gem5
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
iew_impl.hh
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2010-2013 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  */
43 
44 #ifndef __CPU_O3_IEW_IMPL_IMPL_HH__
45 #define __CPU_O3_IEW_IMPL_IMPL_HH__
46 
47 // @todo: Fix the instantaneous communication among all the stages within
48 // iew. There's a clear delay between issue and execute, yet backwards
49 // communication happens simultaneously.
50 
51 #include <queue>
52 
53 #include "arch/utility.hh"
54 #include "config/the_isa.hh"
55 #include "cpu/checker/cpu.hh"
56 #include "cpu/o3/fu_pool.hh"
57 #include "cpu/o3/iew.hh"
58 #include "cpu/timebuf.hh"
59 #include "debug/Activity.hh"
60 #include "debug/Drain.hh"
61 #include "debug/IEW.hh"
62 #include "debug/O3PipeView.hh"
63 #include "params/DerivO3CPU.hh"
64 
65 using namespace std;
66 
67 template<class Impl>
68 DefaultIEW<Impl>::DefaultIEW(O3CPU *_cpu, DerivO3CPUParams *params)
69  : issueToExecQueue(params->backComSize, params->forwardComSize),
70  cpu(_cpu),
71  instQueue(_cpu, this, params),
72  ldstQueue(_cpu, this, params),
73  fuPool(params->fuPool),
74  commitToIEWDelay(params->commitToIEWDelay),
75  renameToIEWDelay(params->renameToIEWDelay),
76  issueToExecuteDelay(params->issueToExecuteDelay),
77  dispatchWidth(params->dispatchWidth),
78  issueWidth(params->issueWidth),
79  wbWidth(params->wbWidth),
80  numThreads(params->numThreads)
81 {
82  if (dispatchWidth > Impl::MaxWidth)
83  fatal("dispatchWidth (%d) is larger than compiled limit (%d),\n"
84  "\tincrease MaxWidth in src/cpu/o3/impl.hh\n",
85  dispatchWidth, static_cast<int>(Impl::MaxWidth));
86  if (issueWidth > Impl::MaxWidth)
87  fatal("issueWidth (%d) is larger than compiled limit (%d),\n"
88  "\tincrease MaxWidth in src/cpu/o3/impl.hh\n",
89  issueWidth, static_cast<int>(Impl::MaxWidth));
90  if (wbWidth > Impl::MaxWidth)
91  fatal("wbWidth (%d) is larger than compiled limit (%d),\n"
92  "\tincrease MaxWidth in src/cpu/o3/impl.hh\n",
93  wbWidth, static_cast<int>(Impl::MaxWidth));
94 
95  _status = Active;
97  wbStatus = Idle;
98 
99  // Setup wire to read instructions coming from issue.
101 
102  // Instruction queue needs the queue between issue and execute.
103  instQueue.setIssueToExecuteQueue(&issueToExecQueue);
104 
105  for (ThreadID tid = 0; tid < numThreads; tid++) {
106  dispatchStatus[tid] = Running;
107  fetchRedirect[tid] = false;
108  }
109 
110  updateLSQNextCycle = false;
111 
112  skidBufferMax = (renameToIEWDelay + 1) * params->renameWidth;
113 }
114 
115 template <class Impl>
116 std::string
118 {
119  return cpu->name() + ".iew";
120 }
121 
122 template <class Impl>
123 void
125 {
126  ppDispatch = new ProbePointArg<DynInstPtr>(cpu->getProbeManager(), "Dispatch");
127  ppMispredict = new ProbePointArg<DynInstPtr>(cpu->getProbeManager(), "Mispredict");
132  ppExecute = new ProbePointArg<DynInstPtr>(cpu->getProbeManager(),
133  "Execute");
138  ppToCommit = new ProbePointArg<DynInstPtr>(cpu->getProbeManager(),
139  "ToCommit");
140 }
141 
142 template <class Impl>
143 void
145 {
146  using namespace Stats;
147 
148  instQueue.regStats();
149  ldstQueue.regStats();
150 
151  iewIdleCycles
152  .name(name() + ".iewIdleCycles")
153  .desc("Number of cycles IEW is idle");
154 
155  iewSquashCycles
156  .name(name() + ".iewSquashCycles")
157  .desc("Number of cycles IEW is squashing");
158 
159  iewBlockCycles
160  .name(name() + ".iewBlockCycles")
161  .desc("Number of cycles IEW is blocking");
162 
163  iewUnblockCycles
164  .name(name() + ".iewUnblockCycles")
165  .desc("Number of cycles IEW is unblocking");
166 
167  iewDispatchedInsts
168  .name(name() + ".iewDispatchedInsts")
169  .desc("Number of instructions dispatched to IQ");
170 
171  iewDispSquashedInsts
172  .name(name() + ".iewDispSquashedInsts")
173  .desc("Number of squashed instructions skipped by dispatch");
174 
175  iewDispLoadInsts
176  .name(name() + ".iewDispLoadInsts")
177  .desc("Number of dispatched load instructions");
178 
179  iewDispStoreInsts
180  .name(name() + ".iewDispStoreInsts")
181  .desc("Number of dispatched store instructions");
182 
183  iewDispNonSpecInsts
184  .name(name() + ".iewDispNonSpecInsts")
185  .desc("Number of dispatched non-speculative instructions");
186 
187  iewIQFullEvents
188  .name(name() + ".iewIQFullEvents")
189  .desc("Number of times the IQ has become full, causing a stall");
190 
191  iewLSQFullEvents
192  .name(name() + ".iewLSQFullEvents")
193  .desc("Number of times the LSQ has become full, causing a stall");
194 
195  memOrderViolationEvents
196  .name(name() + ".memOrderViolationEvents")
197  .desc("Number of memory order violations");
198 
199  predictedTakenIncorrect
200  .name(name() + ".predictedTakenIncorrect")
201  .desc("Number of branches that were predicted taken incorrectly");
202 
203  predictedNotTakenIncorrect
204  .name(name() + ".predictedNotTakenIncorrect")
205  .desc("Number of branches that were predicted not taken incorrectly");
206 
207  branchMispredicts
208  .name(name() + ".branchMispredicts")
209  .desc("Number of branch mispredicts detected at execute");
210 
211  branchMispredicts = predictedTakenIncorrect + predictedNotTakenIncorrect;
212 
213  iewExecutedInsts
214  .name(name() + ".iewExecutedInsts")
215  .desc("Number of executed instructions");
216 
217  iewExecLoadInsts
218  .init(cpu->numThreads)
219  .name(name() + ".iewExecLoadInsts")
220  .desc("Number of load instructions executed")
221  .flags(total);
222 
223  iewExecSquashedInsts
224  .name(name() + ".iewExecSquashedInsts")
225  .desc("Number of squashed instructions skipped in execute");
226 
227  iewExecutedSwp
228  .init(cpu->numThreads)
229  .name(name() + ".exec_swp")
230  .desc("number of swp insts executed")
231  .flags(total);
232 
233  iewExecutedNop
234  .init(cpu->numThreads)
235  .name(name() + ".exec_nop")
236  .desc("number of nop insts executed")
237  .flags(total);
238 
239  iewExecutedRefs
240  .init(cpu->numThreads)
241  .name(name() + ".exec_refs")
242  .desc("number of memory reference insts executed")
243  .flags(total);
244 
245  iewExecutedBranches
246  .init(cpu->numThreads)
247  .name(name() + ".exec_branches")
248  .desc("Number of branches executed")
249  .flags(total);
250 
251  iewExecStoreInsts
252  .name(name() + ".exec_stores")
253  .desc("Number of stores executed")
254  .flags(total);
255  iewExecStoreInsts = iewExecutedRefs - iewExecLoadInsts;
256 
257  iewExecRate
258  .name(name() + ".exec_rate")
259  .desc("Inst execution rate")
260  .flags(total);
261 
262  iewExecRate = iewExecutedInsts / cpu->numCycles;
263 
264  iewInstsToCommit
265  .init(cpu->numThreads)
266  .name(name() + ".wb_sent")
267  .desc("cumulative count of insts sent to commit")
268  .flags(total);
269 
270  writebackCount
271  .init(cpu->numThreads)
272  .name(name() + ".wb_count")
273  .desc("cumulative count of insts written-back")
274  .flags(total);
275 
276  producerInst
277  .init(cpu->numThreads)
278  .name(name() + ".wb_producers")
279  .desc("num instructions producing a value")
280  .flags(total);
281 
282  consumerInst
283  .init(cpu->numThreads)
284  .name(name() + ".wb_consumers")
285  .desc("num instructions consuming a value")
286  .flags(total);
287 
288  wbFanout
289  .name(name() + ".wb_fanout")
290  .desc("average fanout of values written-back")
291  .flags(total);
292 
293  wbFanout = producerInst / consumerInst;
294 
295  wbRate
296  .name(name() + ".wb_rate")
297  .desc("insts written-back per cycle")
298  .flags(total);
299  wbRate = writebackCount / cpu->numCycles;
300 }
301 
302 template<class Impl>
303 void
305 {
306  for (ThreadID tid = 0; tid < numThreads; tid++) {
307  toRename->iewInfo[tid].usedIQ = true;
308  toRename->iewInfo[tid].freeIQEntries =
309  instQueue.numFreeEntries(tid);
310 
311  toRename->iewInfo[tid].usedLSQ = true;
312  toRename->iewInfo[tid].freeLQEntries = ldstQueue.numFreeLoadEntries(tid);
313  toRename->iewInfo[tid].freeSQEntries = ldstQueue.numFreeStoreEntries(tid);
314  }
315 
316  // Initialize the checker's dcache port here
317  if (cpu->checker) {
318  cpu->checker->setDcachePort(&cpu->getDataPort());
319  }
320 
321  cpu->activateStage(O3CPU::IEWIdx);
322 }
323 
324 template<class Impl>
325 void
327 {
328  timeBuffer = tb_ptr;
329 
330  // Setup wire to read information from time buffer, from commit.
331  fromCommit = timeBuffer->getWire(-commitToIEWDelay);
332 
333  // Setup wire to write information back to previous stages.
334  toRename = timeBuffer->getWire(0);
335 
336  toFetch = timeBuffer->getWire(0);
337 
338  // Instruction queue also needs main time buffer.
339  instQueue.setTimeBuffer(tb_ptr);
340 }
341 
342 template<class Impl>
343 void
345 {
346  renameQueue = rq_ptr;
347 
348  // Setup wire to read information from rename queue.
349  fromRename = renameQueue->getWire(-renameToIEWDelay);
350 }
351 
352 template<class Impl>
353 void
355 {
356  iewQueue = iq_ptr;
357 
358  // Setup wire to write instructions to commit.
359  toCommit = iewQueue->getWire(0);
360 }
361 
362 template<class Impl>
363 void
365 {
366  activeThreads = at_ptr;
367 
368  ldstQueue.setActiveThreads(at_ptr);
369  instQueue.setActiveThreads(at_ptr);
370 }
371 
372 template<class Impl>
373 void
375 {
376  scoreboard = sb_ptr;
377 }
378 
379 template <class Impl>
380 bool
382 {
383  bool drained = ldstQueue.isDrained() && instQueue.isDrained();
384 
385  for (ThreadID tid = 0; tid < numThreads; tid++) {
386  if (!insts[tid].empty()) {
387  DPRINTF(Drain, "%i: Insts not empty.\n", tid);
388  drained = false;
389  }
390  if (!skidBuffer[tid].empty()) {
391  DPRINTF(Drain, "%i: Skid buffer not empty.\n", tid);
392  drained = false;
393  }
394  drained = drained && dispatchStatus[tid] == Running;
395  }
396 
397  // Also check the FU pool as instructions are "stored" in FU
398  // completion events until they are done and not accounted for
399  // above
400  if (drained && !fuPool->isDrained()) {
401  DPRINTF(Drain, "FU pool still busy.\n");
402  drained = false;
403  }
404 
405  return drained;
406 }
407 
408 template <class Impl>
409 void
411 {
412  assert(isDrained());
413 
414  instQueue.drainSanityCheck();
415  ldstQueue.drainSanityCheck();
416 }
417 
418 template <class Impl>
419 void
421 {
422  // Reset all state.
423  _status = Active;
424  exeStatus = Running;
425  wbStatus = Idle;
426 
427  instQueue.takeOverFrom();
428  ldstQueue.takeOverFrom();
429  fuPool->takeOverFrom();
430 
431  startupStage();
432  cpu->activityThisCycle();
433 
434  for (ThreadID tid = 0; tid < numThreads; tid++) {
435  dispatchStatus[tid] = Running;
436  fetchRedirect[tid] = false;
437  }
438 
439  updateLSQNextCycle = false;
440 
441  for (int i = 0; i < issueToExecQueue.getSize(); ++i) {
442  issueToExecQueue.advance();
443  }
444 }
445 
446 template<class Impl>
447 void
449 {
450  DPRINTF(IEW, "[tid:%i]: Squashing all instructions.\n", tid);
451 
452  // Tell the IQ to start squashing.
453  instQueue.squash(tid);
454 
455  // Tell the LDSTQ to start squashing.
456  ldstQueue.squash(fromCommit->commitInfo[tid].doneSeqNum, tid);
457  updatedQueues = true;
458 
459  // Clear the skid buffer in case it has any data in it.
460  DPRINTF(IEW, "[tid:%i]: Removing skidbuffer instructions until [sn:%i].\n",
461  tid, fromCommit->commitInfo[tid].doneSeqNum);
462 
463  while (!skidBuffer[tid].empty()) {
464  if (skidBuffer[tid].front()->isLoad()) {
465  toRename->iewInfo[tid].dispatchedToLQ++;
466  }
467  if (skidBuffer[tid].front()->isStore()) {
468  toRename->iewInfo[tid].dispatchedToSQ++;
469  }
470 
471  toRename->iewInfo[tid].dispatched++;
472 
473  skidBuffer[tid].pop();
474  }
475 
476  emptyRenameInsts(tid);
477 }
478 
479 template<class Impl>
480 void
482 {
483  DPRINTF(IEW, "[tid:%i]: Squashing from a specific instruction, PC: %s "
484  "[sn:%i].\n", tid, inst->pcState(), inst->seqNum);
485 
486  if (!toCommit->squash[tid] ||
487  inst->seqNum < toCommit->squashedSeqNum[tid]) {
488  toCommit->squash[tid] = true;
489  toCommit->squashedSeqNum[tid] = inst->seqNum;
490  toCommit->branchTaken[tid] = inst->pcState().branching();
491 
492  TheISA::PCState pc = inst->pcState();
493  TheISA::advancePC(pc, inst->staticInst);
494 
495  toCommit->pc[tid] = pc;
496  toCommit->mispredictInst[tid] = inst;
497  toCommit->includeSquashInst[tid] = false;
498 
499  wroteToTimeBuffer = true;
500  }
501 
502 }
503 
504 template<class Impl>
505 void
507 {
508  DPRINTF(IEW, "[tid:%i]: Memory violation, squashing violator and younger "
509  "insts, PC: %s [sn:%i].\n", tid, inst->pcState(), inst->seqNum);
510  // Need to include inst->seqNum in the following comparison to cover the
511  // corner case when a branch misprediction and a memory violation for the
512  // same instruction (e.g. load PC) are detected in the same cycle. In this
513  // case the memory violator should take precedence over the branch
514  // misprediction because it requires the violator itself to be included in
515  // the squash.
516  if (!toCommit->squash[tid] ||
517  inst->seqNum <= toCommit->squashedSeqNum[tid]) {
518  toCommit->squash[tid] = true;
519 
520  toCommit->squashedSeqNum[tid] = inst->seqNum;
521  toCommit->pc[tid] = inst->pcState();
522  toCommit->mispredictInst[tid] = NULL;
523 
524  // Must include the memory violator in the squash.
525  toCommit->includeSquashInst[tid] = true;
526 
527  wroteToTimeBuffer = true;
528  }
529 }
530 
531 template<class Impl>
532 void
534 {
535  DPRINTF(IEW, "[tid:%u]: Blocking.\n", tid);
536 
537  if (dispatchStatus[tid] != Blocked &&
538  dispatchStatus[tid] != Unblocking) {
539  toRename->iewBlock[tid] = true;
540  wroteToTimeBuffer = true;
541  }
542 
543  // Add the current inputs to the skid buffer so they can be
544  // reprocessed when this stage unblocks.
545  skidInsert(tid);
546 
547  dispatchStatus[tid] = Blocked;
548 }
549 
550 template<class Impl>
551 void
553 {
554  DPRINTF(IEW, "[tid:%i]: Reading instructions out of the skid "
555  "buffer %u.\n",tid, tid);
556 
557  // If the skid bufffer is empty, signal back to previous stages to unblock.
558  // Also switch status to running.
559  if (skidBuffer[tid].empty()) {
560  toRename->iewUnblock[tid] = true;
561  wroteToTimeBuffer = true;
562  DPRINTF(IEW, "[tid:%i]: Done unblocking.\n",tid);
563  dispatchStatus[tid] = Running;
564  }
565 }
566 
567 template<class Impl>
568 void
570 {
571  instQueue.wakeDependents(inst);
572 }
573 
574 template<class Impl>
575 void
577 {
578  instQueue.rescheduleMemInst(inst);
579 }
580 
581 template<class Impl>
582 void
584 {
585  instQueue.replayMemInst(inst);
586 }
587 
588 template<class Impl>
589 void
591 {
592  instQueue.blockMemInst(inst);
593 }
594 
595 template<class Impl>
596 void
598 {
599  instQueue.cacheUnblocked();
600 }
601 
602 template<class Impl>
603 void
605 {
606  // This function should not be called after writebackInsts in a
607  // single cycle. That will cause problems with an instruction
608  // being added to the queue to commit without being processed by
609  // writebackInsts prior to being sent to commit.
610 
611  // First check the time slot that this instruction will write
612  // to. If there are free write ports at the time, then go ahead
613  // and write the instruction to that time. If there are not,
614  // keep looking back to see where's the first time there's a
615  // free slot.
616  while ((*iewQueue)[wbCycle].insts[wbNumInst]) {
617  ++wbNumInst;
618  if (wbNumInst == wbWidth) {
619  ++wbCycle;
620  wbNumInst = 0;
621  }
622  }
623 
624  DPRINTF(IEW, "Current wb cycle: %i, width: %i, numInst: %i\nwbActual:%i\n",
625  wbCycle, wbWidth, wbNumInst, wbCycle * wbWidth + wbNumInst);
626  // Add finished instruction to queue to commit.
627  (*iewQueue)[wbCycle].insts[wbNumInst] = inst;
628  (*iewQueue)[wbCycle].size++;
629 }
630 
631 template <class Impl>
632 unsigned
634 {
635  unsigned inst_count = 0;
636 
637  for (int i=0; i<fromRename->size; i++) {
638  if (!fromRename->insts[i]->isSquashed())
639  inst_count++;
640  }
641 
642  return inst_count;
643 }
644 
645 template<class Impl>
646 void
648 {
649  DynInstPtr inst = NULL;
650 
651  while (!insts[tid].empty()) {
652  inst = insts[tid].front();
653 
654  insts[tid].pop();
655 
656  DPRINTF(IEW,"[tid:%i]: Inserting [sn:%lli] PC:%s into "
657  "dispatch skidBuffer %i\n",tid, inst->seqNum,
658  inst->pcState(),tid);
659 
660  skidBuffer[tid].push(inst);
661  }
662 
663  assert(skidBuffer[tid].size() <= skidBufferMax &&
664  "Skidbuffer Exceeded Max Size");
665 }
666 
667 template<class Impl>
668 int
670 {
671  int max=0;
672 
673  list<ThreadID>::iterator threads = activeThreads->begin();
674  list<ThreadID>::iterator end = activeThreads->end();
675 
676  while (threads != end) {
677  ThreadID tid = *threads++;
678  unsigned thread_count = skidBuffer[tid].size();
679  if (max < thread_count)
680  max = thread_count;
681  }
682 
683  return max;
684 }
685 
686 template<class Impl>
687 bool
689 {
690  list<ThreadID>::iterator threads = activeThreads->begin();
691  list<ThreadID>::iterator end = activeThreads->end();
692 
693  while (threads != end) {
694  ThreadID tid = *threads++;
695 
696  if (!skidBuffer[tid].empty())
697  return false;
698  }
699 
700  return true;
701 }
702 
703 template <class Impl>
704 void
706 {
707  bool any_unblocking = false;
708 
709  list<ThreadID>::iterator threads = activeThreads->begin();
710  list<ThreadID>::iterator end = activeThreads->end();
711 
712  while (threads != end) {
713  ThreadID tid = *threads++;
714 
715  if (dispatchStatus[tid] == Unblocking) {
716  any_unblocking = true;
717  break;
718  }
719  }
720 
721  // If there are no ready instructions waiting to be scheduled by the IQ,
722  // and there's no stores waiting to write back, and dispatch is not
723  // unblocking, then there is no internal activity for the IEW stage.
724  instQueue.intInstQueueReads++;
725  if (_status == Active && !instQueue.hasReadyInsts() &&
726  !ldstQueue.willWB() && !any_unblocking) {
727  DPRINTF(IEW, "IEW switching to idle\n");
728 
729  deactivateStage();
730 
731  _status = Inactive;
732  } else if (_status == Inactive && (instQueue.hasReadyInsts() ||
733  ldstQueue.willWB() ||
734  any_unblocking)) {
735  // Otherwise there is internal activity. Set to active.
736  DPRINTF(IEW, "IEW switching to active\n");
737 
738  activateStage();
739 
740  _status = Active;
741  }
742 }
743 
744 template <class Impl>
745 void
747 {
748  instQueue.resetEntries();
749  ldstQueue.resetEntries();
750 }
751 
752 template <class Impl>
753 bool
755 {
756  bool ret_val(false);
757 
758  if (fromCommit->commitInfo[tid].robSquashing) {
759  DPRINTF(IEW,"[tid:%i]: Stall from Commit stage detected.\n",tid);
760  ret_val = true;
761  } else if (instQueue.isFull(tid)) {
762  DPRINTF(IEW,"[tid:%i]: Stall: IQ is full.\n",tid);
763  ret_val = true;
764  }
765 
766  return ret_val;
767 }
768 
769 template <class Impl>
770 void
772 {
773  // Check if there's a squash signal, squash if there is
774  // Check stall signals, block if there is.
775  // If status was Blocked
776  // if so then go to unblocking
777  // If status was Squashing
778  // check if squashing is not high. Switch to running this cycle.
779 
780  if (fromCommit->commitInfo[tid].squash) {
781  squash(tid);
782 
783  if (dispatchStatus[tid] == Blocked ||
784  dispatchStatus[tid] == Unblocking) {
785  toRename->iewUnblock[tid] = true;
786  wroteToTimeBuffer = true;
787  }
788 
789  dispatchStatus[tid] = Squashing;
790  fetchRedirect[tid] = false;
791  return;
792  }
793 
794  if (fromCommit->commitInfo[tid].robSquashing) {
795  DPRINTF(IEW, "[tid:%i]: ROB is still squashing.\n", tid);
796 
797  dispatchStatus[tid] = Squashing;
798  emptyRenameInsts(tid);
799  wroteToTimeBuffer = true;
800  }
801 
802  if (checkStall(tid)) {
803  block(tid);
804  dispatchStatus[tid] = Blocked;
805  return;
806  }
807 
808  if (dispatchStatus[tid] == Blocked) {
809  // Status from previous cycle was blocked, but there are no more stall
810  // conditions. Switch over to unblocking.
811  DPRINTF(IEW, "[tid:%i]: Done blocking, switching to unblocking.\n",
812  tid);
813 
814  dispatchStatus[tid] = Unblocking;
815 
816  unblock(tid);
817 
818  return;
819  }
820 
821  if (dispatchStatus[tid] == Squashing) {
822  // Switch status to running if rename isn't being told to block or
823  // squash this cycle.
824  DPRINTF(IEW, "[tid:%i]: Done squashing, switching to running.\n",
825  tid);
826 
827  dispatchStatus[tid] = Running;
828 
829  return;
830  }
831 }
832 
833 template <class Impl>
834 void
836 {
837  int insts_from_rename = fromRename->size;
838 #ifdef DEBUG
839  for (ThreadID tid = 0; tid < numThreads; tid++)
840  assert(insts[tid].empty());
841 #endif
842  for (int i = 0; i < insts_from_rename; ++i) {
843  insts[fromRename->insts[i]->threadNumber].push(fromRename->insts[i]);
844  }
845 }
846 
847 template <class Impl>
848 void
850 {
851  DPRINTF(IEW, "[tid:%i]: Removing incoming rename instructions\n", tid);
852 
853  while (!insts[tid].empty()) {
854 
855  if (insts[tid].front()->isLoad()) {
856  toRename->iewInfo[tid].dispatchedToLQ++;
857  }
858  if (insts[tid].front()->isStore()) {
859  toRename->iewInfo[tid].dispatchedToSQ++;
860  }
861 
862  toRename->iewInfo[tid].dispatched++;
863 
864  insts[tid].pop();
865  }
866 }
867 
868 template <class Impl>
869 void
871 {
872  cpu->wakeCPU();
873 }
874 
875 template <class Impl>
876 void
878 {
879  DPRINTF(Activity, "Activity this cycle.\n");
880  cpu->activityThisCycle();
881 }
882 
883 template <class Impl>
884 inline void
886 {
887  DPRINTF(Activity, "Activating stage.\n");
888  cpu->activateStage(O3CPU::IEWIdx);
889 }
890 
891 template <class Impl>
892 inline void
894 {
895  DPRINTF(Activity, "Deactivating stage.\n");
896  cpu->deactivateStage(O3CPU::IEWIdx);
897 }
898 
899 template<class Impl>
900 void
902 {
903  // If status is Running or idle,
904  // call dispatchInsts()
905  // If status is Unblocking,
906  // buffer any instructions coming from rename
907  // continue trying to empty skid buffer
908  // check if stall conditions have passed
909 
910  if (dispatchStatus[tid] == Blocked) {
911  ++iewBlockCycles;
912 
913  } else if (dispatchStatus[tid] == Squashing) {
914  ++iewSquashCycles;
915  }
916 
917  // Dispatch should try to dispatch as many instructions as its bandwidth
918  // will allow, as long as it is not currently blocked.
919  if (dispatchStatus[tid] == Running ||
920  dispatchStatus[tid] == Idle) {
921  DPRINTF(IEW, "[tid:%i] Not blocked, so attempting to run "
922  "dispatch.\n", tid);
923 
924  dispatchInsts(tid);
925  } else if (dispatchStatus[tid] == Unblocking) {
926  // Make sure that the skid buffer has something in it if the
927  // status is unblocking.
928  assert(!skidsEmpty());
929 
930  // If the status was unblocking, then instructions from the skid
931  // buffer were used. Remove those instructions and handle
932  // the rest of unblocking.
933  dispatchInsts(tid);
934 
935  ++iewUnblockCycles;
936 
937  if (validInstsFromRename()) {
938  // Add the current inputs to the skid buffer so they can be
939  // reprocessed when this stage unblocks.
940  skidInsert(tid);
941  }
942 
943  unblock(tid);
944  }
945 }
946 
947 template <class Impl>
948 void
950 {
951  // Obtain instructions from skid buffer if unblocking, or queue from rename
952  // otherwise.
953  std::queue<DynInstPtr> &insts_to_dispatch =
954  dispatchStatus[tid] == Unblocking ?
955  skidBuffer[tid] : insts[tid];
956 
957  int insts_to_add = insts_to_dispatch.size();
958 
959  DynInstPtr inst;
960  bool add_to_iq = false;
961  int dis_num_inst = 0;
962 
963  // Loop through the instructions, putting them in the instruction
964  // queue.
965  for ( ; dis_num_inst < insts_to_add &&
966  dis_num_inst < dispatchWidth;
967  ++dis_num_inst)
968  {
969  inst = insts_to_dispatch.front();
970 
971  if (dispatchStatus[tid] == Unblocking) {
972  DPRINTF(IEW, "[tid:%i]: Issue: Examining instruction from skid "
973  "buffer\n", tid);
974  }
975 
976  // Make sure there's a valid instruction there.
977  assert(inst);
978 
979  DPRINTF(IEW, "[tid:%i]: Issue: Adding PC %s [sn:%lli] [tid:%i] to "
980  "IQ.\n",
981  tid, inst->pcState(), inst->seqNum, inst->threadNumber);
982 
983  // Be sure to mark these instructions as ready so that the
984  // commit stage can go ahead and execute them, and mark
985  // them as issued so the IQ doesn't reprocess them.
986 
987  // Check for squashed instructions.
988  if (inst->isSquashed()) {
989  DPRINTF(IEW, "[tid:%i]: Issue: Squashed instruction encountered, "
990  "not adding to IQ.\n", tid);
991 
992  ++iewDispSquashedInsts;
993 
994  insts_to_dispatch.pop();
995 
996  //Tell Rename That An Instruction has been processed
997  if (inst->isLoad()) {
998  toRename->iewInfo[tid].dispatchedToLQ++;
999  }
1000  if (inst->isStore()) {
1001  toRename->iewInfo[tid].dispatchedToSQ++;
1002  }
1003 
1004  toRename->iewInfo[tid].dispatched++;
1005 
1006  continue;
1007  }
1008 
1009  // Check for full conditions.
1010  if (instQueue.isFull(tid)) {
1011  DPRINTF(IEW, "[tid:%i]: Issue: IQ has become full.\n", tid);
1012 
1013  // Call function to start blocking.
1014  block(tid);
1015 
1016  // Set unblock to false. Special case where we are using
1017  // skidbuffer (unblocking) instructions but then we still
1018  // get full in the IQ.
1019  toRename->iewUnblock[tid] = false;
1020 
1021  ++iewIQFullEvents;
1022  break;
1023  }
1024 
1025  // Check LSQ if inst is LD/ST
1026  if ((inst->isLoad() && ldstQueue.lqFull(tid)) ||
1027  (inst->isStore() && ldstQueue.sqFull(tid))) {
1028  DPRINTF(IEW, "[tid:%i]: Issue: %s has become full.\n",tid,
1029  inst->isLoad() ? "LQ" : "SQ");
1030 
1031  // Call function to start blocking.
1032  block(tid);
1033 
1034  // Set unblock to false. Special case where we are using
1035  // skidbuffer (unblocking) instructions but then we still
1036  // get full in the IQ.
1037  toRename->iewUnblock[tid] = false;
1038 
1039  ++iewLSQFullEvents;
1040  break;
1041  }
1042 
1043  // Otherwise issue the instruction just fine.
1044  if (inst->isLoad()) {
1045  DPRINTF(IEW, "[tid:%i]: Issue: Memory instruction "
1046  "encountered, adding to LSQ.\n", tid);
1047 
1048  // Reserve a spot in the load store queue for this
1049  // memory access.
1050  ldstQueue.insertLoad(inst);
1051 
1052  ++iewDispLoadInsts;
1053 
1054  add_to_iq = true;
1055 
1056  toRename->iewInfo[tid].dispatchedToLQ++;
1057  } else if (inst->isStore()) {
1058  DPRINTF(IEW, "[tid:%i]: Issue: Memory instruction "
1059  "encountered, adding to LSQ.\n", tid);
1060 
1061  ldstQueue.insertStore(inst);
1062 
1063  ++iewDispStoreInsts;
1064 
1065  if (inst->isStoreConditional()) {
1066  // Store conditionals need to be set as "canCommit()"
1067  // so that commit can process them when they reach the
1068  // head of commit.
1069  // @todo: This is somewhat specific to Alpha.
1070  inst->setCanCommit();
1071  instQueue.insertNonSpec(inst);
1072  add_to_iq = false;
1073 
1074  ++iewDispNonSpecInsts;
1075  } else {
1076  add_to_iq = true;
1077  }
1078 
1079  toRename->iewInfo[tid].dispatchedToSQ++;
1080  } else if (inst->isMemBarrier() || inst->isWriteBarrier()) {
1081  // Same as non-speculative stores.
1082  inst->setCanCommit();
1083  instQueue.insertBarrier(inst);
1084  add_to_iq = false;
1085  } else if (inst->isNop()) {
1086  DPRINTF(IEW, "[tid:%i]: Issue: Nop instruction encountered, "
1087  "skipping.\n", tid);
1088 
1089  inst->setIssued();
1090  inst->setExecuted();
1091  inst->setCanCommit();
1092 
1093  instQueue.recordProducer(inst);
1094 
1095  iewExecutedNop[tid]++;
1096 
1097  add_to_iq = false;
1098  } else {
1099  assert(!inst->isExecuted());
1100  add_to_iq = true;
1101  }
1102 
1103  if (inst->isNonSpeculative()) {
1104  DPRINTF(IEW, "[tid:%i]: Issue: Nonspeculative instruction "
1105  "encountered, skipping.\n", tid);
1106 
1107  // Same as non-speculative stores.
1108  inst->setCanCommit();
1109 
1110  // Specifically insert it as nonspeculative.
1111  instQueue.insertNonSpec(inst);
1112 
1113  ++iewDispNonSpecInsts;
1114 
1115  add_to_iq = false;
1116  }
1117 
1118  // If the instruction queue is not full, then add the
1119  // instruction.
1120  if (add_to_iq) {
1121  instQueue.insert(inst);
1122  }
1123 
1124  insts_to_dispatch.pop();
1125 
1126  toRename->iewInfo[tid].dispatched++;
1127 
1128  ++iewDispatchedInsts;
1129 
1130 #if TRACING_ON
1131  inst->dispatchTick = curTick() - inst->fetchTick;
1132 #endif
1133  ppDispatch->notify(inst);
1134  }
1135 
1136  if (!insts_to_dispatch.empty()) {
1137  DPRINTF(IEW,"[tid:%i]: Issue: Bandwidth Full. Blocking.\n", tid);
1138  block(tid);
1139  toRename->iewUnblock[tid] = false;
1140  }
1141 
1142  if (dispatchStatus[tid] == Idle && dis_num_inst) {
1143  dispatchStatus[tid] = Running;
1144 
1145  updatedQueues = true;
1146  }
1147 
1148  dis_num_inst = 0;
1149 }
1150 
1151 template <class Impl>
1152 void
1154 {
1155  int inst = 0;
1156 
1157  std::cout << "Available Instructions: ";
1158 
1159  while (fromIssue->insts[inst]) {
1160 
1161  if (inst%3==0) std::cout << "\n\t";
1162 
1163  std::cout << "PC: " << fromIssue->insts[inst]->pcState()
1164  << " TN: " << fromIssue->insts[inst]->threadNumber
1165  << " SN: " << fromIssue->insts[inst]->seqNum << " | ";
1166 
1167  inst++;
1168 
1169  }
1170 
1171  std::cout << "\n";
1172 }
1173 
1174 template <class Impl>
1175 void
1177 {
1178  wbNumInst = 0;
1179  wbCycle = 0;
1180 
1181  list<ThreadID>::iterator threads = activeThreads->begin();
1182  list<ThreadID>::iterator end = activeThreads->end();
1183 
1184  while (threads != end) {
1185  ThreadID tid = *threads++;
1186  fetchRedirect[tid] = false;
1187  }
1188 
1189  // Uncomment this if you want to see all available instructions.
1190  // @todo This doesn't actually work anymore, we should fix it.
1191 // printAvailableInsts();
1192 
1193  // Execute/writeback any instructions that are available.
1194  int insts_to_execute = fromIssue->size;
1195  int inst_num = 0;
1196  for (; inst_num < insts_to_execute;
1197  ++inst_num) {
1198 
1199  DPRINTF(IEW, "Execute: Executing instructions from IQ.\n");
1200 
1201  DynInstPtr inst = instQueue.getInstToExecute();
1202 
1203  DPRINTF(IEW, "Execute: Processing PC %s, [tid:%i] [sn:%i].\n",
1204  inst->pcState(), inst->threadNumber,inst->seqNum);
1205 
1206  // Notify potential listeners that this instruction has started
1207  // executing
1208  ppExecute->notify(inst);
1209 
1210  // Check if the instruction is squashed; if so then skip it
1211  if (inst->isSquashed()) {
1212  DPRINTF(IEW, "Execute: Instruction was squashed. PC: %s, [tid:%i]"
1213  " [sn:%i]\n", inst->pcState(), inst->threadNumber,
1214  inst->seqNum);
1215 
1216  // Consider this instruction executed so that commit can go
1217  // ahead and retire the instruction.
1218  inst->setExecuted();
1219 
1220  // Not sure if I should set this here or just let commit try to
1221  // commit any squashed instructions. I like the latter a bit more.
1222  inst->setCanCommit();
1223 
1224  ++iewExecSquashedInsts;
1225 
1226  continue;
1227  }
1228 
1229  Fault fault = NoFault;
1230 
1231  // Execute instruction.
1232  // Note that if the instruction faults, it will be handled
1233  // at the commit stage.
1234  if (inst->isMemRef()) {
1235  DPRINTF(IEW, "Execute: Calculating address for memory "
1236  "reference.\n");
1237 
1238  // Tell the LDSTQ to execute this instruction (if it is a load).
1239  if (inst->isLoad()) {
1240  // Loads will mark themselves as executed, and their writeback
1241  // event adds the instruction to the queue to commit
1242  fault = ldstQueue.executeLoad(inst);
1243 
1244  if (inst->isTranslationDelayed() &&
1245  fault == NoFault) {
1246  // A hw page table walk is currently going on; the
1247  // instruction must be deferred.
1248  DPRINTF(IEW, "Execute: Delayed translation, deferring "
1249  "load.\n");
1250  instQueue.deferMemInst(inst);
1251  continue;
1252  }
1253 
1254  if (inst->isDataPrefetch() || inst->isInstPrefetch()) {
1255  inst->fault = NoFault;
1256  }
1257  } else if (inst->isStore()) {
1258  fault = ldstQueue.executeStore(inst);
1259 
1260  if (inst->isTranslationDelayed() &&
1261  fault == NoFault) {
1262  // A hw page table walk is currently going on; the
1263  // instruction must be deferred.
1264  DPRINTF(IEW, "Execute: Delayed translation, deferring "
1265  "store.\n");
1266  instQueue.deferMemInst(inst);
1267  continue;
1268  }
1269 
1270  // If the store had a fault then it may not have a mem req
1271  if (fault != NoFault || !inst->readPredicate() ||
1272  !inst->isStoreConditional()) {
1273  // If the instruction faulted, then we need to send it along
1274  // to commit without the instruction completing.
1275  // Send this instruction to commit, also make sure iew stage
1276  // realizes there is activity.
1277  inst->setExecuted();
1278  instToCommit(inst);
1279  activityThisCycle();
1280  }
1281 
1282  // Store conditionals will mark themselves as
1283  // executed, and their writeback event will add the
1284  // instruction to the queue to commit.
1285  } else {
1286  panic("Unexpected memory type!\n");
1287  }
1288 
1289  } else {
1290  // If the instruction has already faulted, then skip executing it.
1291  // Such case can happen when it faulted during ITLB translation.
1292  // If we execute the instruction (even if it's a nop) the fault
1293  // will be replaced and we will lose it.
1294  if (inst->getFault() == NoFault) {
1295  inst->execute();
1296  if (!inst->readPredicate())
1297  inst->forwardOldRegs();
1298  }
1299 
1300  inst->setExecuted();
1301 
1302  instToCommit(inst);
1303  }
1304 
1305  updateExeInstStats(inst);
1306 
1307  // Check if branch prediction was correct, if not then we need
1308  // to tell commit to squash in flight instructions. Only
1309  // handle this if there hasn't already been something that
1310  // redirects fetch in this group of instructions.
1311 
1312  // This probably needs to prioritize the redirects if a different
1313  // scheduler is used. Currently the scheduler schedules the oldest
1314  // instruction first, so the branch resolution order will be correct.
1315  ThreadID tid = inst->threadNumber;
1316 
1317  if (!fetchRedirect[tid] ||
1318  !toCommit->squash[tid] ||
1319  toCommit->squashedSeqNum[tid] > inst->seqNum) {
1320 
1321  // Prevent testing for misprediction on load instructions,
1322  // that have not been executed.
1323  bool loadNotExecuted = !inst->isExecuted() && inst->isLoad();
1324 
1325  if (inst->mispredicted() && !loadNotExecuted) {
1326  fetchRedirect[tid] = true;
1327 
1328  DPRINTF(IEW, "Execute: Branch mispredict detected.\n");
1329  DPRINTF(IEW, "Predicted target was PC: %s.\n",
1330  inst->readPredTarg());
1331  DPRINTF(IEW, "Execute: Redirecting fetch to PC: %s.\n",
1332  inst->pcState());
1333  // If incorrect, then signal the ROB that it must be squashed.
1334  squashDueToBranch(inst, tid);
1335 
1336  ppMispredict->notify(inst);
1337 
1338  if (inst->readPredTaken()) {
1339  predictedTakenIncorrect++;
1340  } else {
1341  predictedNotTakenIncorrect++;
1342  }
1343  } else if (ldstQueue.violation(tid)) {
1344  assert(inst->isMemRef());
1345  // If there was an ordering violation, then get the
1346  // DynInst that caused the violation. Note that this
1347  // clears the violation signal.
1348  DynInstPtr violator;
1349  violator = ldstQueue.getMemDepViolator(tid);
1350 
1351  DPRINTF(IEW, "LDSTQ detected a violation. Violator PC: %s "
1352  "[sn:%lli], inst PC: %s [sn:%lli]. Addr is: %#x.\n",
1353  violator->pcState(), violator->seqNum,
1354  inst->pcState(), inst->seqNum, inst->physEffAddrLow);
1355 
1356  fetchRedirect[tid] = true;
1357 
1358  // Tell the instruction queue that a violation has occured.
1359  instQueue.violation(inst, violator);
1360 
1361  // Squash.
1362  squashDueToMemOrder(violator, tid);
1363 
1364  ++memOrderViolationEvents;
1365  }
1366  } else {
1367  // Reset any state associated with redirects that will not
1368  // be used.
1369  if (ldstQueue.violation(tid)) {
1370  assert(inst->isMemRef());
1371 
1372  DynInstPtr violator = ldstQueue.getMemDepViolator(tid);
1373 
1374  DPRINTF(IEW, "LDSTQ detected a violation. Violator PC: "
1375  "%s, inst PC: %s. Addr is: %#x.\n",
1376  violator->pcState(), inst->pcState(),
1377  inst->physEffAddrLow);
1378  DPRINTF(IEW, "Violation will not be handled because "
1379  "already squashing\n");
1380 
1381  ++memOrderViolationEvents;
1382  }
1383  }
1384  }
1385 
1386  // Update and record activity if we processed any instructions.
1387  if (inst_num) {
1388  if (exeStatus == Idle) {
1389  exeStatus = Running;
1390  }
1391 
1392  updatedQueues = true;
1393 
1394  cpu->activityThisCycle();
1395  }
1396 
1397  // Need to reset this in case a writeback event needs to write into the
1398  // iew queue. That way the writeback event will write into the correct
1399  // spot in the queue.
1400  wbNumInst = 0;
1401 
1402 }
1403 
1404 template <class Impl>
1405 void
1407 {
1408  // Loop through the head of the time buffer and wake any
1409  // dependents. These instructions are about to write back. Also
1410  // mark scoreboard that this instruction is finally complete.
1411  // Either have IEW have direct access to scoreboard, or have this
1412  // as part of backwards communication.
1413  for (int inst_num = 0; inst_num < wbWidth &&
1414  toCommit->insts[inst_num]; inst_num++) {
1415  DynInstPtr inst = toCommit->insts[inst_num];
1416  ThreadID tid = inst->threadNumber;
1417 
1418  DPRINTF(IEW, "Sending instructions to commit, [sn:%lli] PC %s.\n",
1419  inst->seqNum, inst->pcState());
1420 
1421  iewInstsToCommit[tid]++;
1422  // Notify potential listeners that execution is complete for this
1423  // instruction.
1424  ppToCommit->notify(inst);
1425 
1426  // Some instructions will be sent to commit without having
1427  // executed because they need commit to handle them.
1428  // E.g. Strictly ordered loads have not actually executed when they
1429  // are first sent to commit. Instead commit must tell the LSQ
1430  // when it's ready to execute the strictly ordered load.
1431  if (!inst->isSquashed() && inst->isExecuted() && inst->getFault() == NoFault) {
1432  int dependents = instQueue.wakeDependents(inst);
1433 
1434  for (int i = 0; i < inst->numDestRegs(); i++) {
1435  //mark as Ready
1436  DPRINTF(IEW,"Setting Destination Register %i\n",
1437  inst->renamedDestRegIdx(i));
1438  scoreboard->setReg(inst->renamedDestRegIdx(i));
1439  }
1440 
1441  if (dependents) {
1442  producerInst[tid]++;
1443  consumerInst[tid]+= dependents;
1444  }
1445  writebackCount[tid]++;
1446  }
1447  }
1448 }
1449 
1450 template<class Impl>
1451 void
1453 {
1454  wbNumInst = 0;
1455  wbCycle = 0;
1456 
1457  wroteToTimeBuffer = false;
1458  updatedQueues = false;
1459 
1460  sortInsts();
1461 
1462  // Free function units marked as being freed this cycle.
1463  fuPool->processFreeUnits();
1464 
1465  list<ThreadID>::iterator threads = activeThreads->begin();
1466  list<ThreadID>::iterator end = activeThreads->end();
1467 
1468  // Check stall and squash signals, dispatch any instructions.
1469  while (threads != end) {
1470  ThreadID tid = *threads++;
1471 
1472  DPRINTF(IEW,"Issue: Processing [tid:%i]\n",tid);
1473 
1474  checkSignalsAndUpdate(tid);
1475  dispatch(tid);
1476  }
1477 
1478  if (exeStatus != Squashing) {
1479  executeInsts();
1480 
1481  writebackInsts();
1482 
1483  // Have the instruction queue try to schedule any ready instructions.
1484  // (In actuality, this scheduling is for instructions that will
1485  // be executed next cycle.)
1486  instQueue.scheduleReadyInsts();
1487 
1488  // Also should advance its own time buffers if the stage ran.
1489  // Not the best place for it, but this works (hopefully).
1490  issueToExecQueue.advance();
1491  }
1492 
1493  bool broadcast_free_entries = false;
1494 
1495  if (updatedQueues || exeStatus == Running || updateLSQNextCycle) {
1496  exeStatus = Idle;
1497  updateLSQNextCycle = false;
1498 
1499  broadcast_free_entries = true;
1500  }
1501 
1502  // Writeback any stores using any leftover bandwidth.
1503  ldstQueue.writebackStores();
1504 
1505  // Check the committed load/store signals to see if there's a load
1506  // or store to commit. Also check if it's being told to execute a
1507  // nonspeculative instruction.
1508  // This is pretty inefficient...
1509 
1510  threads = activeThreads->begin();
1511  while (threads != end) {
1512  ThreadID tid = (*threads++);
1513 
1514  DPRINTF(IEW,"Processing [tid:%i]\n",tid);
1515 
1516  // Update structures based on instructions committed.
1517  if (fromCommit->commitInfo[tid].doneSeqNum != 0 &&
1518  !fromCommit->commitInfo[tid].squash &&
1519  !fromCommit->commitInfo[tid].robSquashing) {
1520 
1521  ldstQueue.commitStores(fromCommit->commitInfo[tid].doneSeqNum,tid);
1522 
1523  ldstQueue.commitLoads(fromCommit->commitInfo[tid].doneSeqNum,tid);
1524 
1525  updateLSQNextCycle = true;
1526  instQueue.commit(fromCommit->commitInfo[tid].doneSeqNum,tid);
1527  }
1528 
1529  if (fromCommit->commitInfo[tid].nonSpecSeqNum != 0) {
1530 
1531  //DPRINTF(IEW,"NonspecInst from thread %i",tid);
1532  if (fromCommit->commitInfo[tid].strictlyOrdered) {
1533  instQueue.replayMemInst(
1534  fromCommit->commitInfo[tid].strictlyOrderedLoad);
1535  fromCommit->commitInfo[tid].strictlyOrderedLoad->setAtCommit();
1536  } else {
1537  instQueue.scheduleNonSpec(
1538  fromCommit->commitInfo[tid].nonSpecSeqNum);
1539  }
1540  }
1541 
1542  if (broadcast_free_entries) {
1543  toFetch->iewInfo[tid].iqCount =
1544  instQueue.getCount(tid);
1545  toFetch->iewInfo[tid].ldstqCount =
1546  ldstQueue.getCount(tid);
1547 
1548  toRename->iewInfo[tid].usedIQ = true;
1549  toRename->iewInfo[tid].freeIQEntries =
1550  instQueue.numFreeEntries(tid);
1551  toRename->iewInfo[tid].usedLSQ = true;
1552 
1553  toRename->iewInfo[tid].freeLQEntries =
1554  ldstQueue.numFreeLoadEntries(tid);
1555  toRename->iewInfo[tid].freeSQEntries =
1556  ldstQueue.numFreeStoreEntries(tid);
1557 
1558  wroteToTimeBuffer = true;
1559  }
1560 
1561  DPRINTF(IEW, "[tid:%i], Dispatch dispatched %i instructions.\n",
1562  tid, toRename->iewInfo[tid].dispatched);
1563  }
1564 
1565  DPRINTF(IEW, "IQ has %i free entries (Can schedule: %i). "
1566  "LQ has %i free entries. SQ has %i free entries.\n",
1567  instQueue.numFreeEntries(), instQueue.hasReadyInsts(),
1568  ldstQueue.numFreeLoadEntries(), ldstQueue.numFreeStoreEntries());
1569 
1570  updateStatus();
1571 
1572  if (wroteToTimeBuffer) {
1573  DPRINTF(Activity, "Activity this cycle.\n");
1574  cpu->activityThisCycle();
1575  }
1576 }
1577 
1578 template <class Impl>
1579 void
1581 {
1582  ThreadID tid = inst->threadNumber;
1583 
1584  iewExecutedInsts++;
1585 
1586 #if TRACING_ON
1587  if (DTRACE(O3PipeView)) {
1588  inst->completeTick = curTick() - inst->fetchTick;
1589  }
1590 #endif
1591 
1592  //
1593  // Control operations
1594  //
1595  if (inst->isControl())
1596  iewExecutedBranches[tid]++;
1597 
1598  //
1599  // Memory operations
1600  //
1601  if (inst->isMemRef()) {
1602  iewExecutedRefs[tid]++;
1603 
1604  if (inst->isLoad()) {
1605  iewExecLoadInsts[tid]++;
1606  }
1607  }
1608 }
1609 
1610 template <class Impl>
1611 void
1613 {
1614  ThreadID tid = inst->threadNumber;
1615 
1616  if (!fetchRedirect[tid] ||
1617  !toCommit->squash[tid] ||
1618  toCommit->squashedSeqNum[tid] > inst->seqNum) {
1619 
1620  if (inst->mispredicted()) {
1621  fetchRedirect[tid] = true;
1622 
1623  DPRINTF(IEW, "Execute: Branch mispredict detected.\n");
1624  DPRINTF(IEW, "Predicted target was PC:%#x, NPC:%#x.\n",
1625  inst->predInstAddr(), inst->predNextInstAddr());
1626  DPRINTF(IEW, "Execute: Redirecting fetch to PC: %#x,"
1627  " NPC: %#x.\n", inst->nextInstAddr(),
1628  inst->nextInstAddr());
1629  // If incorrect, then signal the ROB that it must be squashed.
1630  squashDueToBranch(inst, tid);
1631 
1632  if (inst->readPredTaken()) {
1633  predictedTakenIncorrect++;
1634  } else {
1635  predictedNotTakenIncorrect++;
1636  }
1637  }
1638  }
1639 }
1640 
1641 #endif//__CPU_O3_IEW_IMPL_IMPL_HH__
Cycles renameToIEWDelay
Rename to IEW delay.
Definition: iew.hh:381
#define DPRINTF(x,...)
Definition: trace.hh:212
void setTimeBuffer(TimeBuffer< TimeStruct > *tb_ptr)
Sets main time buffer used for backwards communication.
Definition: iew_impl.hh:326
unsigned issueWidth
Width of issue, in instructions.
Definition: iew.hh:394
bool fetchRedirect[Impl::MaxThreads]
Records if there is a fetch redirect on this cycle for each thread.
Definition: iew.hh:370
StageStatus exeStatus
Execute status.
Definition: iew.hh:122
decltype(nullptr) constexpr NoFault
Definition: types.hh:189
void setIEWQueue(TimeBuffer< IEWStruct > *iq_ptr)
Sets time buffer to pass on instructions to commit.
Definition: iew_impl.hh:354
const std::string & name()
Definition: trace.cc:49
Bitfield< 7 > i
Definition: miscregs.hh:1378
void setRenameQueue(TimeBuffer< RenameStruct > *rq_ptr)
Sets time buffer for getting instructions coming from rename.
Definition: iew_impl.hh:344
#define panic(...)
Definition: misc.hh:153
void startupStage()
Initializes stage; sends back the number of free IQ and LSQ entries.
Definition: iew_impl.hh:304
void unblock(ThreadID tid)
Unblocks Dispatch if the skid buffer is empty, and signals back to other stages to unblock...
Definition: iew_impl.hh:552
void squashDueToMemOrder(DynInstPtr &inst, ThreadID tid)
Sends commit proper information for a squash due to a memory order violation.
Definition: iew_impl.hh:506
void checkMisprediction(DynInstPtr &inst)
Check misprediction.
Definition: iew_impl.hh:1612
ThreadID numThreads
Number of active threads.
Definition: iew.hh:410
void printAvailableInsts()
Debug function to print instructions that are issued this cycle.
Definition: iew_impl.hh:1153
Cycles issueToExecuteDelay
Issue to execute delay.
Definition: iew.hh:388
DefaultIEW(O3CPU *_cpu, DerivO3CPUParams *params)
Constructs a DefaultIEW with the given parameters.
Definition: iew_impl.hh:68
void blockMemInst(DynInstPtr &inst)
Moves memory instruction onto the list of cache blocked instructions.
Definition: iew_impl.hh:590
void squash(ThreadID tid)
Squashes instructions in IEW for a specific thread.
Definition: iew_impl.hh:448
bool checkStall(ThreadID tid)
Checks if any of the stall conditions are currently true.
Definition: iew_impl.hh:754
unsigned dispatchWidth
Width of dispatch, in instructions.
Definition: iew.hh:391
StageStatus wbStatus
Writeback status.
Definition: iew.hh:124
unsigned validInstsFromRename()
Returns the number of valid, non-squashed instructions coming from rename to dispatch.
Definition: iew_impl.hh:633
Implements a simple scoreboard to track which registers are ready.
Definition: scoreboard.hh:56
void cacheUnblocked()
Notifies that the cache has become unblocked.
Definition: iew_impl.hh:597
IQ instQueue
Instruction queue.
Definition: iew.hh:356
TimeBuffer< IssueStruct >::wire fromIssue
Wire to read information from the issue stage time queue.
Definition: iew.hh:322
Tick curTick()
The current simulated tick.
Definition: core.hh:47
void activateStage()
Tells CPU that the IEW stage is active and running.
Definition: iew_impl.hh:885
#define DTRACE(x)
Definition: trace.hh:210
void instToCommit(DynInstPtr &inst)
Sends an instruction to commit through the time buffer.
Definition: iew_impl.hh:604
void dispatchInsts(ThreadID tid)
Dispatches instructions to IQ and LSQ.
Definition: iew_impl.hh:949
Status _status
Overall stage status.
Definition: iew.hh:118
TimeBuffer< IssueStruct > issueToExecQueue
Issue stage queue.
Definition: iew.hh:319
void takeOverFrom()
Takes over from another CPU's thread.
Definition: iew_impl.hh:420
void tick()
Ticks IEW stage, causing Dispatch, the IQ, the LSQ, Execute, and Writeback to run for one cycle...
Definition: iew_impl.hh:1452
bool isDrained() const
Has the stage drained?
Definition: iew_impl.hh:381
void emptyRenameInsts(ThreadID tid)
Removes instructions from rename from a thread's instruction list.
Definition: iew_impl.hh:849
#define fatal(...)
Definition: misc.hh:163
void checkSignalsAndUpdate(ThreadID tid)
Processes inputs and changes state accordingly.
Definition: iew_impl.hh:771
unsigned wbWidth
Writeback width.
Definition: iew.hh:407
void wakeCPU()
Tells the CPU to wakeup if it has descheduled itself due to no activity.
Definition: iew_impl.hh:870
void advancePC(PCState &pc, const StaticInstPtr &inst)
Definition: utility.hh:108
bool updateLSQNextCycle
Records if the LSQ needs to be updated on the next cycle, so that IEW knows if there will be activity...
Definition: iew.hh:366
const FlagsType total
Print the total.
Definition: info.hh:49
void squashDueToBranch(DynInstPtr &inst, ThreadID tid)
Sends commit proper information for a squash due to a branch mispredict.
Definition: iew_impl.hh:481
void activityThisCycle()
Reports to the CPU that there is activity this cycle.
Definition: iew_impl.hh:877
int16_t ThreadID
Thread index/ID type.
Definition: types.hh:171
void setActiveThreads(std::list< ThreadID > *at_ptr)
Sets pointer to list of active threads.
Definition: iew_impl.hh:364
void regStats()
Registers statistics.
Definition: iew_impl.hh:144
int size()
Definition: pagetable.hh:146
int skidCount()
Returns the max of the number of entries in all of the skid buffers.
Definition: iew_impl.hh:669
void setScoreboard(Scoreboard *sb_ptr)
Sets pointer to the scoreboard.
Definition: iew_impl.hh:374
void executeInsts()
Executes instructions.
Definition: iew_impl.hh:1176
void drainSanityCheck() const
Perform sanity checks after a drain.
Definition: iew_impl.hh:410
GenericISA::SimplePCState< MachInst > PCState
Definition: types.hh:43
void writebackInsts()
Writebacks instructions.
Definition: iew_impl.hh:1406
void sortInsts()
Sorts instructions coming from rename into lists separated by thread.
Definition: iew_impl.hh:835
void dispatch(ThreadID tid)
Determines proper actions to take given Dispatch's status.
Definition: iew_impl.hh:901
void skidInsert(ThreadID tid)
Inserts unused instructions of a thread into the skid buffer.
Definition: iew_impl.hh:647
wire getWire(int idx)
Definition: timebuf.hh:232
void rescheduleMemInst(DynInstPtr &inst)
Tells memory dependence unit that a memory instruction needs to be rescheduled.
Definition: iew_impl.hh:576
void deactivateStage()
Tells CPU that the IEW stage is inactive and idle.
Definition: iew_impl.hh:893
IntReg pc
Definition: remote_gdb.hh:91
bool skidsEmpty()
Returns if all of the skid buffers are empty.
Definition: iew_impl.hh:688
unsigned skidBufferMax
Maximum size of the skid buffer.
Definition: iew.hh:416
Impl::DynInstPtr DynInstPtr
Definition: iew.hh:85
void wakeDependents(DynInstPtr &inst)
Wakes all dependents of a completed instruction.
Definition: iew_impl.hh:569
void resetEntries()
Resets entries of the IQ and the LSQ.
Definition: iew_impl.hh:746
StageStatus dispatchStatus[Impl::MaxThreads]
Dispatch status.
Definition: iew.hh:120
void updateExeInstStats(DynInstPtr &inst)
Updates execution stats based on the instruction.
Definition: iew_impl.hh:1580
void block(ThreadID tid)
Sets Dispatch to blocked, and signals back to other stages to block.
Definition: iew_impl.hh:533
Impl::O3CPU O3CPU
Definition: iew.hh:86
std::shared_ptr< FaultBase > Fault
Definition: types.hh:184
void regProbePoints()
Registers probes.
Definition: iew_impl.hh:124
std::string name() const
Returns the name of the DefaultIEW stage.
Definition: iew_impl.hh:117
void updateStatus()
Updates overall IEW status based on all of the stages' statuses.
Definition: iew_impl.hh:705
void replayMemInst(DynInstPtr &inst)
Re-executes all rescheduled memory instructions.
Definition: iew_impl.hh:583

Generated on Fri Jun 9 2017 13:03:43 for gem5 by doxygen 1.8.6