh"
55#include "params/DerivO3CPU.hh"
56#include "sim/core.hh"
57
58// clang complains about std::set being overloaded with Packet::set if
59// we open up the entire namespace std
60using std::list;
61
62template <class Impl>
63InstructionQueue<Impl>::FUCompletion::FUCompletion(DynInstPtr &_inst,
64    int fu_idx, InstructionQueue<Impl> *iq_ptr)
65    : Event(Stat_Event_Pri, AutoDelete),
66      inst(_inst), fuIdx(fu_idx), iqPtr(iq_ptr), freeFU(false)
67{
68}
69
70template <class Impl>
71void
72InstructionQueue<Impl>::FUCompletion::process()
73{
74    iqPtr->processFUCompletion(inst, freeFU ? fuIdx : -1);
75    inst = NULL;
76}
77
78
79template <class Impl>
80const char *
81InstructionQueue<Impl>::FUCompletion::description() const
82{
83    return "Functional unit completion";
84}
85
86template <class Impl>
87InstructionQueue<Impl>::InstructionQueue(O3CPU *cpu_ptr, IEW *iew_ptr,
88                                         DerivO3CPUParams *params)
89    : cpu(cpu_ptr),
90      iewStage(iew_ptr),
91      fuPool(params->fuPool),
92      numEntries(params->numIQEntries),
93      totalWidth(params->issueWidth),
94      commitToIEWDelay(params->commitToIEWDelay)
95{
96    assert(fuPool);
97
98    numThreads = params->numThreads;
99
100    // Set the number of total physical registers
101    numPhysRegs = params->numPhysIntRegs + params->numPhysFloatRegs +
102        params->numPhysCCRegs;
103
104    //Create an entry for each physical register within the
105    //dependency graph.
106    dependGraph.resize(numPhysRegs);
107
108    // Resize the register scoreboard.
109    regScoreboard.resize(numPhysRegs);
110
111    //Initialize Mem Dependence Units
112    for (ThreadID tid = 0; tid < numThreads; tid++) {
113        memDepUnit[tid].init(params, tid);
114        memDepUnit[tid].setIQ(this);
115    }
116
117    resetState();
118
119    std::string policy = params->smtIQPolicy;
120
121    //Convert string to lowercase
122    std::transform(policy.begin(), policy.end(), policy.begin(),
123                   (int(*)(int)) tolower);
124
125    //Figure out resource sharing policy
126    if (policy == "dynamic") {
127        iqPolicy = Dynamic;
128
129        //Set Max Entries to Total ROB Capacity
130        for (ThreadID tid = 0; tid < numThreads; tid++) {
131            maxEntries[tid] = numEntries;
132        }
133
134    } else if (policy == "partitioned") {
135        iqPolicy = Partitioned;
136
137        //@todo:make work if part_amt doesnt divide evenly.
138        int part_amt = numEntries / numThreads;
139
140        //Divide ROB up evenly
141        for (ThreadID tid = 0; tid < numThreads; tid++) {
142            maxEntries[tid] = part_amt;
143        }
144
145        DPRINTF(IQ, "IQ sharing policy set to Partitioned:"
146                "%i entries per thread.\n",part_amt);
147    } else if (policy == "threshold") {
148        iqPolicy = Threshold;
149
150        double threshold =  (double)params->smtIQThreshold / 100;
151
152        int thresholdIQ = (int)((double)threshold * numEntries);
153
154        //Divide up by threshold amount
155        for (ThreadID tid = 0; tid < numThreads; tid++) {
156            maxEntries[tid] = thresholdIQ;
157        }
158
159        DPRINTF(IQ, "IQ sharing policy set to Threshold:"
160                "%i entries per thread.\n",thresholdIQ);
161   } else {
162       assert(0 && "Invalid IQ Sharing Policy.Options Are:{Dynamic,"
163              "Partitioned, Threshold}");
164   }
165}
166
167template <class Impl>
168InstructionQueue<Impl>::~InstructionQueue()
169{
170    dependGraph.reset();
171#ifdef DEBUG
172    cprintf("Nodes traversed: %i, removed: %i\n",
173            dependGraph.nodesTraversed, dependGraph.nodesRemoved);
174#endif
175}
176
177template <class Impl>
178std::string
179InstructionQueue<Impl>::name() const
180{
181    return cpu->name() + ".iq";
182}
183
184template <class Impl>
185void
186InstructionQueue<Impl>::regStats()
187{
188    using namespace Stats;
189    iqInstsAdded
190        .name(name() + ".iqInstsAdded")
191        .desc("Number of instructions added to the IQ (excludes non-spec)")
192        .prereq(iqInstsAdded);
193
194    iqNonSpecInstsAdded
195        .name(name() + ".iqNonSpecInstsAdded")
196        .desc("Number of non-speculative instructions added to the IQ")
197        .prereq(iqNonSpecInstsAdded);
198
199    iqInstsIssued
200        .name(name() + ".iqInstsIssued")
201        .desc("Number of instructions issued")
202        .prereq(iqInstsIssued);
203
204    iqIntInstsIssued
205        .name(name() + ".iqIntInstsIssued")
206        .desc("Number of integer instructions issued")
207        .prereq(iqIntInstsIssued);
208
209    iqFloatInstsIssued
210        .name(name() + ".iqFloatInstsIssued")
211        .desc("Number of float instructions issued")
212        .prereq(iqFloatInstsIssued);
213
214    iqBranchInstsIssued
215        .name(name() + ".iqBranchInstsIssued")
216        .desc("Number of branch instructions issued")
217        .prereq(iqBranchInstsIssued);
218
219    iqMemInstsIssued
220        .name(name() + ".iqMemInstsIssued")
221        .desc("Number of memory instructions issued")
222        .prereq(iqMemInstsIssued);
223
224    iqMiscInstsIssued
225        .name(name() + ".iqMiscInstsIssued")
226        .desc("Number of miscellaneous instructions issued")
227        .prereq(iqMiscInstsIssued);
228
229    iqSquashedInstsIssued
230        .name(name() + ".iqSquashedInstsIssued")
231        .desc("Number of squashed instructions issued")
232        .prereq(iqSquashedInstsIssued);
233
234    iqSquashedInstsExamined
235        .name(name() + ".iqSquashedInstsExamined")
236        .desc("Number of squashed instructions iterated over during squash;"
237              " mainly for profiling")
238        .prereq(iqSquashedInstsExamined);
239
240    iqSquashedOperandsExamined
241        .name(name() + ".iqSquashedOperandsExamined")
242        .desc("Number of squashed operands that are examined and possibly "
243              "removed from graph")
244        .prereq(iqSquashedOperandsExamined);
245
246    iqSquashedNonSpecRemoved
247        .name(name() + ".iqSquashedNonSpecRemoved")
248        .desc("Number of squashed non-spec instructions that were removed")
249        .prereq(iqSquashedNonSpecRemoved);
250/*
251    queueResDist
252        .init(Num_OpClasses, 0, 99, 2)
253        .name(name() + ".IQ:residence:")
254        .desc("cycles from dispatch to issue")
255        .flags(total | pdf | cdf )
256        ;
257    for (int i = 0; i < Num_OpClasses; ++i) {
258        queueResDist.subname(i, opClassStrings[i]);
259    }
260*/
261    numIssuedDist
262        .init(0,totalWidth,1)
263        .name(name() + ".issued_per_cycle")
264        .desc("Number of insts issued each cycle")
265        .flags(pdf)
266        ;
267/*
268    dist_unissued
269        .init(Num_OpClasses+2)
270        .name(name() + ".unissued_cause")
271        .desc("Reason ready instruction not issued")
272        .flags(pdf | dist)
273        ;
274    for (int i=0; i < (Num_OpClasses + 2); ++i) {
275        dist_unissued.subname(i, unissued_names[i]);
276    }
277*/
278    statIssuedInstType
279        .init(numThreads,Enums::Num_OpClass)
280        .name(name() + ".FU_type")
281        .desc("Type of FU issued")
282        .flags(total | pdf | dist)
283        ;
284    statIssuedInstType.ysubnames(Enums::OpClassStrings);
285
286    //
287    //  How long did instructions for a particular FU type wait prior to issue
288    //
289/*
290    issueDelayDist
291        .init(Num_OpClasses,0,99,2)
292        .name(name() + ".")
293        .desc("cycles from operands ready to issue")
294        .flags(pdf | cdf)
295        ;
296
297    for (int i=0; i<Num_OpClasses; ++i) {
298        std::stringstream subname;
299        subname << opClassStrings[i] << "_delay";
300        issueDelayDist.subname(i, subname.str());
301    }
302*/
303    issueRate
304        .name(name() + ".rate")
305        .desc("Inst issue rate")
306        .flags(total)
307        ;
308    issueRate = iqInstsIssued / cpu->numCycles;
309
310    statFuBusy
311        .init(Num_OpClasses)
312        .name(name() + ".fu_full")
313        .desc("attempts to use FU when none available")
314        .flags(pdf | dist)
315        ;
316    for (int i=0; i < Num_OpClasses; ++i) {
317        statFuBusy.subname(i, Enums::OpClassStrings[i]);
318    }
319
320    fuBusy
321        .init(numThreads)
322        .name(name() + ".fu_busy_cnt")
323        .desc("FU busy when requested")
324        .flags(total)
325        ;
326
327    fuBusyRate
328        .name(name() + ".fu_busy_rate")
329        .desc("FU busy rate (busy events/executed inst)")
330        .flags(total)
331        ;
332    fuBusyRate = fuBusy / iqInstsIssued;
333
334    for (ThreadID tid = 0; tid < numThreads; tid++) {
335        // Tell mem dependence unit to reg stats as well.
336        memDepUnit[tid].regStats();
337    }
338
339    intInstQueueReads
340        .name(name() + ".int_inst_queue_reads")
341        .desc("Number of integer instruction queue reads")
342        .flags(total);
343
344    intInstQueueWrites
345        .name(name() + ".int_inst_queue_writes")
346        .desc("Number of integer instruction queue writes")
347        .flags(total);
348
349    intInstQueueWakeupAccesses
350        .name(name() + ".int_inst_queue_wakeup_accesses")
351        .desc("Number of integer instruction queue wakeup accesses")
352        .flags(total);
353
354    fpInstQueueReads
355        .name(name() + ".fp_inst_queue_reads")
356        .desc("Number of floating instruction queue reads")
357        .flags(total);
358
359    fpInstQueueWrites
360        .name(name() + ".fp_inst_queue_writes")
361        .desc("Number of floating instruction queue writes")
362        .flags(total);
363
364    fpInstQueueWakeupQccesses
365        .name(name() + ".fp_inst_queue_wakeup_accesses")
366        .desc("Number of floating instruction queue wakeup accesses")
367        .flags(total);
368
369    intAluAccesses
370        .name(name() + ".int_alu_accesses")
371        .desc("Number of integer alu accesses")
372        .flags(total);
373
374    fpAluAccesses
375        .name(name() + ".fp_alu_accesses")
376        .desc("Number of floating point alu accesses")
377        .flags(total);
378
379}
380
381template <class Impl>
382void
383InstructionQueue<Impl>::resetState()
384{
385    //Initialize thread IQ counts
386    for (ThreadID tid = 0; tid <numThreads; tid++) {
387        count[tid] = 0;
388        instList[tid].clear();
389    }
390
391    // Initialize the number of free IQ entries.
392    freeEntries = numEntries;
393
394    // Note that in actuality, the registers corresponding to the logical
395    // registers start off as ready.  However this doesn't matter for the
396    // IQ as the instruction should have been correctly told if those
397    // registers are ready in rename.  Thus it can all be initialized as
398    // unready.
399    for (int i = 0; i < numPhysRegs; ++i) {
400        regScoreboard[i] = false;
401    }
402
403    for (ThreadID tid = 0; tid < numThreads; ++tid) {
404        squashedSeqNum[tid] = 0;
405    }
406
407    for (int i = 0; i < Num_OpClasses; ++i) {
408        while (!readyInsts[i].empty())
409            readyInsts[i].pop();
410        queueOnList[i] = false;
411        readyIt[i] = listOrder.end();
412    }
413    nonSpecInsts.clear();
414    listOrder.clear();
415    deferredMemInsts.clear();
416    blockedMemInsts.clear();
417    retryMemInsts.clear();
418    wbOutstanding = 0;
419}
420
421template <class Impl>
422void
423InstructionQueue<Impl>::setActiveThreads(list<ThreadID> *at_ptr)
424{
425    activeThreads = at_ptr;
426}
427
428template <class Impl>
429void
430InstructionQueue<Impl>::setIssueToExecuteQueue(TimeBuffer<IssueStruct> *i2e_ptr)
431{
432      issueToExecuteQueue = i2e_ptr;
433}
434
435template <class Impl>
436void
437InstructionQueue<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr)
438{
439    timeBuffer = tb_ptr;
440
441    fromCommit = timeBuffer->getWire(-commitToIEWDelay);
442}
443
444template <class Impl>
445bool
446InstructionQueue<Impl>::isDrained() const
447{
448    bool drained = dependGraph.empty() &&
449                   instsToExecute.empty() &&
450                   wbOutstanding == 0;
451    for (ThreadID tid = 0; tid < numThreads; ++tid)
452        drained = drained && memDepUnit[tid].isDrained();
453
454    return drained;
455}
456
457template <class Impl>
458void
459InstructionQueue<Impl>::drainSanityCheck() const
460{
461    assert(dependGraph.empty());
462    assert(instsToExecute.empty());
463    for (ThreadID tid = 0; tid < numThreads; ++tid)
464        memDepUnit[tid].drainSanityCheck();
465}
466
467template <class Impl>
468void
469InstructionQueue<Impl>::takeOverFrom()
470{
471    resetState();
472}
473
474template <class Impl>
475int
476InstructionQueue<Impl>::entryAmount(ThreadID num_threads)
477{
478    if (iqPolicy == Partitioned) {
479        return numEntries / num_threads;
480    } else {
481        return 0;
482    }
483}
484
485
486template <class Impl>
487void
488InstructionQueue<Impl>::resetEntries()
489{
490    if (iqPolicy != Dynamic || numThreads > 1) {
491        int active_threads = activeThreads->size();
492
493        list<ThreadID>::iterator threads = activeThreads->begin();
494        list<ThreadID>::iterator end = activeThreads->end();
495
496        while (threads != end) {
497            ThreadID tid = *threads++;
498
499            if (iqPolicy == Partitioned) {
500                maxEntries[tid] = numEntries / active_threads;
501            } else if (iqPolicy == Threshold && active_threads == 1) {
502                maxEntries[tid] = numEntries;
503            }
504        }
505    }
506}
507
508template <class Impl>
509unsigned
510InstructionQueue<Impl>::numFreeEntries()
511{
512    return freeEntries;
513}
514
515template <class Impl>
516unsigned
517InstructionQueue<Impl>::numFreeEntries(ThreadID tid)
518{
519    return maxEntries[tid] - count[tid];
520}
521
522// Might want to do something more complex if it knows how many instructions
523// will be issued this cycle.
524template <class Impl>
525bool
526InstructionQueue<Impl>::isFull()
527{
528    if (freeEntries == 0) {
529        return(true);
530    } else {
531        return(false);
532    }
533}
534
535template <class Impl>
536bool
537InstructionQueue<Impl>::isFull(ThreadID tid)
538{
539    if (numFreeEntries(tid) == 0) {
540        return(true);
541    } else {
542        return(false);
543    }
544}
545
546template <class Impl>
547bool
548InstructionQueue<Impl>::hasReadyInsts()
549{
550    if (!listOrder.empty()) {
551        return true;
552    }
553
554    for (int i = 0; i < Num_OpClasses; ++i) {
555        if (!readyInsts[i].empty()) {
556            return true;
557        }
558    }
559
560    return false;
561}
562
563template <class Impl>
564void
565InstructionQueue<Impl>::insert(DynInstPtr &new_inst)
566{
567    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
568    // Make sure the instruction is valid
569    assert(new_inst);
570
571    DPRINTF(IQ, "Adding instruction [sn:%lli] PC %s to the IQ.\n",
572            new_inst->seqNum, new_inst->pcState());
573
574    assert(freeEntries != 0);
575
576    instList[new_inst->threadNumber].push_back(new_inst);
577
578    --freeEntries;
579
580    new_inst->setInIQ();
581
582    // Look through its source registers (physical regs), and mark any
583    // dependencies.
584    addToDependents(new_inst);
585
586    // Have this instruction set itself as the producer of its destination
587    // register(s).
588    addToProducers(new_inst);
589
590    if (new_inst->isMemRef()) {
591        memDepUnit[new_inst->threadNumber].insert(new_inst);
592    } else {
593        addIfReady(new_inst);
594    }
595
596    ++iqInstsAdded;
597
598    count[new_inst->threadNumber]++;
599
600    assert(freeEntries == (numEntries - countInsts()));
601}
602
603template <class Impl>
604void
605InstructionQueue<Impl>::insertNonSpec(DynInstPtr &new_inst)
606{
607    // @todo: Clean up this code; can do it by setting inst as unable
608    // to issue, then calling normal insert on the inst.
609    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
610
611    assert(new_inst);
612
613    nonSpecInsts[new_inst->seqNum] = new_inst;
614
615    DPRINTF(IQ, "Adding non-speculative instruction [sn:%lli] PC %s "
616            "to the IQ.\n",
617            new_inst->seqNum, new_inst->pcState());
618
619    assert(freeEntries != 0);
620
621    instList[new_inst->threadNumber].push_back(new_inst);
622
623    --freeEntries;
624
625    new_inst->setInIQ();
626
627    // Have this instruction set itself as the producer of its destination
628    // register(s).
629    addToProducers(new_inst);
630
631    // If it's a memory instruction, add it to the memory dependency
632    // unit.
633    if (new_inst->isMemRef()) {
634        memDepUnit[new_inst->threadNumber].insertNonSpec(new_inst);
635    }
636
637    ++iqNonSpecInstsAdded;
638
639    count[new_inst->threadNumber]++;
640
641    assert(freeEntries == (numEntries - countInsts()));
642}
643
644template <class Impl>
645void
646InstructionQueue<Impl>::insertBarrier(DynInstPtr &barr_inst)
647{
648    memDepUnit[barr_inst->threadNumber].insertBarrier(barr_inst);
649
650    insertNonSpec(barr_inst);
651}
652
653template <class Impl>
654typename Impl::DynInstPtr
655InstructionQueue<Impl>::getInstToExecute()
656{
657    assert(!instsToExecute.empty());
658    DynInstPtr inst = instsToExecute.front();
659    instsToExecute.pop_front();
660    if (inst->isFloating()){
661        fpInstQueueReads++;
662    } else {
663        intInstQueueReads++;
664    }
665    return inst;
666}
667
668template <class Impl>
669void
670InstructionQueue<Impl>::addToOrderList(OpClass op_class)
671{
672    assert(!readyInsts[op_class].empty());
673
674    ListOrderEntry queue_entry;
675
676    queue_entry.queueType = op_class;
677
678    queue_entry.oldestInst = readyInsts[op_class].top()->seqNum;
679
680    ListOrderIt list_it = listOrder.begin();
681    ListOrderIt list_end_it = listOrder.end();
682
683    while (list_it != list_end_it) {
684        if ((*list_it).oldestInst > queue_entry.oldestInst) {
685            break;
686        }
687
688        list_it++;
689    }
690
691    readyIt[op_class] = listOrder.insert(list_it, queue_entry);
692    queueOnList[op_class] = true;
693}
694
695template <class Impl>
696void
697InstructionQueue<Impl>::moveToYoungerInst(ListOrderIt list_order_it)
698{
699    // Get iterator of next item on the list
700    // Delete the original iterator
701    // Determine if the next item is either the end of the list or younger
702    // than the new instruction.  If so, then add in a new iterator right here.
703    // If not, then move along.
704    ListOrderEntry queue_entry;
705    OpClass op_class = (*list_order_it).queueType;
706    ListOrderIt next_it = list_order_it;
707
708    ++next_it;
709
710    queue_entry.queueType = op_class;
711    queue_entry.oldestInst = readyInsts[op_class].top()->seqNum;
712
713    while (next_it != listOrder.end() &&
714           (*next_it).oldestInst < queue_entry.oldestInst) {
715        ++next_it;
716    }
717
718    readyIt[op_class] = listOrder.insert(next_it, queue_entry);
719}
720
721template <class Impl>
722void
723InstructionQueue<Impl>::processFUCompletion(DynInstPtr &inst, int fu_idx)
724{
725    DPRINTF(IQ, "Processing FU completion [sn:%lli]\n", inst->seqNum);
726    assert(!cpu->switchedOut());
727    // The CPU could have been sleeping until this op completed (*extremely*
728    // long latency op).  Wake it if it was.  This may be overkill.
729   --wbOutstanding;
730    iewStage->wakeCPU();
731
732    if (fu_idx > -1)
733        fuPool->freeUnitNextCycle(fu_idx);
734
735    // @todo: Ensure that these FU Completions happen at the beginning
736    // of a cycle, otherwise they could add too many instructions to
737    // the queue.
738    issueToExecuteQueue->access(-1)->size++;
739    instsToExecute.push_back(inst);
740}
741
742// @todo: Figure out a better way to remove the squashed items from the
743// lists.  Checking the top item of each list to see if it's squashed
744// wastes time and forces jumps.
745template <class Impl>
746void
747InstructionQueue<Impl>::scheduleReadyInsts()
748{
749    DPRINTF(IQ, "Attempting to schedule ready instructions from "
750            "the IQ.\n");
751
752    IssueStruct *i2e_info = issueToExecuteQueue->access(0);
753
754    DynInstPtr mem_inst;
755    while (mem_inst = getDeferredMemInstToExecute()) {
756        addReadyMemInst(mem_inst);
757    }
758
759    // See if any cache blocked instructions are able to be executed
760    while (mem_inst = getBlockedMemInstToExecute()) {
761        addReadyMemInst(mem_inst);
762    }
763
764    // Have iterator to head of the list
765    // While I haven't exceeded bandwidth or reached the end of the list,
766    // Try to get a FU that can do what this op needs.
767    // If successful, change the oldestInst to the new top of the list, put
768    // the queue in the proper place in the list.
769    // Increment the iterator.
770    // This will avoid trying to schedule a certain op class if there are no
771    // FUs that handle it.
772    int total_issued = 0;
773    ListOrderIt order_it = listOrder.begin();
774    ListOrderIt order_end_it = listOrder.end();
775
776    while (total_issued < totalWidth && order_it != order_end_it) {
777        OpClass op_class = (*order_it).queueType;
778
779        assert(!readyInsts[op_class].empty());
780
781        DynInstPtr issuing_inst = readyInsts[op_class].top();
782
783        issuing_inst->isFloating() ? fpInstQueueReads++ : intInstQueueReads++;
784
785        assert(issuing_inst->seqNum == (*order_it).oldestInst);
786
787        if (issuing_inst->isSquashed()) {
788            readyInsts[op_class].pop();
789
790            if (!readyInsts[op_class].empty()) {
791                moveToYoungerInst(order_it);
792            } else {
793                readyIt[op_class] = listOrder.end();
794                queueOnList[op_class] = false;
795            }
796
797            listOrder.erase(order_it++);
798
799            ++iqSquashedInstsIssued;
800
801            continue;
802        }
803
804        int idx = FUPool::NoCapableFU;
805        Cycles op_latency = Cycles(1);
806        ThreadID tid = issuing_inst->threadNumber;
807
808        if (op_class != No_OpClass) {
809            idx = fuPool->getUnit(op_class);
810            issuing_inst->isFloating() ? fpAluAccesses++ : intAluAccesses++;
811            if (idx > FUPool::NoFreeFU) {
812                op_latency = fuPool->getOpLatency(op_class);
813            }
814        }
815
816        // If we have an instruction that doesn't require a FU, or a
817        // valid FU, then schedule for execution.
818        if (idx != FUPool::NoFreeFU) {
819            if (op_latency == Cycles(1)) {
820                i2e_info->size++;
821                instsToExecute.push_back(issuing_inst);
822
823                // Add the FU onto the list of FU's to be freed next
824                // cycle if we used one.
825                if (idx >= 0)
826                    fuPool->freeUnitNextCycle(idx);
827            } else {
828                bool pipelined = fuPool->isPipelined(op_class);
829                // Generate completion event for the FU
830                ++wbOutstanding;
831                FUCompletion *execution = new FUCompletion(issuing_inst,
832                                                           idx, this);
833
834                cpu->schedule(execution,
835                              cpu->clockEdge(Cycles(op_latency - 1)));
836
837                if (!pipelined) {
838                    // If FU isn't pipelined, then it must be freed
839                    // upon the execution completing.
840                    execution->setFreeFU();
841                } else {
842                    // Add the FU onto the list of FU's to be freed next cycle.
843                    fuPool->freeUnitNextCycle(idx);
844                }
845            }
846
847            DPRINTF(IQ, "Thread %i: Issuing instruction PC %s "
848                    "[sn:%lli]\n",
849                    tid, issuing_inst->pcState(),
850                    issuing_inst->seqNum);
851
852            readyInsts[op_class].pop();
853
854            if (!readyInsts[op_class].empty()) {
855                moveToYoungerInst(order_it);
856            } else {
857                readyIt[op_class] = listOrder.end();
858                queueOnList[op_class] = false;
859            }
860
861            issuing_inst->setIssued();
862            ++total_issued;
863
864#if TRACING_ON
865            issuing_inst->issueTick = curTick() - issuing_inst->fetchTick;
866#endif
867
868            if (!issuing_inst->isMemRef()) {
869                // Memory instructions can not be freed from the IQ until they
870                // complete.
871                ++freeEntries;
872                count[tid]--;
873                issuing_inst->clearInIQ();
874            } else {
875                memDepUnit[tid].issue(issuing_inst);
876            }
877
878            listOrder.erase(order_it++);
879            statIssuedInstType[tid][op_class]++;
880        } else {
881            statFuBusy[op_class]++;
882            fuBusy[tid]++;
883            ++order_it;
884        }
885    }
886
887    numIssuedDist.sample(total_issued);
888    iqInstsIssued+= total_issued;
889
890    // If we issued any instructions, tell the CPU we had activity.
891    // @todo If the way deferred memory instructions are handeled due to
892    // translation changes then the deferredMemInsts condition should be removed
893    // from the code below.
894    if (total_issued || !retryMemInsts.empty() || !deferredMemInsts.empty()) {
895        cpu->activityThisCycle();
896    } else {
897        DPRINTF(IQ, "Not able to schedule any instructions.\n");
898    }
899}
900
901template <class Impl>
902void
903InstructionQueue<Impl>::scheduleNonSpec(const InstSeqNum &inst)
904{
905    DPRINTF(IQ, "Marking nonspeculative instruction [sn:%lli] as ready "
906            "to execute.\n", inst);
907
908    NonSpecMapIt inst_it = nonSpecInsts.find(inst);
909
910    assert(inst_it != nonSpecInsts.end());
911
912    ThreadID tid = (*inst_it).second->threadNumber;
913
914    (*inst_it).second->setAtCommit();
915
916    (*inst_it).second->setCanIssue();
917
918    if (!(*inst_it).second->isMemRef()) {
919        addIfReady((*inst_it).second);
920    } else {
921        memDepUnit[tid].nonSpecInstReady((*inst_it).second);
922    }
923
924    (*inst_it).second = NULL;
925
926    nonSpecInsts.erase(inst_it);
927}
928
929template <class Impl>
930void
931InstructionQueue<Impl>::commit(const InstSeqNum &inst, ThreadID tid)
932{
933    DPRINTF(IQ, "[tid:%i]: Committing instructions older than [sn:%i]\n",
934            tid,inst);
935
936    ListIt iq_it = instList[tid].begin();
937
938    while (iq_it != instList[tid].end() &&
939           (*iq_it)->seqNum <= inst) {
940        ++iq_it;
941        instList[tid].pop_front();
942    }
943
944    assert(freeEntries == (numEntries - countInsts()));
945}
946
947template <class Impl>
948int
949InstructionQueue<Impl>::wakeDependents(DynInstPtr &completed_inst)
950{
951    int dependents = 0;
952
953    // The instruction queue here takes care of both floating and int ops
954    if (completed_inst->isFloating()) {
955        fpInstQueueWakeupQccesses++;
956    } else {
957        intInstQueueWakeupAccesses++;
958    }
959
960    DPRINTF(IQ, "Waking dependents of completed instruction.\n");
961
962    assert(!completed_inst->isSquashed());
963
964    // Tell the memory dependence unit to wake any dependents on this
965    // instruction if it is a memory instruction.  Also complete the memory
966    // instruction at this point since we know it executed without issues.
967    // @todo: Might want to rename "completeMemInst" to something that
968    // indicates that it won't need to be replayed, and call this
969    // earlier.  Might not be a big deal.
970    if (completed_inst->isMemRef()) {
971        memDepUnit[completed_inst->threadNumber].wakeDependents(completed_inst);
972        completeMemInst(completed_inst);
973    } else if (completed_inst->isMemBarrier() ||
974               completed_inst->isWriteBarrier()) {
975        memDepUnit[completed_inst->threadNumber].completeBarrier(completed_inst);
976    }
977
978    for (int dest_reg_idx = 0;
979         dest_reg_idx < completed_inst->numDestRegs();
980         dest_reg_idx++)
981    {
982        PhysRegIdPtr dest_reg =
983            completed_inst->renamedDestRegIdx(dest_reg_idx);
984
985        // Special case of uniq or control registers.  They are not
986        // handled by the IQ and thus have no dependency graph entry.
987        if (dest_reg->isFixedMapping()) {
988            DPRINTF(IQ, "Reg %d [%s] is part of a fix mapping, skipping\n",
989                    dest_reg->index(), dest_reg->className());
990            continue;
991        }
992
993        DPRINTF(IQ, "Waking any dependents on register %i (%s).\n",
994                dest_reg->index(),
995                dest_reg->className());
996
997        //Go through the dependency chain, marking the registers as
998        //ready within the waiting instructions.
999        DynInstPtr dep_inst = dependGraph.pop(dest_reg->flatIndex());
1000
1001        while (dep_inst) {
1002            DPRINTF(IQ, "Waking up a dependent instruction, [sn:%lli] "
1003                    "PC %s.\n", dep_inst->seqNum, dep_inst->pcState());
1004
1005            // Might want to give more information to the instruction
1006            // so that it knows which of its source registers is
1007            // ready.  However that would mean that the dependency
1008            // graph entries would need to hold the src_reg_idx.
1009            dep_inst->markSrcRegReady();
1010
1011            addIfReady(dep_inst);
1012
1013            dep_inst = dependGraph.pop(dest_reg->flatIndex());
1014
1015            ++dependents;
1016        }
1017
1018        // Reset the head node now that all of its dependents have
1019        // been woken up.
1020        assert(dependGraph.empty(dest_reg->flatIndex()));
1021        dependGraph.clearInst(dest_reg->flatIndex());
1022
1023        // Mark the scoreboard as having that register ready.
1024        regScoreboard[dest_reg->flatIndex()] = true;
1025    }
1026    return dependents;
1027}
1028
1029template <class Impl>
1030void
1031InstructionQueue<Impl>::addReadyMemInst(DynInstPtr &ready_inst)
1032{
1033    OpClass op_class = ready_inst->opClass();
1034
1035    readyInsts[op_class].push(ready_inst);
1036
1037    // Will need to reorder the list if either a queue is not on the list,
1038    // or it has an older instruction than last time.
1039    if (!queueOnList[op_class]) {
1040        addToOrderList(op_class);
1041    } else if (readyInsts[op_class].top()->seqNum  <
1042               (*readyIt[op_class]).oldestInst) {
1043        listOrder.erase(readyIt[op_class]);
1044        addToOrderList(op_class);
1045    }
1046
1047    DPRINTF(IQ, "Instruction is ready to issue, putting it onto "
1048            "the ready list, PC %s opclass:%i [sn:%lli].\n",
1049            ready_inst->pcState(), op_class, ready_inst->seqNum);
1050}
1051
1052template <class Impl>
1053void
1054InstructionQueue<Impl>::rescheduleMemInst(DynInstPtr &resched_inst)
1055{
1056    DPRINTF(IQ, "Rescheduling mem inst [sn:%lli]\n", resched_inst->seqNum);
1057
1058    // Reset DTB translation state
1059    resched_inst->translationStarted(false);
1060    resched_inst->translationCompleted(false);
1061
1062    resched_inst->clearCanIssue();
1063    memDepUnit[resched_inst->threadNumber].reschedule(resched_inst);
1064}
1065
1066template <class Impl>
1067void
1068InstructionQueue<Impl>::replayMemInst(DynInstPtr &replay_inst)
1069{
1070    memDepUnit[replay_inst->threadNumber].replay();
1071}
1072
1073template <class Impl>
1074void
1075InstructionQueue<Impl>::completeMemInst(DynInstPtr &completed_inst)
1076{
1077    ThreadID tid = completed_inst->threadNumber;
1078
1079    DPRINTF(IQ, "Completing mem instruction PC: %s [sn:%lli]\n",
1080            completed_inst->pcState(), completed_inst->seqNum);
1081
1082    ++freeEntries;
1083
1084    completed_inst->memOpDone(true);
1085
1086    memDepUnit[tid].completed(completed_inst);
1087    count[tid]--;
1088}
1089
1090template <class Impl>
1091void
1092InstructionQueue<Impl>::deferMemInst(DynInstPtr &deferred_inst)
1093{
1094    deferredMemInsts.push_back(deferred_inst);
1095}
1096
1097template <class Impl>
1098void
1099InstructionQueue<Impl>::blockMemInst(DynInstPtr &blocked_inst)
1100{
1101    blocked_inst->translationStarted(false);
1102    blocked_inst->translationCompleted(false);
1103
1104    blocked_inst->clearIssued();
1105    blocked_inst->clearCanIssue();
1106    blockedMemInsts.push_back(blocked_inst);
1107}
1108
1109template <class Impl>
1110void
1111InstructionQueue<Impl>::cacheUnblocked()
1112{
1113    retryMemInsts.splice(retryMemInsts.end(), blockedMemInsts);
1114    // Get the CPU ticking again
1115    cpu->wakeCPU();
1116}
1117
1118template <class Impl>
1119typename Impl::DynInstPtr
1120InstructionQueue<Impl>::getDeferredMemInstToExecute()
1121{
1122    for (ListIt it = deferredMemInsts.begin(); it != deferredMemInsts.end();
1123         ++it) {
1124        if ((*it)->translationCompleted() || (*it)->isSquashed()) {
1125            DynInstPtr mem_inst = *it;
1126            deferredMemInsts.erase(it);
1127            return mem_inst;
1128        }
1129    }
1130    return nullptr;
1131}
1132
1133template <class Impl>
1134typename Impl::DynInstPtr
1135InstructionQueue<Impl>::getBlockedMemInstToExecute()
1136{
1137    if (retryMemInsts.empty()) {
1138        return nullptr;
1139    } else {
1140        DynInstPtr mem_inst = retryMemInsts.front();
1141        retryMemInsts.pop_front();
1142        return mem_inst;
1143    }
1144}
1145
1146template <class Impl>
1147void
1148InstructionQueue<Impl>::violation(DynInstPtr &store,
1149                                  DynInstPtr &faulting_load)
1150{
1151    intInstQueueWrites++;
1152    memDepUnit[store->threadNumber].violation(store, faulting_load);
1153}
1154
1155template <class Impl>
1156void
1157InstructionQueue<Impl>::squash(ThreadID tid)
1158{
1159    DPRINTF(IQ, "[tid:%i]: Starting to squash instructions in "
1160            "the IQ.\n", tid);
1161
1162    // Read instruction sequence number of last instruction out of the
1163    // time buffer.
1164    squashedSeqNum[tid] = fromCommit->commitInfo[tid].doneSeqNum;
1165
1166    doSquash(tid);
1167
1168    // Also tell the memory dependence unit to squash.
1169    memDepUnit[tid].squash(squashedSeqNum[tid], tid);
1170}
1171
1172template <class Impl>
1173void
1174InstructionQueue<Impl>::doSquash(ThreadID tid)
1175{
1176    // Start at the tail.
1177    ListIt squash_it = instList[tid].end();
1178    --squash_it;
1179
1180    DPRINTF(IQ, "[tid:%i]: Squashing until sequence number %i!\n",
1181            tid, squashedSeqNum[tid]);
1182
1183    // Squash any instructions younger than the squashed sequence number
1184    // given.
1185    while (squash_it != instList[tid].end() &&
1186           (*squash_it)->seqNum > squashedSeqNum[tid]) {
1187
1188        DynInstPtr squashed_inst = (*squash_it);
1189        squashed_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
1190
1191        // Only handle the instruction if it actually is in the IQ and
1192        // hasn't already been squashed in the IQ.
1193        if (squashed_inst->threadNumber != tid ||
1194            squashed_inst->isSquashedInIQ()) {
1195            --squash_it;
1196            continue;
1197        }
1198
1199        if (!squashed_inst->isIssued() ||
1200            (squashed_inst->isMemRef() &&
1201             !squashed_inst->memOpDone())) {
1202
1203            DPRINTF(IQ, "[tid:%i]: Instruction [sn:%lli] PC %s squashed.\n",
1204                    tid, squashed_inst->seqNum, squashed_inst->pcState());
1205
1206            bool is_acq_rel = squashed_inst->isMemBarrier() &&
1207                         (squashed_inst->isLoad() ||
1208                           (squashed_inst->isStore() &&
1209                             !squashed_inst->isStoreConditional()));
1210
1211            // Remove the instruction from the dependency list.
1212            if (is_acq_rel ||
1213                (!squashed_inst->isNonSpeculative() &&
1214                 !squashed_inst->isStoreConditional() &&
1215                 !squashed_inst->isMemBarrier() &&
1216                 !squashed_inst->isWriteBarrier())) {
1217
1218                for (int src_reg_idx = 0;
1219                     src_reg_idx < squashed_inst->numSrcRegs();
1220                     src_reg_idx++)
1221                {
1222                    PhysRegIdPtr src_reg =
1223                        squashed_inst->renamedSrcRegIdx(src_reg_idx);
1224
1225                    // Only remove it from the dependency graph if it
1226                    // was placed there in the first place.
1227
1228                    // Instead of doing a linked list traversal, we
1229                    // can just remove these squashed instructions
1230                    // either at issue time, or when the register is
1231                    // overwritten.  The only downside to this is it
1232                    // leaves more room for error.
1233
1234                    if (!squashed_inst->isReadySrcRegIdx(src_reg_idx) &&
1235                        !src_reg->isFixedMapping()) {
1236                        dependGraph.remove(src_reg->flatIndex(),
1237                                           squashed_inst);
1238                    }
1239
1240
1241                    ++iqSquashedOperandsExamined;
1242                }
1243            } else if (!squashed_inst->isStoreConditional() ||
1244                       !squashed_inst->isCompleted()) {
1245                NonSpecMapIt ns_inst_it =
1246                    nonSpecInsts.find(squashed_inst->seqNum);
1247
1248                // we remove non-speculative instructions from
1249                // nonSpecInsts already when they are ready, and so we
1250                // cannot always expect to find them
1251                if (ns_inst_it == nonSpecInsts.end()) {
1252                    // loads that became ready but stalled on a
1253                    // blocked cache are alreayd removed from
1254                    // nonSpecInsts, and have not faulted
1255                    assert(squashed_inst->getFault() != NoFault ||
1256                           squashed_inst->isMemRef());
1257                } else {
1258
1259                    (*ns_inst_it).second = NULL;
1260
1261                    nonSpecInsts.erase(ns_inst_it);
1262
1263                    ++iqSquashedNonSpecRemoved;
1264                }
1265            }
1266
1267            // Might want to also clear out the head of the dependency graph.
1268
1269            // Mark it as squashed within the IQ.
1270            squashed_inst->setSquashedInIQ();
1271
1272            // @todo: Remove this hack where several statuses are set so the
1273            // inst will flow through the rest of the pipeline.
1274            squashed_inst->setIssued();
1275            squashed_inst->setCanCommit();
1276            squashed_inst->clearInIQ();
1277
1278            //Update Thread IQ Count
1279            count[squashed_inst->threadNumber]--;
1280
1281            ++freeEntries;
1282        }
1283
1284        instList[tid].erase(squash_it--);
1285        ++iqSquashedInstsExamined;
1286    }
1287}
1288
1289template <class Impl>
1290bool
1291InstructionQueue<Impl>::addToDependents(DynInstPtr &new_inst)
1292{
1293    // Loop through the instruction's source registers, adding
1294    // them to the dependency list if they are not ready.
1295    int8_t total_src_regs = new_inst->numSrcRegs();
1296    bool return_val = false;
1297
1298    for (int src_reg_idx = 0;
1299         src_reg_idx < total_src_regs;
1300         src_reg_idx++)
1301    {
1302        // Only add it to the dependency graph if it's not ready.
1303        if (!new_inst->isReadySrcRegIdx(src_reg_idx)) {
1304            PhysRegIdPtr src_reg = new_inst->renamedSrcRegIdx(src_reg_idx);
1305
1306            // Check the IQ's scoreboard to make sure the register
1307            // hasn't become ready while the instruction was in flight
1308            // between stages.  Only if it really isn't ready should
1309            // it be added to the dependency graph.
1310            if (src_reg->isFixedMapping()) {
1311                continue;
1312            } else if (!regScoreboard[src_reg->flatIndex()]) {
1313                DPRINTF(IQ, "Instruction PC %s has src reg %i (%s) that "
1314                        "is being added to the dependency chain.\n",
1315                        new_inst->pcState(), src_reg->index(),
1316                        src_reg->className());
1317
1318                dependGraph.insert(src_reg->flatIndex(), new_inst);
1319
1320                // Change the return value to indicate that something
1321                // was added to the dependency graph.
1322                return_val = true;
1323            } else {
1324                DPRINTF(IQ, "Instruction PC %s has src reg %i (%s) that "
1325                        "became ready before it reached the IQ.\n",
1326                        new_inst->pcState(), src_reg->index(),
1327                        src_reg->className());
1328                // Mark a register ready within the instruction.
1329                new_inst->markSrcRegReady(src_reg_idx);
1330            }
1331        }
1332    }
1333
1334    return return_val;
1335}
1336
1337template <class Impl>
1338void
1339InstructionQueue<Impl>::addToProducers(DynInstPtr &new_inst)
1340{
1341    // Nothing really needs to be marked when an instruction becomes
1342    // the producer of a register's value, but for convenience a ptr
1343    // to the producing instruction will be placed in the head node of
1344    // the dependency links.
1345    int8_t total_dest_regs = new_inst->numDestRegs();
1346
1347    for (int dest_reg_idx = 0;
1348         dest_reg_idx < total_dest_regs;
1349         dest_reg_idx++)
1350    {
1351        PhysRegIdPtr dest_reg = new_inst->renamedDestRegIdx(dest_reg_idx);
1352
1353        // Some registers have fixed mapping, and there is no need to track
1354        // dependencies as these instructions must be executed at commit.
1355        if (dest_reg->isFixedMapping()) {
1356            continue;
1357        }
1358
1359        if (!dependGraph.empty(dest_reg->flatIndex())) {
1360            dependGraph.dump();
1361            panic("Dependency graph %i (%s) (flat: %i) not empty!",
1362                  dest_reg->index(), dest_reg->className(),
1363                  dest_reg->flatIndex());
1364        }
1365
1366        dependGraph.setInst(dest_reg->flatIndex(), new_inst);
1367
1368        // Mark the scoreboard to say it's not yet ready.
1369        regScoreboard[dest_reg->flatIndex()] = false;
1370    }
1371}
1372
1373template <class Impl>
1374void
1375InstructionQueue<Impl>::addIfReady(DynInstPtr &inst)
1376{
1377    // If the instruction now has all of its source registers
1378    // available, then add it to the list of ready instructions.
1379    if (inst->readyToIssue()) {
1380
1381        //Add the instruction to the proper ready list.
1382        if (inst->isMemRef()) {
1383
1384            DPRINTF(IQ, "Checking if memory instruction can issue.\n");
1385
1386            // Message to the mem dependence unit that this instruction has
1387            // its registers ready.
1388            memDepUnit[inst->threadNumber].regsReady(inst);
1389
1390            return;
1391        }
1392
1393        OpClass op_class = inst->opClass();
1394
1395        DPRINTF(IQ, "Instruction is ready to issue, putting it onto "
1396                "the ready list, PC %s opclass:%i [sn:%lli].\n",
1397                inst->pcState(), op_class, inst->seqNum);
1398
1399        readyInsts[op_class].push(inst);
1400
1401        // Will need to reorder the list if either a queue is not on the list,
1402        // or it has an older instruction than last time.
1403        if (!queueOnList[op_class]) {
1404            addToOrderList(op_class);
1405        } else if (readyInsts[op_class].top()->seqNum  <
1406                   (*readyIt[op_class]).oldestInst) {
1407            listOrder.erase(readyIt[op_class]);
1408            addToOrderList(op_class);
1409        }
1410    }
1411}
1412
1413template <class Impl>
1414int
1415InstructionQueue<Impl>::countInsts()
1416{
1417#if 0
1418    //ksewell:This works but definitely could use a cleaner write
1419    //with a more intuitive way of counting. Right now it's
1420    //just brute force ....
1421    // Change the #if if you want to use this method.
1422    int total_insts = 0;
1423
1424    for (ThreadID tid = 0; tid < numThreads; ++tid) {
1425        ListIt count_it = instList[tid].begin();
1426
1427        while (count_it != instList[tid].end()) {
1428            if (!(*count_it)->isSquashed() && !(*count_it)->isSquashedInIQ()) {
1429                if (!(*count_it)->isIssued()) {
1430                    ++total_insts;
1431                } else if ((*count_it)->isMemRef() &&
1432                           !(*count_it)->memOpDone) {
1433                    // Loads that have not been marked as executed still count
1434                    // towards the total instructions.
1435                    ++total_insts;
1436                }
1437            }
1438
1439            ++count_it;
1440        }
1441    }
1442
1443    return total_insts;
1444#else
1445    return numEntries - freeEntries;
1446#endif
1447}
1448
1449template <class Impl>
1450void
1451InstructionQueue<Impl>::dumpLists()
1452{
1453    for (int i = 0; i < Num_OpClasses; ++i) {
1454        cprintf("Ready list %i size: %i\n", i, readyInsts[i].size());
1455
1456        cprintf("\n");
1457    }
1458
1459    cprintf("Non speculative list size: %i\n", nonSpecInsts.size());
1460
1461    NonSpecMapIt non_spec_it = nonSpecInsts.begin();
1462    NonSpecMapIt non_spec_end_it = nonSpecInsts.end();
1463
1464    cprintf("Non speculative list: ");
1465
1466    while (non_spec_it != non_spec_end_it) {
1467        cprintf("%s [sn:%lli]", (*non_spec_it).second->pcState(),
1468                (*non_spec_it).second->seqNum);
1469        ++non_spec_it;
1470    }
1471
1472    cprintf("\n");
1473
1474    ListOrderIt list_order_it = listOrder.begin();
1475    ListOrderIt list_order_end_it = listOrder.end();
1476    int i = 1;
1477
1478    cprintf("List order: ");
1479
1480    while (list_order_it != list_order_end_it) {
1481        cprintf("%i OpClass:%i [sn:%lli] ", i, (*list_order_it).queueType,
1482                (*list_order_it).oldestInst);
1483
1484        ++list_order_it;
1485        ++i;
1486    }
1487
1488    cprintf("\n");
1489}
1490
1491
1492template <class Impl>
1493void
1494InstructionQueue<Impl>::dumpInsts()
1495{
1496    for (ThreadID tid = 0; tid < numThreads; ++tid) {
1497        int num = 0;
1498        int valid_num = 0;
1499        ListIt inst_list_it = instList[tid].begin();
1500
1501        while (inst_list_it != instList[tid].end()) {
1502            cprintf("Instruction:%i\n", num);
1503            if (!(*inst_list_it)->isSquashed()) {
1504                if (!(*inst_list_it)->isIssued()) {
1505                    ++valid_num;
1506                    cprintf("Count:%i\n", valid_num);
1507                } else if ((*inst_list_it)->isMemRef() &&
1508                           !(*inst_list_it)->memOpDone()) {
1509                    // Loads that have not been marked as executed
1510                    // still count towards the total instructions.
1511                    ++valid_num;
1512                    cprintf("Count:%i\n", valid_num);
1513                }
1514            }
1515
1516            cprintf("PC: %s\n[sn:%lli]\n[tid:%i]\n"
1517                    "Issued:%i\nSquashed:%i\n",
1518                    (*inst_list_it)->pcState(),
1519                    (*inst_list_it)->seqNum,
1520                    (*inst_list_it)->threadNumber,
1521                    (*inst_list_it)->isIssued(),
1522                    (*inst_list_it)->isSquashed());
1523
1524            if ((*inst_list_it)->isMemRef()) {
1525                cprintf("MemOpDone:%i\n", (*inst_list_it)->memOpDone());
1526            }
1527
1528            cprintf("\n");
1529
1530            inst_list_it++;
1531            ++num;
1532        }
1533    }
1534
1535    cprintf("Insts to Execute list:\n");
1536
1537    int num = 0;
1538    int valid_num = 0;
1539    ListIt inst_list_it = instsToExecute.begin();
1540
1541    while (inst_list_it != instsToExecute.end())
1542    {
1543        cprintf("Instruction:%i\n",
1544                num);
1545        if (!(*inst_list_it)->isSquashed()) {
1546            if (!(*inst_list_it)->isIssued()) {
1547                ++valid_num;
1548                cprintf("Count:%i\n", valid_num);
1549            } else if ((*inst_list_it)->isMemRef() &&
1550                       !(*inst_list_it)->memOpDone()) {
1551                // Loads that have not been marked as executed
1552                // still count towards the total instructions.
1553                ++valid_num;
1554                cprintf("Count:%i\n", valid_num);
1555            }
1556        }
1557
1558        cprintf("PC: %s\n[sn:%lli]\n[tid:%i]\n"
1559                "Issued:%i\nSquashed:%i\n",
1560                (*inst_list_it)->pcState(),
1561                (*inst_list_it)->seqNum,
1562                (*inst_list_it)->threadNumber,
1563                (*inst_list_it)->isIssued(),
1564                (*inst_list_it)->isSquashed());
1565
1566        if ((*inst_list_it)->isMemRef()) {
1567            cprintf("MemOpDone:%i\n", (*inst_list_it)->memOpDone());
1568        }
1569
1570        cprintf("\n");
1571
1572        inst_list_it++;
1573        ++num;
1574    }
1575}
1576
1577#endif//__CPU_O3_INST_QUEUE_IMPL_HH__
1578