Merge more changes in from head.
[gem5.git] / src / cpu / o3 / iew_impl.hh
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 // @todo: Fix the instantaneous communication among all the stages within
32 // iew. There's a clear delay between issue and execute, yet backwards
33 // communication happens simultaneously.
34
35 #include <queue>
36
37 #include "base/timebuf.hh"
38 #include "cpu/o3/fu_pool.hh"
39 #include "cpu/o3/iew.hh"
40
41 template<class Impl>
42 DefaultIEW<Impl>::DefaultIEW(O3CPU *_cpu, Params *params)
43 : issueToExecQueue(params->backComSize, params->forwardComSize),
44 cpu(_cpu),
45 instQueue(_cpu, this, params),
46 ldstQueue(_cpu, this, params),
47 fuPool(params->fuPool),
48 commitToIEWDelay(params->commitToIEWDelay),
49 renameToIEWDelay(params->renameToIEWDelay),
50 issueToExecuteDelay(params->issueToExecuteDelay),
51 dispatchWidth(params->dispatchWidth),
52 issueWidth(params->issueWidth),
53 wbOutstanding(0),
54 wbWidth(params->wbWidth),
55 numThreads(params->numberOfThreads),
56 switchedOut(false)
57 {
58 _status = Active;
59 exeStatus = Running;
60 wbStatus = Idle;
61
62 // Setup wire to read instructions coming from issue.
63 fromIssue = issueToExecQueue.getWire(-issueToExecuteDelay);
64
65 // Instruction queue needs the queue between issue and execute.
66 instQueue.setIssueToExecuteQueue(&issueToExecQueue);
67
68 for (int i=0; i < numThreads; i++) {
69 dispatchStatus[i] = Running;
70 stalls[i].commit = false;
71 fetchRedirect[i] = false;
72 }
73
74 wbMax = wbWidth * params->wbDepth;
75
76 updateLSQNextCycle = false;
77
78 ableToIssue = true;
79
80 skidBufferMax = (3 * (renameToIEWDelay * params->renameWidth)) + issueWidth;
81 }
82
83 template <class Impl>
84 std::string
85 DefaultIEW<Impl>::name() const
86 {
87 return cpu->name() + ".iew";
88 }
89
90 template <class Impl>
91 void
92 DefaultIEW<Impl>::regStats()
93 {
94 using namespace Stats;
95
96 instQueue.regStats();
97 ldstQueue.regStats();
98
99 iewIdleCycles
100 .name(name() + ".iewIdleCycles")
101 .desc("Number of cycles IEW is idle");
102
103 iewSquashCycles
104 .name(name() + ".iewSquashCycles")
105 .desc("Number of cycles IEW is squashing");
106
107 iewBlockCycles
108 .name(name() + ".iewBlockCycles")
109 .desc("Number of cycles IEW is blocking");
110
111 iewUnblockCycles
112 .name(name() + ".iewUnblockCycles")
113 .desc("Number of cycles IEW is unblocking");
114
115 iewDispatchedInsts
116 .name(name() + ".iewDispatchedInsts")
117 .desc("Number of instructions dispatched to IQ");
118
119 iewDispSquashedInsts
120 .name(name() + ".iewDispSquashedInsts")
121 .desc("Number of squashed instructions skipped by dispatch");
122
123 iewDispLoadInsts
124 .name(name() + ".iewDispLoadInsts")
125 .desc("Number of dispatched load instructions");
126
127 iewDispStoreInsts
128 .name(name() + ".iewDispStoreInsts")
129 .desc("Number of dispatched store instructions");
130
131 iewDispNonSpecInsts
132 .name(name() + ".iewDispNonSpecInsts")
133 .desc("Number of dispatched non-speculative instructions");
134
135 iewIQFullEvents
136 .name(name() + ".iewIQFullEvents")
137 .desc("Number of times the IQ has become full, causing a stall");
138
139 iewLSQFullEvents
140 .name(name() + ".iewLSQFullEvents")
141 .desc("Number of times the LSQ has become full, causing a stall");
142
143 memOrderViolationEvents
144 .name(name() + ".memOrderViolationEvents")
145 .desc("Number of memory order violations");
146
147 predictedTakenIncorrect
148 .name(name() + ".predictedTakenIncorrect")
149 .desc("Number of branches that were predicted taken incorrectly");
150
151 predictedNotTakenIncorrect
152 .name(name() + ".predictedNotTakenIncorrect")
153 .desc("Number of branches that were predicted not taken incorrectly");
154
155 branchMispredicts
156 .name(name() + ".branchMispredicts")
157 .desc("Number of branch mispredicts detected at execute");
158
159 branchMispredicts = predictedTakenIncorrect + predictedNotTakenIncorrect;
160
161 iewExecutedInsts
162 .name(name() + ".iewExecutedInsts")
163 .desc("Number of executed instructions");
164
165 iewExecLoadInsts
166 .init(cpu->number_of_threads)
167 .name(name() + ".iewExecLoadInsts")
168 .desc("Number of load instructions executed")
169 .flags(total);
170
171 iewExecSquashedInsts
172 .name(name() + ".iewExecSquashedInsts")
173 .desc("Number of squashed instructions skipped in execute");
174
175 iewExecutedSwp
176 .init(cpu->number_of_threads)
177 .name(name() + ".EXEC:swp")
178 .desc("number of swp insts executed")
179 .flags(total);
180
181 iewExecutedNop
182 .init(cpu->number_of_threads)
183 .name(name() + ".EXEC:nop")
184 .desc("number of nop insts executed")
185 .flags(total);
186
187 iewExecutedRefs
188 .init(cpu->number_of_threads)
189 .name(name() + ".EXEC:refs")
190 .desc("number of memory reference insts executed")
191 .flags(total);
192
193 iewExecutedBranches
194 .init(cpu->number_of_threads)
195 .name(name() + ".EXEC:branches")
196 .desc("Number of branches executed")
197 .flags(total);
198
199 iewExecStoreInsts
200 .name(name() + ".EXEC:stores")
201 .desc("Number of stores executed")
202 .flags(total);
203 iewExecStoreInsts = iewExecutedRefs - iewExecLoadInsts;
204
205 iewExecRate
206 .name(name() + ".EXEC:rate")
207 .desc("Inst execution rate")
208 .flags(total);
209
210 iewExecRate = iewExecutedInsts / cpu->numCycles;
211
212 iewInstsToCommit
213 .init(cpu->number_of_threads)
214 .name(name() + ".WB:sent")
215 .desc("cumulative count of insts sent to commit")
216 .flags(total);
217
218 writebackCount
219 .init(cpu->number_of_threads)
220 .name(name() + ".WB:count")
221 .desc("cumulative count of insts written-back")
222 .flags(total);
223
224 producerInst
225 .init(cpu->number_of_threads)
226 .name(name() + ".WB:producers")
227 .desc("num instructions producing a value")
228 .flags(total);
229
230 consumerInst
231 .init(cpu->number_of_threads)
232 .name(name() + ".WB:consumers")
233 .desc("num instructions consuming a value")
234 .flags(total);
235
236 wbPenalized
237 .init(cpu->number_of_threads)
238 .name(name() + ".WB:penalized")
239 .desc("number of instrctions required to write to 'other' IQ")
240 .flags(total);
241
242 wbPenalizedRate
243 .name(name() + ".WB:penalized_rate")
244 .desc ("fraction of instructions written-back that wrote to 'other' IQ")
245 .flags(total);
246
247 wbPenalizedRate = wbPenalized / writebackCount;
248
249 wbFanout
250 .name(name() + ".WB:fanout")
251 .desc("average fanout of values written-back")
252 .flags(total);
253
254 wbFanout = producerInst / consumerInst;
255
256 wbRate
257 .name(name() + ".WB:rate")
258 .desc("insts written-back per cycle")
259 .flags(total);
260 wbRate = writebackCount / cpu->numCycles;
261 }
262
263 template<class Impl>
264 void
265 DefaultIEW<Impl>::initStage()
266 {
267 for (int tid=0; tid < numThreads; tid++) {
268 toRename->iewInfo[tid].usedIQ = true;
269 toRename->iewInfo[tid].freeIQEntries =
270 instQueue.numFreeEntries(tid);
271
272 toRename->iewInfo[tid].usedLSQ = true;
273 toRename->iewInfo[tid].freeLSQEntries =
274 ldstQueue.numFreeEntries(tid);
275 }
276
277 cpu->activateStage(O3CPU::IEWIdx);
278 }
279
280 template<class Impl>
281 void
282 DefaultIEW<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr)
283 {
284 timeBuffer = tb_ptr;
285
286 // Setup wire to read information from time buffer, from commit.
287 fromCommit = timeBuffer->getWire(-commitToIEWDelay);
288
289 // Setup wire to write information back to previous stages.
290 toRename = timeBuffer->getWire(0);
291
292 toFetch = timeBuffer->getWire(0);
293
294 // Instruction queue also needs main time buffer.
295 instQueue.setTimeBuffer(tb_ptr);
296 }
297
298 template<class Impl>
299 void
300 DefaultIEW<Impl>::setRenameQueue(TimeBuffer<RenameStruct> *rq_ptr)
301 {
302 renameQueue = rq_ptr;
303
304 // Setup wire to read information from rename queue.
305 fromRename = renameQueue->getWire(-renameToIEWDelay);
306 }
307
308 template<class Impl>
309 void
310 DefaultIEW<Impl>::setIEWQueue(TimeBuffer<IEWStruct> *iq_ptr)
311 {
312 iewQueue = iq_ptr;
313
314 // Setup wire to write instructions to commit.
315 toCommit = iewQueue->getWire(0);
316 }
317
318 template<class Impl>
319 void
320 DefaultIEW<Impl>::setActiveThreads(std::list<unsigned> *at_ptr)
321 {
322 activeThreads = at_ptr;
323
324 ldstQueue.setActiveThreads(at_ptr);
325 instQueue.setActiveThreads(at_ptr);
326 }
327
328 template<class Impl>
329 void
330 DefaultIEW<Impl>::setScoreboard(Scoreboard *sb_ptr)
331 {
332 scoreboard = sb_ptr;
333 }
334
335 template <class Impl>
336 bool
337 DefaultIEW<Impl>::drain()
338 {
339 // IEW is ready to drain at any time.
340 cpu->signalDrained();
341 return true;
342 }
343
344 template <class Impl>
345 void
346 DefaultIEW<Impl>::resume()
347 {
348 }
349
350 template <class Impl>
351 void
352 DefaultIEW<Impl>::switchOut()
353 {
354 // Clear any state.
355 switchedOut = true;
356 assert(insts[0].empty());
357 assert(skidBuffer[0].empty());
358
359 instQueue.switchOut();
360 ldstQueue.switchOut();
361 fuPool->switchOut();
362
363 for (int i = 0; i < numThreads; i++) {
364 while (!insts[i].empty())
365 insts[i].pop();
366 while (!skidBuffer[i].empty())
367 skidBuffer[i].pop();
368 }
369 }
370
371 template <class Impl>
372 void
373 DefaultIEW<Impl>::takeOverFrom()
374 {
375 // Reset all state.
376 _status = Active;
377 exeStatus = Running;
378 wbStatus = Idle;
379 switchedOut = false;
380
381 instQueue.takeOverFrom();
382 ldstQueue.takeOverFrom();
383 fuPool->takeOverFrom();
384
385 initStage();
386 cpu->activityThisCycle();
387
388 for (int i=0; i < numThreads; i++) {
389 dispatchStatus[i] = Running;
390 stalls[i].commit = false;
391 fetchRedirect[i] = false;
392 }
393
394 updateLSQNextCycle = false;
395
396 for (int i = 0; i < issueToExecQueue.getSize(); ++i) {
397 issueToExecQueue.advance();
398 }
399 }
400
401 template<class Impl>
402 void
403 DefaultIEW<Impl>::squash(unsigned tid)
404 {
405 DPRINTF(IEW, "[tid:%i]: Squashing all instructions.\n",
406 tid);
407
408 // Tell the IQ to start squashing.
409 instQueue.squash(tid);
410
411 // Tell the LDSTQ to start squashing.
412 ldstQueue.squash(fromCommit->commitInfo[tid].doneSeqNum, tid);
413 updatedQueues = true;
414
415 // Clear the skid buffer in case it has any data in it.
416 DPRINTF(IEW, "[tid:%i]: Removing skidbuffer instructions until [sn:%i].\n",
417 tid, fromCommit->commitInfo[tid].doneSeqNum);
418
419 while (!skidBuffer[tid].empty()) {
420 if (skidBuffer[tid].front()->isLoad() ||
421 skidBuffer[tid].front()->isStore() ) {
422 toRename->iewInfo[tid].dispatchedToLSQ++;
423 }
424
425 toRename->iewInfo[tid].dispatched++;
426
427 skidBuffer[tid].pop();
428 }
429
430 emptyRenameInsts(tid);
431 }
432
433 template<class Impl>
434 void
435 DefaultIEW<Impl>::squashDueToBranch(DynInstPtr &inst, unsigned tid)
436 {
437 DPRINTF(IEW, "[tid:%i]: Squashing from a specific instruction, PC: %#x "
438 "[sn:%i].\n", tid, inst->readPC(), inst->seqNum);
439
440 toCommit->squash[tid] = true;
441 toCommit->squashedSeqNum[tid] = inst->seqNum;
442 toCommit->mispredPC[tid] = inst->readPC();
443 toCommit->branchMispredict[tid] = true;
444
445 #if ISA_HAS_DELAY_SLOT
446 int instSize = sizeof(TheISA::MachInst);
447 toCommit->branchTaken[tid] =
448 !(inst->readNextPC() + instSize == inst->readNextNPC() &&
449 (inst->readNextPC() == inst->readPC() + instSize ||
450 inst->readNextPC() == inst->readPC() + 2 * instSize));
451 #else
452 toCommit->branchTaken[tid] = inst->readNextPC() !=
453 (inst->readPC() + sizeof(TheISA::MachInst));
454 #endif
455 toCommit->nextPC[tid] = inst->readNextPC();
456 toCommit->nextNPC[tid] = inst->readNextNPC();
457 toCommit->nextMicroPC[tid] = inst->readNextMicroPC();
458
459 toCommit->includeSquashInst[tid] = false;
460
461 wroteToTimeBuffer = true;
462 }
463
464 template<class Impl>
465 void
466 DefaultIEW<Impl>::squashDueToMemOrder(DynInstPtr &inst, unsigned tid)
467 {
468 DPRINTF(IEW, "[tid:%i]: Squashing from a specific instruction, "
469 "PC: %#x [sn:%i].\n", tid, inst->readPC(), inst->seqNum);
470
471 toCommit->squash[tid] = true;
472 toCommit->squashedSeqNum[tid] = inst->seqNum;
473 toCommit->nextPC[tid] = inst->readNextPC();
474 toCommit->nextNPC[tid] = inst->readNextNPC();
475 toCommit->branchMispredict[tid] = false;
476
477 toCommit->includeSquashInst[tid] = false;
478
479 wroteToTimeBuffer = true;
480 }
481
482 template<class Impl>
483 void
484 DefaultIEW<Impl>::squashDueToMemBlocked(DynInstPtr &inst, unsigned tid)
485 {
486 DPRINTF(IEW, "[tid:%i]: Memory blocked, squashing load and younger insts, "
487 "PC: %#x [sn:%i].\n", tid, inst->readPC(), inst->seqNum);
488
489 toCommit->squash[tid] = true;
490 toCommit->squashedSeqNum[tid] = inst->seqNum;
491 toCommit->nextPC[tid] = inst->readPC();
492 toCommit->nextNPC[tid] = inst->readNextPC();
493 toCommit->branchMispredict[tid] = false;
494
495 // Must include the broadcasted SN in the squash.
496 toCommit->includeSquashInst[tid] = true;
497
498 ldstQueue.setLoadBlockedHandled(tid);
499
500 wroteToTimeBuffer = true;
501 }
502
503 template<class Impl>
504 void
505 DefaultIEW<Impl>::block(unsigned tid)
506 {
507 DPRINTF(IEW, "[tid:%u]: Blocking.\n", tid);
508
509 if (dispatchStatus[tid] != Blocked &&
510 dispatchStatus[tid] != Unblocking) {
511 toRename->iewBlock[tid] = true;
512 wroteToTimeBuffer = true;
513 }
514
515 // Add the current inputs to the skid buffer so they can be
516 // reprocessed when this stage unblocks.
517 skidInsert(tid);
518
519 dispatchStatus[tid] = Blocked;
520 }
521
522 template<class Impl>
523 void
524 DefaultIEW<Impl>::unblock(unsigned tid)
525 {
526 DPRINTF(IEW, "[tid:%i]: Reading instructions out of the skid "
527 "buffer %u.\n",tid, tid);
528
529 // If the skid bufffer is empty, signal back to previous stages to unblock.
530 // Also switch status to running.
531 if (skidBuffer[tid].empty()) {
532 toRename->iewUnblock[tid] = true;
533 wroteToTimeBuffer = true;
534 DPRINTF(IEW, "[tid:%i]: Done unblocking.\n",tid);
535 dispatchStatus[tid] = Running;
536 }
537 }
538
539 template<class Impl>
540 void
541 DefaultIEW<Impl>::wakeDependents(DynInstPtr &inst)
542 {
543 instQueue.wakeDependents(inst);
544 }
545
546 template<class Impl>
547 void
548 DefaultIEW<Impl>::rescheduleMemInst(DynInstPtr &inst)
549 {
550 instQueue.rescheduleMemInst(inst);
551 }
552
553 template<class Impl>
554 void
555 DefaultIEW<Impl>::replayMemInst(DynInstPtr &inst)
556 {
557 instQueue.replayMemInst(inst);
558 }
559
560 template<class Impl>
561 void
562 DefaultIEW<Impl>::instToCommit(DynInstPtr &inst)
563 {
564 // This function should not be called after writebackInsts in a
565 // single cycle. That will cause problems with an instruction
566 // being added to the queue to commit without being processed by
567 // writebackInsts prior to being sent to commit.
568
569 // First check the time slot that this instruction will write
570 // to. If there are free write ports at the time, then go ahead
571 // and write the instruction to that time. If there are not,
572 // keep looking back to see where's the first time there's a
573 // free slot.
574 while ((*iewQueue)[wbCycle].insts[wbNumInst]) {
575 ++wbNumInst;
576 if (wbNumInst == wbWidth) {
577 ++wbCycle;
578 wbNumInst = 0;
579 }
580
581 assert((wbCycle * wbWidth + wbNumInst) <= wbMax);
582 }
583
584 DPRINTF(IEW, "Current wb cycle: %i, width: %i, numInst: %i\nwbActual:%i\n",
585 wbCycle, wbWidth, wbNumInst, wbCycle * wbWidth + wbNumInst);
586 // Add finished instruction to queue to commit.
587 (*iewQueue)[wbCycle].insts[wbNumInst] = inst;
588 (*iewQueue)[wbCycle].size++;
589 }
590
591 template <class Impl>
592 unsigned
593 DefaultIEW<Impl>::validInstsFromRename()
594 {
595 unsigned inst_count = 0;
596
597 for (int i=0; i<fromRename->size; i++) {
598 if (!fromRename->insts[i]->isSquashed())
599 inst_count++;
600 }
601
602 return inst_count;
603 }
604
605 template<class Impl>
606 void
607 DefaultIEW<Impl>::skidInsert(unsigned tid)
608 {
609 DynInstPtr inst = NULL;
610
611 while (!insts[tid].empty()) {
612 inst = insts[tid].front();
613
614 insts[tid].pop();
615
616 DPRINTF(Decode,"[tid:%i]: Inserting [sn:%lli] PC:%#x into "
617 "dispatch skidBuffer %i\n",tid, inst->seqNum,
618 inst->readPC(),tid);
619
620 skidBuffer[tid].push(inst);
621 }
622
623 assert(skidBuffer[tid].size() <= skidBufferMax &&
624 "Skidbuffer Exceeded Max Size");
625 }
626
627 template<class Impl>
628 int
629 DefaultIEW<Impl>::skidCount()
630 {
631 int max=0;
632
633 std::list<unsigned>::iterator threads = activeThreads->begin();
634 std::list<unsigned>::iterator end = activeThreads->end();
635
636 while (threads != end) {
637 unsigned tid = *threads++;
638 unsigned thread_count = skidBuffer[tid].size();
639 if (max < thread_count)
640 max = thread_count;
641 }
642
643 return max;
644 }
645
646 template<class Impl>
647 bool
648 DefaultIEW<Impl>::skidsEmpty()
649 {
650 std::list<unsigned>::iterator threads = activeThreads->begin();
651 std::list<unsigned>::iterator end = activeThreads->end();
652
653 while (threads != end) {
654 unsigned tid = *threads++;
655
656 if (!skidBuffer[tid].empty())
657 return false;
658 }
659
660 return true;
661 }
662
663 template <class Impl>
664 void
665 DefaultIEW<Impl>::updateStatus()
666 {
667 bool any_unblocking = false;
668
669 std::list<unsigned>::iterator threads = activeThreads->begin();
670 std::list<unsigned>::iterator end = activeThreads->end();
671
672 while (threads != end) {
673 unsigned tid = *threads++;
674
675 if (dispatchStatus[tid] == Unblocking) {
676 any_unblocking = true;
677 break;
678 }
679 }
680
681 // If there are no ready instructions waiting to be scheduled by the IQ,
682 // and there's no stores waiting to write back, and dispatch is not
683 // unblocking, then there is no internal activity for the IEW stage.
684 if (_status == Active && !instQueue.hasReadyInsts() &&
685 !ldstQueue.willWB() && !any_unblocking) {
686 DPRINTF(IEW, "IEW switching to idle\n");
687
688 deactivateStage();
689
690 _status = Inactive;
691 } else if (_status == Inactive && (instQueue.hasReadyInsts() ||
692 ldstQueue.willWB() ||
693 any_unblocking)) {
694 // Otherwise there is internal activity. Set to active.
695 DPRINTF(IEW, "IEW switching to active\n");
696
697 activateStage();
698
699 _status = Active;
700 }
701 }
702
703 template <class Impl>
704 void
705 DefaultIEW<Impl>::resetEntries()
706 {
707 instQueue.resetEntries();
708 ldstQueue.resetEntries();
709 }
710
711 template <class Impl>
712 void
713 DefaultIEW<Impl>::readStallSignals(unsigned tid)
714 {
715 if (fromCommit->commitBlock[tid]) {
716 stalls[tid].commit = true;
717 }
718
719 if (fromCommit->commitUnblock[tid]) {
720 assert(stalls[tid].commit);
721 stalls[tid].commit = false;
722 }
723 }
724
725 template <class Impl>
726 bool
727 DefaultIEW<Impl>::checkStall(unsigned tid)
728 {
729 bool ret_val(false);
730
731 if (stalls[tid].commit) {
732 DPRINTF(IEW,"[tid:%i]: Stall from Commit stage detected.\n",tid);
733 ret_val = true;
734 } else if (instQueue.isFull(tid)) {
735 DPRINTF(IEW,"[tid:%i]: Stall: IQ is full.\n",tid);
736 ret_val = true;
737 } else if (ldstQueue.isFull(tid)) {
738 DPRINTF(IEW,"[tid:%i]: Stall: LSQ is full\n",tid);
739
740 if (ldstQueue.numLoads(tid) > 0 ) {
741
742 DPRINTF(IEW,"[tid:%i]: LSQ oldest load: [sn:%i] \n",
743 tid,ldstQueue.getLoadHeadSeqNum(tid));
744 }
745
746 if (ldstQueue.numStores(tid) > 0) {
747
748 DPRINTF(IEW,"[tid:%i]: LSQ oldest store: [sn:%i] \n",
749 tid,ldstQueue.getStoreHeadSeqNum(tid));
750 }
751
752 ret_val = true;
753 } else if (ldstQueue.isStalled(tid)) {
754 DPRINTF(IEW,"[tid:%i]: Stall: LSQ stall detected.\n",tid);
755 ret_val = true;
756 }
757
758 return ret_val;
759 }
760
761 template <class Impl>
762 void
763 DefaultIEW<Impl>::checkSignalsAndUpdate(unsigned tid)
764 {
765 // Check if there's a squash signal, squash if there is
766 // Check stall signals, block if there is.
767 // If status was Blocked
768 // if so then go to unblocking
769 // If status was Squashing
770 // check if squashing is not high. Switch to running this cycle.
771
772 readStallSignals(tid);
773
774 if (fromCommit->commitInfo[tid].squash) {
775 squash(tid);
776
777 if (dispatchStatus[tid] == Blocked ||
778 dispatchStatus[tid] == Unblocking) {
779 toRename->iewUnblock[tid] = true;
780 wroteToTimeBuffer = true;
781 }
782
783 dispatchStatus[tid] = Squashing;
784
785 fetchRedirect[tid] = false;
786 return;
787 }
788
789 if (fromCommit->commitInfo[tid].robSquashing) {
790 DPRINTF(IEW, "[tid:%i]: ROB is still squashing.\n", tid);
791
792 dispatchStatus[tid] = Squashing;
793
794 emptyRenameInsts(tid);
795 wroteToTimeBuffer = true;
796 return;
797 }
798
799 if (checkStall(tid)) {
800 block(tid);
801 dispatchStatus[tid] = Blocked;
802 return;
803 }
804
805 if (dispatchStatus[tid] == Blocked) {
806 // Status from previous cycle was blocked, but there are no more stall
807 // conditions. Switch over to unblocking.
808 DPRINTF(IEW, "[tid:%i]: Done blocking, switching to unblocking.\n",
809 tid);
810
811 dispatchStatus[tid] = Unblocking;
812
813 unblock(tid);
814
815 return;
816 }
817
818 if (dispatchStatus[tid] == Squashing) {
819 // Switch status to running if rename isn't being told to block or
820 // squash this cycle.
821 DPRINTF(IEW, "[tid:%i]: Done squashing, switching to running.\n",
822 tid);
823
824 dispatchStatus[tid] = Running;
825
826 return;
827 }
828 }
829
830 template <class Impl>
831 void
832 DefaultIEW<Impl>::sortInsts()
833 {
834 int insts_from_rename = fromRename->size;
835 #ifdef DEBUG
836 for (int i = 0; i < numThreads; i++)
837 assert(insts[i].empty());
838 #endif
839 for (int i = 0; i < insts_from_rename; ++i) {
840 insts[fromRename->insts[i]->threadNumber].push(fromRename->insts[i]);
841 }
842 }
843
844 template <class Impl>
845 void
846 DefaultIEW<Impl>::emptyRenameInsts(unsigned tid)
847 {
848 DPRINTF(IEW, "[tid:%i]: Removing incoming rename instructions\n", tid);
849
850 while (!insts[tid].empty()) {
851
852 if (insts[tid].front()->isLoad() ||
853 insts[tid].front()->isStore() ) {
854 toRename->iewInfo[tid].dispatchedToLSQ++;
855 }
856
857 toRename->iewInfo[tid].dispatched++;
858
859 insts[tid].pop();
860 }
861 }
862
863 template <class Impl>
864 void
865 DefaultIEW<Impl>::wakeCPU()
866 {
867 cpu->wakeCPU();
868 }
869
870 template <class Impl>
871 void
872 DefaultIEW<Impl>::activityThisCycle()
873 {
874 DPRINTF(Activity, "Activity this cycle.\n");
875 cpu->activityThisCycle();
876 }
877
878 template <class Impl>
879 inline void
880 DefaultIEW<Impl>::activateStage()
881 {
882 DPRINTF(Activity, "Activating stage.\n");
883 cpu->activateStage(O3CPU::IEWIdx);
884 }
885
886 template <class Impl>
887 inline void
888 DefaultIEW<Impl>::deactivateStage()
889 {
890 DPRINTF(Activity, "Deactivating stage.\n");
891 cpu->deactivateStage(O3CPU::IEWIdx);
892 }
893
894 template<class Impl>
895 void
896 DefaultIEW<Impl>::dispatch(unsigned tid)
897 {
898 // If status is Running or idle,
899 // call dispatchInsts()
900 // If status is Unblocking,
901 // buffer any instructions coming from rename
902 // continue trying to empty skid buffer
903 // check if stall conditions have passed
904
905 if (dispatchStatus[tid] == Blocked) {
906 ++iewBlockCycles;
907
908 } else if (dispatchStatus[tid] == Squashing) {
909 ++iewSquashCycles;
910 }
911
912 // Dispatch should try to dispatch as many instructions as its bandwidth
913 // will allow, as long as it is not currently blocked.
914 if (dispatchStatus[tid] == Running ||
915 dispatchStatus[tid] == Idle) {
916 DPRINTF(IEW, "[tid:%i] Not blocked, so attempting to run "
917 "dispatch.\n", tid);
918
919 dispatchInsts(tid);
920 } else if (dispatchStatus[tid] == Unblocking) {
921 // Make sure that the skid buffer has something in it if the
922 // status is unblocking.
923 assert(!skidsEmpty());
924
925 // If the status was unblocking, then instructions from the skid
926 // buffer were used. Remove those instructions and handle
927 // the rest of unblocking.
928 dispatchInsts(tid);
929
930 ++iewUnblockCycles;
931
932 if (validInstsFromRename() && dispatchedAllInsts) {
933 // Add the current inputs to the skid buffer so they can be
934 // reprocessed when this stage unblocks.
935 skidInsert(tid);
936 }
937
938 unblock(tid);
939 }
940 }
941
942 template <class Impl>
943 void
944 DefaultIEW<Impl>::dispatchInsts(unsigned tid)
945 {
946 dispatchedAllInsts = true;
947
948 // Obtain instructions from skid buffer if unblocking, or queue from rename
949 // otherwise.
950 std::queue<DynInstPtr> &insts_to_dispatch =
951 dispatchStatus[tid] == Unblocking ?
952 skidBuffer[tid] : insts[tid];
953
954 int insts_to_add = insts_to_dispatch.size();
955
956 DynInstPtr inst;
957 bool add_to_iq = false;
958 int dis_num_inst = 0;
959
960 // Loop through the instructions, putting them in the instruction
961 // queue.
962 for ( ; dis_num_inst < insts_to_add &&
963 dis_num_inst < dispatchWidth;
964 ++dis_num_inst)
965 {
966 inst = insts_to_dispatch.front();
967
968 if (dispatchStatus[tid] == Unblocking) {
969 DPRINTF(IEW, "[tid:%i]: Issue: Examining instruction from skid "
970 "buffer\n", tid);
971 }
972
973 // Make sure there's a valid instruction there.
974 assert(inst);
975
976 DPRINTF(IEW, "[tid:%i]: Issue: Adding PC %#x [sn:%lli] [tid:%i] to "
977 "IQ.\n",
978 tid, inst->readPC(), inst->seqNum, inst->threadNumber);
979
980 // Be sure to mark these instructions as ready so that the
981 // commit stage can go ahead and execute them, and mark
982 // them as issued so the IQ doesn't reprocess them.
983
984 // Check for squashed instructions.
985 if (inst->isSquashed()) {
986 DPRINTF(IEW, "[tid:%i]: Issue: Squashed instruction encountered, "
987 "not adding to IQ.\n", tid);
988
989 ++iewDispSquashedInsts;
990
991 insts_to_dispatch.pop();
992
993 //Tell Rename That An Instruction has been processed
994 if (inst->isLoad() || inst->isStore()) {
995 toRename->iewInfo[tid].dispatchedToLSQ++;
996 }
997 toRename->iewInfo[tid].dispatched++;
998
999 continue;
1000 }
1001
1002 // Check for full conditions.
1003 if (instQueue.isFull(tid)) {
1004 DPRINTF(IEW, "[tid:%i]: Issue: IQ has become full.\n", tid);
1005
1006 // Call function to start blocking.
1007 block(tid);
1008
1009 // Set unblock to false. Special case where we are using
1010 // skidbuffer (unblocking) instructions but then we still
1011 // get full in the IQ.
1012 toRename->iewUnblock[tid] = false;
1013
1014 dispatchedAllInsts = false;
1015
1016 ++iewIQFullEvents;
1017 break;
1018 } else if (ldstQueue.isFull(tid)) {
1019 DPRINTF(IEW, "[tid:%i]: Issue: LSQ has become full.\n",tid);
1020
1021 // Call function to start blocking.
1022 block(tid);
1023
1024 // Set unblock to false. Special case where we are using
1025 // skidbuffer (unblocking) instructions but then we still
1026 // get full in the IQ.
1027 toRename->iewUnblock[tid] = false;
1028
1029 dispatchedAllInsts = false;
1030
1031 ++iewLSQFullEvents;
1032 break;
1033 }
1034
1035 // Otherwise issue the instruction just fine.
1036 if (inst->isLoad()) {
1037 DPRINTF(IEW, "[tid:%i]: Issue: Memory instruction "
1038 "encountered, adding to LSQ.\n", tid);
1039
1040 // Reserve a spot in the load store queue for this
1041 // memory access.
1042 ldstQueue.insertLoad(inst);
1043
1044 ++iewDispLoadInsts;
1045
1046 add_to_iq = true;
1047
1048 toRename->iewInfo[tid].dispatchedToLSQ++;
1049 } else if (inst->isStore()) {
1050 DPRINTF(IEW, "[tid:%i]: Issue: Memory instruction "
1051 "encountered, adding to LSQ.\n", tid);
1052
1053 ldstQueue.insertStore(inst);
1054
1055 ++iewDispStoreInsts;
1056
1057 if (inst->isStoreConditional()) {
1058 // Store conditionals need to be set as "canCommit()"
1059 // so that commit can process them when they reach the
1060 // head of commit.
1061 // @todo: This is somewhat specific to Alpha.
1062 inst->setCanCommit();
1063 instQueue.insertNonSpec(inst);
1064 add_to_iq = false;
1065
1066 ++iewDispNonSpecInsts;
1067 } else {
1068 add_to_iq = true;
1069 }
1070
1071 toRename->iewInfo[tid].dispatchedToLSQ++;
1072 } else if (inst->isMemBarrier() || inst->isWriteBarrier()) {
1073 // Same as non-speculative stores.
1074 inst->setCanCommit();
1075 instQueue.insertBarrier(inst);
1076 add_to_iq = false;
1077 } else if (inst->isNop()) {
1078 DPRINTF(IEW, "[tid:%i]: Issue: Nop instruction encountered, "
1079 "skipping.\n", tid);
1080
1081 inst->setIssued();
1082 inst->setExecuted();
1083 inst->setCanCommit();
1084
1085 instQueue.recordProducer(inst);
1086
1087 iewExecutedNop[tid]++;
1088
1089 add_to_iq = false;
1090 } else if (inst->isExecuted()) {
1091 assert(0 && "Instruction shouldn't be executed.\n");
1092 DPRINTF(IEW, "Issue: Executed branch encountered, "
1093 "skipping.\n");
1094
1095 inst->setIssued();
1096 inst->setCanCommit();
1097
1098 instQueue.recordProducer(inst);
1099
1100 add_to_iq = false;
1101 } else {
1102 add_to_iq = true;
1103 }
1104 if (inst->isNonSpeculative()) {
1105 DPRINTF(IEW, "[tid:%i]: Issue: Nonspeculative instruction "
1106 "encountered, skipping.\n", tid);
1107
1108 // Same as non-speculative stores.
1109 inst->setCanCommit();
1110
1111 // Specifically insert it as nonspeculative.
1112 instQueue.insertNonSpec(inst);
1113
1114 ++iewDispNonSpecInsts;
1115
1116 add_to_iq = false;
1117 }
1118
1119 // If the instruction queue is not full, then add the
1120 // instruction.
1121 if (add_to_iq) {
1122 instQueue.insert(inst);
1123 }
1124
1125 insts_to_dispatch.pop();
1126
1127 toRename->iewInfo[tid].dispatched++;
1128
1129 ++iewDispatchedInsts;
1130 }
1131
1132 if (!insts_to_dispatch.empty()) {
1133 DPRINTF(IEW,"[tid:%i]: Issue: Bandwidth Full. Blocking.\n", tid);
1134 block(tid);
1135 toRename->iewUnblock[tid] = false;
1136 }
1137
1138 if (dispatchStatus[tid] == Idle && dis_num_inst) {
1139 dispatchStatus[tid] = Running;
1140
1141 updatedQueues = true;
1142 }
1143
1144 dis_num_inst = 0;
1145 }
1146
1147 template <class Impl>
1148 void
1149 DefaultIEW<Impl>::printAvailableInsts()
1150 {
1151 int inst = 0;
1152
1153 std::cout << "Available Instructions: ";
1154
1155 while (fromIssue->insts[inst]) {
1156
1157 if (inst%3==0) std::cout << "\n\t";
1158
1159 std::cout << "PC: " << fromIssue->insts[inst]->readPC()
1160 << " TN: " << fromIssue->insts[inst]->threadNumber
1161 << " SN: " << fromIssue->insts[inst]->seqNum << " | ";
1162
1163 inst++;
1164
1165 }
1166
1167 std::cout << "\n";
1168 }
1169
1170 template <class Impl>
1171 void
1172 DefaultIEW<Impl>::executeInsts()
1173 {
1174 wbNumInst = 0;
1175 wbCycle = 0;
1176
1177 std::list<unsigned>::iterator threads = activeThreads->begin();
1178 std::list<unsigned>::iterator end = activeThreads->end();
1179
1180 while (threads != end) {
1181 unsigned tid = *threads++;
1182 fetchRedirect[tid] = false;
1183 }
1184
1185 // Uncomment this if you want to see all available instructions.
1186 // printAvailableInsts();
1187
1188 // Execute/writeback any instructions that are available.
1189 int insts_to_execute = fromIssue->size;
1190 int inst_num = 0;
1191 for (; inst_num < insts_to_execute;
1192 ++inst_num) {
1193
1194 DPRINTF(IEW, "Execute: Executing instructions from IQ.\n");
1195
1196 DynInstPtr inst = instQueue.getInstToExecute();
1197
1198 DPRINTF(IEW, "Execute: Processing PC %#x, [tid:%i] [sn:%i].\n",
1199 inst->readPC(), inst->threadNumber,inst->seqNum);
1200
1201 // Check if the instruction is squashed; if so then skip it
1202 if (inst->isSquashed()) {
1203 DPRINTF(IEW, "Execute: Instruction was squashed.\n");
1204
1205 // Consider this instruction executed so that commit can go
1206 // ahead and retire the instruction.
1207 inst->setExecuted();
1208
1209 // Not sure if I should set this here or just let commit try to
1210 // commit any squashed instructions. I like the latter a bit more.
1211 inst->setCanCommit();
1212
1213 ++iewExecSquashedInsts;
1214
1215 decrWb(inst->seqNum);
1216 continue;
1217 }
1218
1219 Fault fault = NoFault;
1220
1221 // Execute instruction.
1222 // Note that if the instruction faults, it will be handled
1223 // at the commit stage.
1224 if (inst->isMemRef() &&
1225 (!inst->isDataPrefetch() && !inst->isInstPrefetch())) {
1226 DPRINTF(IEW, "Execute: Calculating address for memory "
1227 "reference.\n");
1228
1229 // Tell the LDSTQ to execute this instruction (if it is a load).
1230 if (inst->isLoad()) {
1231 // Loads will mark themselves as executed, and their writeback
1232 // event adds the instruction to the queue to commit
1233 fault = ldstQueue.executeLoad(inst);
1234 } else if (inst->isStore()) {
1235 fault = ldstQueue.executeStore(inst);
1236
1237 // If the store had a fault then it may not have a mem req
1238 if (!inst->isStoreConditional() && fault == NoFault) {
1239 inst->setExecuted();
1240
1241 instToCommit(inst);
1242 } else if (fault != NoFault) {
1243 // If the instruction faulted, then we need to send it along to commit
1244 // without the instruction completing.
1245 DPRINTF(IEW, "Store has fault %s! [sn:%lli]\n",
1246 fault->name(), inst->seqNum);
1247
1248 // Send this instruction to commit, also make sure iew stage
1249 // realizes there is activity.
1250 inst->setExecuted();
1251
1252 instToCommit(inst);
1253 activityThisCycle();
1254 }
1255
1256 // Store conditionals will mark themselves as
1257 // executed, and their writeback event will add the
1258 // instruction to the queue to commit.
1259 } else {
1260 panic("Unexpected memory type!\n");
1261 }
1262
1263 } else {
1264 inst->execute();
1265
1266 inst->setExecuted();
1267
1268 instToCommit(inst);
1269 }
1270
1271 updateExeInstStats(inst);
1272
1273 // Check if branch prediction was correct, if not then we need
1274 // to tell commit to squash in flight instructions. Only
1275 // handle this if there hasn't already been something that
1276 // redirects fetch in this group of instructions.
1277
1278 // This probably needs to prioritize the redirects if a different
1279 // scheduler is used. Currently the scheduler schedules the oldest
1280 // instruction first, so the branch resolution order will be correct.
1281 unsigned tid = inst->threadNumber;
1282
1283 if (!fetchRedirect[tid] ||
1284 toCommit->squashedSeqNum[tid] > inst->seqNum) {
1285
1286 if (inst->mispredicted()) {
1287 fetchRedirect[tid] = true;
1288
1289 DPRINTF(IEW, "Execute: Branch mispredict detected.\n");
1290 DPRINTF(IEW, "Predicted target was %#x, %#x.\n",
1291 inst->readPredPC(), inst->readPredNPC());
1292 DPRINTF(IEW, "Execute: Redirecting fetch to PC: %#x,"
1293 " NPC: %#x.\n", inst->readNextPC(),
1294 inst->readNextNPC());
1295 // If incorrect, then signal the ROB that it must be squashed.
1296 squashDueToBranch(inst, tid);
1297
1298 if (inst->readPredTaken()) {
1299 predictedTakenIncorrect++;
1300 } else {
1301 predictedNotTakenIncorrect++;
1302 }
1303 } else if (ldstQueue.violation(tid)) {
1304 assert(inst->isMemRef());
1305 // If there was an ordering violation, then get the
1306 // DynInst that caused the violation. Note that this
1307 // clears the violation signal.
1308 DynInstPtr violator;
1309 violator = ldstQueue.getMemDepViolator(tid);
1310
1311 DPRINTF(IEW, "LDSTQ detected a violation. Violator PC: "
1312 "%#x, inst PC: %#x. Addr is: %#x.\n",
1313 violator->readPC(), inst->readPC(), inst->physEffAddr);
1314
1315 // Ensure the violating instruction is older than
1316 // current squash
1317 /* if (fetchRedirect[tid] &&
1318 violator->seqNum >= toCommit->squashedSeqNum[tid] + 1)
1319 continue;
1320 */
1321 fetchRedirect[tid] = true;
1322
1323 // Tell the instruction queue that a violation has occured.
1324 instQueue.violation(inst, violator);
1325
1326 // Squash.
1327 squashDueToMemOrder(inst,tid);
1328
1329 ++memOrderViolationEvents;
1330 } else if (ldstQueue.loadBlocked(tid) &&
1331 !ldstQueue.isLoadBlockedHandled(tid)) {
1332 fetchRedirect[tid] = true;
1333
1334 DPRINTF(IEW, "Load operation couldn't execute because the "
1335 "memory system is blocked. PC: %#x [sn:%lli]\n",
1336 inst->readPC(), inst->seqNum);
1337
1338 squashDueToMemBlocked(inst, tid);
1339 }
1340 } else {
1341 // Reset any state associated with redirects that will not
1342 // be used.
1343 if (ldstQueue.violation(tid)) {
1344 assert(inst->isMemRef());
1345
1346 DynInstPtr violator = ldstQueue.getMemDepViolator(tid);
1347
1348 DPRINTF(IEW, "LDSTQ detected a violation. Violator PC: "
1349 "%#x, inst PC: %#x. Addr is: %#x.\n",
1350 violator->readPC(), inst->readPC(), inst->physEffAddr);
1351 DPRINTF(IEW, "Violation will not be handled because "
1352 "already squashing\n");
1353
1354 ++memOrderViolationEvents;
1355 }
1356 if (ldstQueue.loadBlocked(tid) &&
1357 !ldstQueue.isLoadBlockedHandled(tid)) {
1358 DPRINTF(IEW, "Load operation couldn't execute because the "
1359 "memory system is blocked. PC: %#x [sn:%lli]\n",
1360 inst->readPC(), inst->seqNum);
1361 DPRINTF(IEW, "Blocked load will not be handled because "
1362 "already squashing\n");
1363
1364 ldstQueue.setLoadBlockedHandled(tid);
1365 }
1366
1367 }
1368 }
1369
1370 // Update and record activity if we processed any instructions.
1371 if (inst_num) {
1372 if (exeStatus == Idle) {
1373 exeStatus = Running;
1374 }
1375
1376 updatedQueues = true;
1377
1378 cpu->activityThisCycle();
1379 }
1380
1381 // Need to reset this in case a writeback event needs to write into the
1382 // iew queue. That way the writeback event will write into the correct
1383 // spot in the queue.
1384 wbNumInst = 0;
1385 }
1386
1387 template <class Impl>
1388 void
1389 DefaultIEW<Impl>::writebackInsts()
1390 {
1391 // Loop through the head of the time buffer and wake any
1392 // dependents. These instructions are about to write back. Also
1393 // mark scoreboard that this instruction is finally complete.
1394 // Either have IEW have direct access to scoreboard, or have this
1395 // as part of backwards communication.
1396 for (int inst_num = 0; inst_num < wbWidth &&
1397 toCommit->insts[inst_num]; inst_num++) {
1398 DynInstPtr inst = toCommit->insts[inst_num];
1399 int tid = inst->threadNumber;
1400
1401 DPRINTF(IEW, "Sending instructions to commit, [sn:%lli] PC %#x.\n",
1402 inst->seqNum, inst->readPC());
1403
1404 iewInstsToCommit[tid]++;
1405
1406 // Some instructions will be sent to commit without having
1407 // executed because they need commit to handle them.
1408 // E.g. Uncached loads have not actually executed when they
1409 // are first sent to commit. Instead commit must tell the LSQ
1410 // when it's ready to execute the uncached load.
1411 if (!inst->isSquashed() && inst->isExecuted() && inst->getFault() == NoFault) {
1412 int dependents = instQueue.wakeDependents(inst);
1413
1414 for (int i = 0; i < inst->numDestRegs(); i++) {
1415 //mark as Ready
1416 DPRINTF(IEW,"Setting Destination Register %i\n",
1417 inst->renamedDestRegIdx(i));
1418 scoreboard->setReg(inst->renamedDestRegIdx(i));
1419 }
1420
1421 if (dependents) {
1422 producerInst[tid]++;
1423 consumerInst[tid]+= dependents;
1424 }
1425 writebackCount[tid]++;
1426 }
1427
1428 decrWb(inst->seqNum);
1429 }
1430 }
1431
1432 template<class Impl>
1433 void
1434 DefaultIEW<Impl>::tick()
1435 {
1436 wbNumInst = 0;
1437 wbCycle = 0;
1438
1439 wroteToTimeBuffer = false;
1440 updatedQueues = false;
1441
1442 sortInsts();
1443
1444 // Free function units marked as being freed this cycle.
1445 fuPool->processFreeUnits();
1446
1447 std::list<unsigned>::iterator threads = activeThreads->begin();
1448 std::list<unsigned>::iterator end = activeThreads->end();
1449
1450 // Check stall and squash signals, dispatch any instructions.
1451 while (threads != end) {
1452 unsigned tid = *threads++;
1453
1454 DPRINTF(IEW,"Issue: Processing [tid:%i]\n",tid);
1455
1456 checkSignalsAndUpdate(tid);
1457 dispatch(tid);
1458 }
1459
1460 if (exeStatus != Squashing) {
1461 executeInsts();
1462
1463 writebackInsts();
1464
1465 // Have the instruction queue try to schedule any ready instructions.
1466 // (In actuality, this scheduling is for instructions that will
1467 // be executed next cycle.)
1468 instQueue.scheduleReadyInsts();
1469
1470 // Also should advance its own time buffers if the stage ran.
1471 // Not the best place for it, but this works (hopefully).
1472 issueToExecQueue.advance();
1473 }
1474
1475 bool broadcast_free_entries = false;
1476
1477 if (updatedQueues || exeStatus == Running || updateLSQNextCycle) {
1478 exeStatus = Idle;
1479 updateLSQNextCycle = false;
1480
1481 broadcast_free_entries = true;
1482 }
1483
1484 // Writeback any stores using any leftover bandwidth.
1485 ldstQueue.writebackStores();
1486
1487 // Check the committed load/store signals to see if there's a load
1488 // or store to commit. Also check if it's being told to execute a
1489 // nonspeculative instruction.
1490 // This is pretty inefficient...
1491
1492 threads = activeThreads->begin();
1493 while (threads != end) {
1494 unsigned tid = (*threads++);
1495
1496 DPRINTF(IEW,"Processing [tid:%i]\n",tid);
1497
1498 // Update structures based on instructions committed.
1499 if (fromCommit->commitInfo[tid].doneSeqNum != 0 &&
1500 !fromCommit->commitInfo[tid].squash &&
1501 !fromCommit->commitInfo[tid].robSquashing) {
1502
1503 ldstQueue.commitStores(fromCommit->commitInfo[tid].doneSeqNum,tid);
1504
1505 ldstQueue.commitLoads(fromCommit->commitInfo[tid].doneSeqNum,tid);
1506
1507 updateLSQNextCycle = true;
1508 instQueue.commit(fromCommit->commitInfo[tid].doneSeqNum,tid);
1509 }
1510
1511 if (fromCommit->commitInfo[tid].nonSpecSeqNum != 0) {
1512
1513 //DPRINTF(IEW,"NonspecInst from thread %i",tid);
1514 if (fromCommit->commitInfo[tid].uncached) {
1515 instQueue.replayMemInst(fromCommit->commitInfo[tid].uncachedLoad);
1516 fromCommit->commitInfo[tid].uncachedLoad->setAtCommit();
1517 } else {
1518 instQueue.scheduleNonSpec(
1519 fromCommit->commitInfo[tid].nonSpecSeqNum);
1520 }
1521 }
1522
1523 if (broadcast_free_entries) {
1524 toFetch->iewInfo[tid].iqCount =
1525 instQueue.getCount(tid);
1526 toFetch->iewInfo[tid].ldstqCount =
1527 ldstQueue.getCount(tid);
1528
1529 toRename->iewInfo[tid].usedIQ = true;
1530 toRename->iewInfo[tid].freeIQEntries =
1531 instQueue.numFreeEntries();
1532 toRename->iewInfo[tid].usedLSQ = true;
1533 toRename->iewInfo[tid].freeLSQEntries =
1534 ldstQueue.numFreeEntries(tid);
1535
1536 wroteToTimeBuffer = true;
1537 }
1538
1539 DPRINTF(IEW, "[tid:%i], Dispatch dispatched %i instructions.\n",
1540 tid, toRename->iewInfo[tid].dispatched);
1541 }
1542
1543 DPRINTF(IEW, "IQ has %i free entries (Can schedule: %i). "
1544 "LSQ has %i free entries.\n",
1545 instQueue.numFreeEntries(), instQueue.hasReadyInsts(),
1546 ldstQueue.numFreeEntries());
1547
1548 updateStatus();
1549
1550 if (wroteToTimeBuffer) {
1551 DPRINTF(Activity, "Activity this cycle.\n");
1552 cpu->activityThisCycle();
1553 }
1554 }
1555
1556 template <class Impl>
1557 void
1558 DefaultIEW<Impl>::updateExeInstStats(DynInstPtr &inst)
1559 {
1560 int thread_number = inst->threadNumber;
1561
1562 //
1563 // Pick off the software prefetches
1564 //
1565 #ifdef TARGET_ALPHA
1566 if (inst->isDataPrefetch())
1567 iewExecutedSwp[thread_number]++;
1568 else
1569 iewIewExecutedcutedInsts++;
1570 #else
1571 iewExecutedInsts++;
1572 #endif
1573
1574 //
1575 // Control operations
1576 //
1577 if (inst->isControl())
1578 iewExecutedBranches[thread_number]++;
1579
1580 //
1581 // Memory operations
1582 //
1583 if (inst->isMemRef()) {
1584 iewExecutedRefs[thread_number]++;
1585
1586 if (inst->isLoad()) {
1587 iewExecLoadInsts[thread_number]++;
1588 }
1589 }
1590 }